# KaizenAI: AI agents and Copilot on the Microsoft stack > KaizenAI is a remote-first, AI-first Microsoft consultancy working worldwide: AI agents, Copilot & Cowork, Dynamics 365, Power Platform and Kaizen PPM. KaizenAI builds AI agents, rolls out Copilot and Cowork, and implements Dynamics 365 and the Power Platform, all on secure Azure foundations. - Founder: Abdul Basit - Works: Remote-first, working with clients worldwide (home base: Lahore, Punjab, Pakistan) - Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ ## Services - [Agentic AI](https://kaizenai.systems/services/ai-agents/): Custom AI agents that act in your systems, with guardrails and humans in the loop. - [Copilot & Cowork](https://kaizenai.systems/services/copilot-cowork/): Make Microsoft 365 Copilot, Copilot Studio and Cowork useful with your own data. - [Dynamics 365 CRM](https://kaizenai.systems/services/dynamics-365/): Sales, Customer Service and Field Service, built around the way your team works. - [Power Platform](https://kaizenai.systems/services/power-platform/): Apps, automation and portals on Dataverse, governed and built to last. - [Systems Integration](https://kaizenai.systems/services/systems-integration/): Connect CRM, telephony, ERP and line-of-business systems so data flows without re-keying. - [Infra & DevSecOps](https://kaizenai.systems/services/infra-devsecops/): Secure Azure foundations and pipelines that ship safely, every time. ## Product: Kaizen PPM Kaizen PPM is a project and portfolio management (PPM) application built natively on Microsoft Power Platform and Dataverse. It takes work from idea to proposal, project and portfolio, and all data stays inside your own Microsoft 365 tenant. [Learn more](https://kaizenai.systems/products/kaizen-ppm/). ## How we work 01. **Discover**: We start from the business problem rather than the licence. We map the process, the data and what "better" would look like in numbers. 02. **Build small**: We ship a working slice in weeks so people can use it, and their feedback shapes the next slice. 03. **Improve continuously**: Kaizen means change for the better. After go-live we measure, tune and automate a little more each cycle. ## Principles - **Your data stays yours**: Solutions live in your own Microsoft 365 and Azure tenant. Nothing is locked into our systems. - **Humans stay in charge**: Our AI agents draft, flag and propose. A person approves anything that matters. - **Built to be maintained**: Source-controlled solutions, documented decisions and deployment pipelines, not one-off clicks. ## Frequently asked questions ### What does KaizenAI do? KaizenAI is an AI-first IT services and product company specialising in the Microsoft ecosystem. We build AI agents, roll out Microsoft 365 Copilot, Copilot Studio and Cowork, implement Dynamics 365 CRM and the Power Platform, and set up secure Azure infrastructure and DevSecOps pipelines. We also build Kaizen PPM, a project portfolio management product on Dataverse. ### Where is KaizenAI based, and who do you work with? We are a remote-first team working with organisations worldwide. Engagements are run in English and delivered remotely, across time zones. Our home base is Lahore, Pakistan. ### How do you use AI safely in business systems? Every AI agent we build is grounded in your own data, security-trimmed to the user it acts for, cites its sources, logs its actions and asks a person before it does anything consequential. We call this keeping a human in the loop. ### Do we need Microsoft 365 Copilot licences to work with you? No. Many of our AI agents run on Azure AI Foundry, Azure OpenAI, Claude or Copilot Studio without per-user Copilot licences. We will recommend whichever option gives the best value for your use case. ### How does an engagement start? With a free consultation. We discuss what you want to improve, then propose a small first step that can be delivered in weeks, not months, and improve from there. That is the kaizen way. --- # KaizenAI services AI and Microsoft services, end to end. Use one practice or combine several; they share the same Dataverse and Azure foundations. ## [Agentic AI](https://kaizenai.systems/services/ai-agents/) We design and build AI agents that read, decide and act in your business systems. Typical work includes document processing, triage, research and back-office automation. Every agent is grounded in your data, tightly scoped and auditable, and asks a person before it does anything important. Technology: Azure AI Foundry, Azure OpenAI, Claude, Model Context Protocol (MCP), Semantic Kernel, Copilot Studio, Python, TypeScript. ## [Copilot & Cowork](https://kaizenai.systems/services/copilot-cowork/) Licences alone do not create value. We prepare your data and permissions, build Copilot Studio agents grounded in your business, and set up Cowork so AI can carry out multi-step work across your Microsoft 365 apps under your rules. Technology: Microsoft 365 Copilot, Copilot Studio, Copilot Cowork, Microsoft Purview, SharePoint, Microsoft Graph. ## [Dynamics 365 CRM](https://kaizenai.systems/services/dynamics-365/) We implement, extend and rescue Dynamics 365 Customer Engagement. That covers clean data models, sensible security and automation people actually use, with nothing customised for its own sake. Technology: Dynamics 365 Sales, Customer Service, Field Service, Dataverse, C# plugins, PCF, Power Automate. ## [Power Platform](https://kaizenai.systems/services/power-platform/) Power Apps, Power Automate, Power Pages and Power BI on top of Dataverse. We build business apps quickly and add the ALM and governance that keep low-code from turning into sprawl. Technology: Power Apps, Power Automate, Power Pages, Power BI, Dataverse, Power Platform Pipelines, pac CLI. ## [Systems Integration](https://kaizenai.systems/services/systems-integration/) We connect CRMs, telephone systems, ERP and line-of-business applications through their APIs: scheduled and event-driven syncs, with duplicate prevention, error handling and monitoring built in. Integrations run in your own Azure tenant and are documented so your team can own them. Technology: Azure Functions (.NET), Logic Apps, Power Automate, REST & OData APIs, OAuth 2.0, Dataverse, Azure Monitor / OpenTelemetry. ## [Infra & DevSecOps](https://kaizenai.systems/services/infra-devsecops/) Cloud infrastructure, identity and delivery pipelines that are secure by default. We build Azure landing zones, harden Entra ID, and automate build, test and deployment for code and Power Platform solutions alike. Technology: Microsoft Azure, Entra ID, Bicep, Terraform, GitHub Actions, Azure DevOps, Microsoft Defender, Intune. --- # Agentic AI > Agentic AI means AI systems that do more than answer questions. They read your data, use tools and take steps in your business systems to finish a task, while a person stays in control of the decisions that matter. We design and build AI agents that read, decide and act in your business systems. Typical work includes document processing, triage, research and back-office automation. Every agent is grounded in your data, tightly scoped and auditable, and asks a person before it does anything important. ## What you get - Hours of repetitive knowledge work handled by agents - Every agent action logged and explainable - Model-agnostic, so you are not locked to one AI vendor ## How we help ### Agent discovery Find the processes where an agent pays back, and rule out the ones where it would not. ### Agent engineering Tool-using agents, retrieval (RAG) over your content, and MCP servers that expose your systems safely. ### Integration Agents wired into Dataverse, Dynamics 365, Teams, email and your APIs. ### Evaluation & guardrails Test suites, approval steps, permission trimming, cost controls and monitoring. ## Example use cases - **Inbox & document triage**: Read incoming emails and attachments, extract data, create records and route exceptions to a person. - **Sales research agent**: Research prospects and draft personalised outreach for a human to approve. ## Tools we use Azure AI Foundry, Azure OpenAI, Claude, Model Context Protocol (MCP), Semantic Kernel, Copilot Studio, Python, TypeScript. ## Frequently asked questions ### What is the difference between a chatbot and an AI agent? A chatbot answers questions. An AI agent can also use tools: it looks up records, fills forms, calls APIs and moves work forward across several steps. We design agents so that anything consequential, such as sending, approving or deleting, waits for a person to confirm. ### Which AI models do you use? We are model-agnostic. We work with Azure OpenAI and Azure AI Foundry, Anthropic Claude and Microsoft Copilot Studio, and choose per use case based on quality, cost, data residency and the customer’s existing Microsoft agreements. ### What is MCP and why does it matter? The Model Context Protocol (MCP) is an open standard for connecting AI models to tools and data. We build MCP servers that expose your systems, such as Dataverse or internal APIs, to AI agents in a controlled, permission-aware way, so the same integration works across several AI clients. ### How do you keep AI agents safe? Every agent is security-trimmed to the user it acts for, cites the records it used, logs its actions, needs approval for writes, and can be switched off per environment. We also test agents against evaluation sets before and after each change. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/ai-agents/ --- # Copilot & Cowork > Microsoft 365 Copilot is Microsoft’s AI assistant across Word, Excel, Outlook and Teams. Copilot Studio lets you build your own agents, and Cowork lets Copilot carry out longer, multi-step tasks for you. Licences alone do not create value. We prepare your data and permissions, build Copilot Studio agents grounded in your business, and set up Cowork so AI can carry out multi-step work across your Microsoft 365 apps under your rules. ## What you get - Copilot answers grounded in your own documents and records - Agents that do the work, not just chat about it - Oversharing risks found and fixed before rollout ## How we help ### Copilot readiness SharePoint and permission hygiene, sensitivity labels and a rollout and adoption plan. ### Copilot Studio agents Custom agents over Dataverse, SharePoint and line-of-business systems, with actions and connectors. ### Cowork enablement Configure Cowork to run long-running, multi-step tasks across Outlook, Teams, Excel and more, with human review built in. ### Copilot in Dynamics 365 Turn on and tune Copilot features across Sales and Customer Service. ## Example use cases - **HR & policy assistant**: Staff ask questions in Teams and get answers that cite the actual policy document. - **Account briefings**: Cowork pulls CRM, email and meeting history into a briefing before every client call. ## Tools we use Microsoft 365 Copilot, Copilot Studio, Copilot Cowork, Microsoft Purview, SharePoint, Microsoft Graph. ## Frequently asked questions ### What does a Copilot readiness assessment cover? We review who can access what in SharePoint, OneDrive and Teams, find overshared content, check sensitivity labels and ownership, and produce a prioritised fix list and a pilot plan. Copilot can see whatever each user can see, so permissions come first. ### What is the difference between Microsoft 365 Copilot and Copilot Studio? Microsoft 365 Copilot is the assistant built into Office apps and Teams. Copilot Studio is the tool for building your own agents with your own knowledge sources, actions and connectors, including agents that work over Dataverse and Dynamics 365. ### Can Copilot work with our Dynamics 365 or Dataverse data? Yes. Copilot features are built into Dynamics 365 Sales and Customer Service, and Copilot Studio agents can query and act on Dataverse tables, always respecting each user’s security roles. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/copilot-cowork/ --- # Dynamics 365 CRM > Dynamics 365 Customer Engagement is Microsoft’s CRM suite. It covers Sales, Customer Service and Field Service and runs on Dataverse, the same data platform as the Power Platform. We implement, extend and rescue Dynamics 365 Customer Engagement. That covers clean data models, sensible security and automation people actually use, with nothing customised for its own sake. ## What you get - One view of every customer, deal and case - Pipelines and SLAs that report themselves - Fewer clicks for the people doing the work ## How we help ### Implementation Discovery, solution design and rollout of D365 Sales, Customer Service and Field Service, configured before customised. ### Customisation & extension Plugins (C#), PCF controls, web resources, business process flows and Dataverse model design. ### Migration From spreadsheets, Salesforce, legacy CRM or on-premises Dynamics, with mapped, validated and reconciled data. ### Health check & rescue An audit of an existing org covering performance, security roles, technical debt and adoption, plus a prioritised fix list. ## Example use cases - **Lead-to-cash visibility**: Qualify leads, track opportunities and hand over to delivery without re-keying data. - **Omnichannel support desk**: Cases from email, web and chat with routing, SLAs and a knowledge base. ## Tools we use Dynamics 365 Sales, Customer Service, Field Service, Dataverse, C# plugins, PCF, Power Automate. ## Frequently asked questions ### Which Dynamics 365 apps do you work with? We focus on Dynamics 365 Customer Engagement: Sales, Customer Service and Field Service, together with the underlying Dataverse platform and Power Platform extensions. ### Can you fix an existing Dynamics 365 system? Yes. Our health check reviews performance, security roles, customisations, plugins and adoption, then gives you a prioritised fix list. We can carry out the fixes or support your team. ### Can you migrate us from Salesforce or spreadsheets? Yes. We map, cleanse and migrate accounts, contacts, opportunities, cases and history, with reconciliation reports so you can prove that nothing was lost. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/dynamics-365/ --- # Power Platform > Microsoft Power Platform is a low-code suite: Power Apps, Power Automate, Power Pages and Power BI, all built on Dataverse. It lets organisations build business apps and automation quickly, without losing enterprise security. Power Apps, Power Automate, Power Pages and Power BI on top of Dataverse. We build business apps quickly and add the ALM and governance that keep low-code from turning into sprawl. ## What you get - Paper and spreadsheet processes turned into apps - Approvals and hand-offs that run themselves - A Center of Excellence that keeps it all governed ## How we help ### Model-driven & canvas apps Choosing the right app type for the job, backed by a properly designed Dataverse schema. ### Power Automate Cloud flows, approvals, integrations and desktop flows (RPA) where APIs do not exist. ### Power Pages Secure external portals for customers, partners and suppliers. ### Governance & ALM Environment strategy, DLP policies, CoE Starter Kit, solution layering and pipelines. ## Example use cases - **Request & approval apps**: Purchase, leave, access or change requests with full audit history. - **Operational dashboards**: Power BI on Dataverse for live operational and executive reporting. ## Tools we use Power Apps, Power Automate, Power Pages, Power BI, Dataverse, Power Platform Pipelines, pac CLI. ## Frequently asked questions ### Should we build a canvas app or a model-driven app? Model-driven apps suit data-heavy processes with many related tables, security roles and forms. Canvas apps suit task-focused, highly designed or mobile-first experiences. Many solutions use both on the same Dataverse data. ### How do you stop Power Platform sprawl? With an environment strategy, data loss prevention (DLP) policies, the CoE Starter Kit for visibility, and source-controlled solutions deployed through pipelines instead of edited directly in production. ### Do we need Dataverse? For anything business-critical, usually yes. Dataverse gives you relational data, row-level security, auditing and a proper ALM story. SharePoint lists are fine for simple, low-risk cases. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/power-platform/ --- # Systems Integration > Systems integration means connecting the applications a business already uses, such as CRM, telephone systems, ERP and finance, so information is entered once and flows reliably wherever it is needed. We connect CRMs, telephone systems, ERP and line-of-business applications through their APIs: scheduled and event-driven syncs, with duplicate prevention, error handling and monitoring built in. Integrations run in your own Azure tenant and are documented so your team can own them. ## What you get - Customer data entered once, available everywhere - Calls identified by customer name, from up-to-date CRM contacts - Integrations you can monitor, rerun and trust ## How we help ### CRM and telephony integration Keep your phone system’s contacts in step with your CRM, so incoming calls show who is calling. Call activity can flow back to the CRM where the phone platform’s API supports it. ### API and data sync Scheduled and event-driven syncs between CRM, ERP and business apps, with field mapping, ID mapping to prevent duplicates, and safe re-runs. ### Dataverse and Dynamics 365 integration Custom APIs, plugins, virtual tables and Power Automate flows that connect Dataverse to the systems around it. ### Monitoring and resilience Health-check endpoints, retries, alerting and telemetry in Azure Monitor, so failures are spotted and fixed before users notice. ## Example use cases - **CRM-to-phone-system contact sync**: For a sales and service business, a .NET Azure Function keeps thousands of CRM customer contacts synchronised into a cloud VoIP telephone system twice a day, with a manual trigger for urgent updates. An ID-mapping store prevents duplicates and connection-test endpoints simplify support. Staff see the customer’s name as the phone rings. - **Order and invoice hand-off**: Won opportunities in CRM create orders in ERP, and invoice status flows back, so sales and finance work from the same facts. ## Tools we use Azure Functions (.NET), Logic Apps, Power Automate, REST & OData APIs, OAuth 2.0, Dataverse, Azure Monitor / OpenTelemetry. ## Frequently asked questions ### Can you integrate our CRM with our telephone system? Yes, if both systems offer an API. We typically sync contacts from the CRM into the phone system so incoming calls are identified by name, and, where the phone platform supports it, write call activity back to the CRM. ### Where do the integrations run? Usually as Azure Functions, Logic Apps or Power Automate flows in your own Azure or Power Platform tenant. Credentials stay in your environment, and you own the code and configuration. ### How do you prevent duplicates and handle failures? We keep an ID mapping between systems so each record is created once and updated afterwards. Every integration has retries, logging and alerting, and can be safely re-run. ### Do you only integrate Microsoft systems? No. Dynamics 365 and Dataverse are our core, but we integrate any system with a documented REST or OData API, including third-party CRMs, telephony platforms and ERPs. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/systems-integration/ --- # Infra & DevSecOps > DevSecOps means building security into every stage of software delivery, from code to deployment, instead of checking it at the end. On Azure, that starts with secure landing zones, strong identity and automated pipelines. Cloud infrastructure, identity and delivery pipelines that are secure by default. We build Azure landing zones, harden Entra ID, and automate build, test and deployment for code and Power Platform solutions alike. ## What you get - Repeatable infrastructure defined as code - Every release built, scanned and deployed by a pipeline - Identity and access that follow least privilege ## How we help ### Azure landing zones Subscriptions, networking, policy and cost management, all set up through Bicep or Terraform. ### Identity & security Entra ID, Conditional Access, PIM, Defender and Intune baselines. ### CI/CD & ALM GitHub Actions and Azure DevOps pipelines, including Power Platform solution deployments. ### DevSecOps Secret scanning, dependency and code scanning, IaC checks and security gates in the pipeline. ## Example use cases - **Power Platform ALM**: Source-controlled solutions deployed dev → test → prod by pipeline, not by hand. - **Secure-by-default tenant**: MFA, Conditional Access and admin separation set up for a growing company. ## Tools we use Microsoft Azure, Entra ID, Bicep, Terraform, GitHub Actions, Azure DevOps, Microsoft Defender, Intune. ## Frequently asked questions ### What is an Azure landing zone? A landing zone is a pre-configured Azure foundation, covering subscriptions, networking, identity, policy and cost controls, so new workloads land in a secure, consistent environment instead of being set up by hand each time. ### Can you set up CI/CD for Power Platform solutions? Yes. We put solutions in source control and deploy them dev → test → prod through GitHub Actions, Azure DevOps or Power Platform Pipelines, with automated checks and approvals. ### Bicep or Terraform? Bicep is Azure-native and simple for Azure-only estates. Terraform suits multi-cloud teams or those already using it. We work with both and follow what your team can maintain. --- Provider: KaizenAI, remote-first, working with clients worldwide. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ Page: https://kaizenai.systems/services/infra-devsecops/ --- # Kaizen PPM: project & portfolio management on Microsoft Dataverse > Kaizen PPM is a project and portfolio management (PPM) application built natively on Microsoft Power Platform and Dataverse. It takes work from idea to proposal, project and portfolio, and all data stays inside your own Microsoft 365 tenant. Take work from idea to proposal, project and portfolio in one Power Platform app. It includes the interactive views PMs expect, the governance PMOs need, and AI that explains instead of guessing. - Data stays 100% inside your own Microsoft 365 tenant - 10 capability areas, from idea to closure - Native Dataverse security, Power BI and Teams ## What does Kaizen PPM include? - **Portfolios & programs**: Group related projects and roll up health, cost and risk for oversight. - **Demand & idea intake**: Capture ideas, cost them as proposals and approve the right work before it starts. - **Interactive scheduling**: A Gantt chart with WBS, dependencies, baselines and drag-to-reschedule, plus a Kanban board. - **Financial management**: Budget, forecast and actuals by period, including labour cost from resource rates. - **Resource management**: Resource requests, capacity heatmaps, bookings and weekly timesheets. - **Risk & issues**: Drag-and-drop 5×5 risk matrix, issue tracking and mitigation to closure. - **Governance registers**: Change requests, decisions, deliverables, stakeholders and lessons learned. - **Status reporting**: Periodic RAG health reports per project, rolled up to an executive dashboard. - **Agile delivery**: Buckets, sprints and labels for teams that work iteratively. - **Data migration**: Built to receive data from Project Online, MS Project, Project for the Web and Planner. ## How does Kaizen PPM use AI? Kaizen PPM is being extended with Microsoft Copilot and Cowork. AI helps you get data in and answers out, and explains the numbers. It never quietly sets a project's health or makes the call for you. 1. **Capture** (You, assisted by AI): Create tasks, risks and status updates from plain language, so there is less typing. 2. **Surface** (You ask, AI retrieves): “Which of my projects are over budget?” is answered in plain English, security-trimmed to you. 3. **Interpret** (AI explains, never sets): Health and slippage are calculated by rules. AI explains them and cites the records behind them. 4. **Decide** (Always a human): AI can propose and rank options. A person approves before anything is committed. AI features are rolling out in waves. Each one is security-trimmed to the user, labelled when AI-assisted, and can be switched off per environment. ## Leaving Project Online or Planner? Every record in Kaizen PPM can carry its source-system lineage: where it came from, its original ID and when it last synced. Supported sources: - Microsoft Project Online - Project for the Web & Planner Premium - Microsoft Planner - Desktop MS Project (XML import & export) - Spreadsheets and CSV ## Frequently asked questions ### What is Kaizen PPM? Kaizen PPM is a project and portfolio management application built on Microsoft Power Platform and Dataverse. It covers idea intake, proposals, projects, programs and portfolios, with scheduling, financials, resources, risks and status reporting. ### Is Kaizen PPM an alternative to Microsoft Project Online? Yes. Kaizen PPM was designed to receive organisations moving off Project Online, Project for the Web, Planner and desktop MS Project. Every migratable record keeps its source-system lineage, and schedules can be imported from and exported to MS Project XML. ### Where is our project data stored? In your own Microsoft Dataverse environment, inside your Microsoft 365 tenant. Kaizen PPM uses standard Dataverse security, so your existing security roles, business units and auditing apply. ### What licences do users need? Kaizen PPM is a model-driven Power App, so each user needs a Power Apps premium or Dynamics 365 licence that includes Dataverse. Contact us for Kaizen PPM product pricing. ### Does Kaizen PPM work with Copilot and Cowork? Copilot and Cowork capabilities are being added in waves. AI helps capture data, answer questions and explain calculated health and slippage. It never sets a project’s status or makes decisions on its own, and every AI feature can be switched off. --- Request a demo: https://kaizenai.systems/contact/?topic=Kaizen%20PPM%20demo --- # AI-ready starter kits from KaizenAI Starter kits, not finished products: working foundations from our Dynamics 365 and Power Platform delivery. Each one shows the approach and saves weeks of set-up. We then plan each organisation's specific requirements with them and adapt the kit to fit. ## How it works 1. **Explore the kit**: See it running, read the code, and decide whether the foundation fits. 2. **Plan your requirements**: A short workshop maps your rules, data and users against the kit. AI helps us turn that into a gap list and a build plan. 3. **Customise and deploy**: We adapt the kit in your tenant, under your publisher, with your security model. You own the result. 4. **Improve continuously**: Start with the smallest useful version, then add rules, reports and automation as you learn. ## Configurable Form Engine > A starting point for moving model-driven form logic out of per-table JavaScript and into Dataverse configuration. Maturity: Foundation · in active development · Code: https://github.com/powerdynamite/configurable-form-engine **The problem.** Every Dynamics 365 project accumulates form scripts: the same "show this, require that" logic written again for each table, and every change needs a developer and a deployment. What you start with: - One web resource that applies visibility, requirement, editability, default values and value locking from Dataverse records - Configuration tables and an unmanaged solution to import - A Copilot Studio agent that drafts rules in plain language and confirms before any write What we tailor for you: - Rules for your own tables, forms and business processes - Extra action types your forms need - Who may change rules, and an approval step before they go live - The agent’s vocabulary, matched to your terminology AI-ready: Ships with a Copilot Studio agent, so rule changes can be planned and drafted conversationally. Built with: Dataverse, Model-driven apps, JavaScript web resource, Copilot Studio. ## Dataverse Audit Insights > A starting point for turning the Dataverse audit trail into usable reports and a record-restore tool. Maturity: Early foundation · validate in your environment · Code: https://github.com/powerdynamite/dataverse-audit-insights **The problem.** Dataverse records a detailed audit trail, but the audit table is not exposed through the Dataverse (TDS) connector, so most organisations never use it. What you start with: - A four-page Power BI report: user activity, tables, trends and compliance - A star-schema model fed from the Dataverse Web API - A build guide for a canvas app that shows field-level changes and restores previous values What we tailor for you: - Which tables and columns matter for your audit and compliance needs - Report pages and measures for your regulators, auditors or adoption goals - Who may restore values, with row-level security in Power BI - Refresh, retention and environment set-up for your tenant AI-ready: Built from plain, readable source (PBIP/TMDL and Markdown guides), so an AI assistant can help extend the model and report. Built with: Power BI (PBIP/TMDL), Dataverse Web API, Power Query, Power Apps, Custom connector. ## PP Governance Hub > A starting point for your own Power Platform governance hub, as a lighter, code-first alternative to the CoE Starter Kit. Maturity: Foundation · private, walkthrough on request **The problem.** Tenant-wide visibility of environments, apps, flows, agents and makers usually means installing and maintaining the full CoE Starter Kit, even when you only need part of it. What you start with: - Daily inventory sync of environments, apps, flows, agents, connectors, makers and licences into Dataverse - A findings engine with nine example governance rules (orphaned resources, stale apps, stopped flows, DLP gaps and more) - A Power Apps Code App to review, acknowledge and resolve findings What we tailor for you: - Governance rules and thresholds that match your policies - Notifications and escalation to the right owners - Your environment strategy, DLP policies and licence cost model - Tenant set-up: app registration, permissions and scheduling AI-ready: Includes AI build guidance (project instructions and a Power Platform setup skill), so Claude can help adapt it to your tenant. Built with: Power Apps Code App (React, TypeScript), Dataverse, C# plugins & Custom APIs, Power Automate, Power Platform Admin API. ## Frequently asked questions ### Are these finished products? No. They are starter kits: working foundations that show the approach and save weeks of set-up. Every organisation’s rules, data and security are different, so each kit is meant to be adapted before it is used in production. ### Can we use them ourselves? The public kits are on GitHub, and your team can explore and extend them. Test thoroughly in a development environment first. The governance hub is not public yet; ask us for a walkthrough. ### What does a customisation engagement look like? A short requirements workshop, a written gap list and plan, then customisation and deployment in your tenant in small increments. You own the resulting solution and source. ### What makes them “AI-ready”? Each kit is structured so AI can help: readable source and build guides, project instructions for assistants such as Claude, and in the form engine’s case a Copilot Studio agent. That makes planning and adapting the kit faster, while people review every change. Plan your version: https://kaizenai.systems/contact/ --- # About KaizenAI KaizenAI is a remote-first, AI-first Microsoft-ecosystem services and product company, working with clients worldwide (home base: Lahore, Punjab, Pakistan). We combine hands-on Dynamics 365 and Power Platform engineering with AI agents that handle the repetitive work. ## Why "Kaizen"? Kaizen (改善) is the Japanese idea of continuous improvement. It means making things a little better every day, involving the people who do the work, and treating each improvement as the starting point for the next. That is how we approach technology: ship something useful early, measure it, and keep improving it. ## Founder **Abdul Basit**, Founder & Principal Consultant. Abdul is a Power Platform and AI solutions architect with eight years of consulting experience, delivering remotely for Microsoft partners across Europe, Australia and the Middle East. He works across Dynamics 365 CE, Dataverse, Copilot Studio and agentic AI, and is the architect behind Kaizen PPM. He also builds AI-ready Power Platform starter kits (https://kaizenai.systems/starter-kits/) and Claude Code skills. Profiles: [website](https://powerdynamite.github.io/) · [linkedin](https://www.linkedin.com/in/mrabdulbasit1999/) · [github](https://github.com/powerdynamite) · [youtube](https://www.youtube.com/@powerdynamite365) · [medium](https://medium.com/@powerdynamite) ## What we value - **Kaizen: change for the better**: small, steady improvements that compound. - **Honest advice**: if a standard feature, a spreadsheet or doing nothing is the right answer, we say so. - **Built to hand over**: you own the source, the documentation and your data, in your own tenant. - **AI with a human in charge**: a person stays accountable for decisions. Contact: mrabdulbasit1999@gmail.com · +92 311 4095179 · +92 340 4949027 · https://kaizenai.systems/contact/ --- # Contact KaizenAI We usually reply within one working day (Pakistan time, UTC+5). - Email: mrabdulbasit1999@gmail.com - Phone: +92 311 4095179 - Phone: +92 340 4949027 - WhatsApp: https://wa.me/923114095179 - Works: Remote-first, working with clients worldwide (home base: Lahore, Punjab, Pakistan) - Contact form: https://kaizenai.systems/contact/ - LinkedIn: https://www.linkedin.com/company/kaizenai-systems/ Topics we can help with: Agentic AI, Copilot & Cowork, Dynamics 365 CRM, Power Platform, Systems Integration, Infra & DevSecOps, Kaizen PPM demo. --- # Claude Cowork vs Copilot Cowork plugins: same idea, different packaging > Anthropic and Microsoft both call their agentic workspace "Cowork", and both use plugins. Here is what carries over between them and what does not. By Abdul Basit · Published 2026-11-01 · Category: Agentic AI Source: https://kaizenai.systems/blog/claude-cowork-vs-copilot-cowork-plugins/ Two products are called "Cowork". **Claude Cowork** is Anthropic's agentic workspace in the Claude app. **Copilot Cowork** is Microsoft's, built with Anthropic and running inside Microsoft 365. Both can be extended with **plugins**, and both use the same underlying idea: package instructions and tool access so the AI behaves like a specialist for your role, team or company. If you build plugins, or are choosing between the two, this is what you need to know. ## What a plugin is in each product **Claude Cowork.** Anthropic introduced Cowork plugins on 30 January 2026. A plugin can bundle four things: - **Skills:** domain knowledge Claude applies automatically when it's relevant - **Connectors:** MCP connections to your tools, such as your CRM or knowledge base - **Slash commands:** actions you trigger yourself, like `/sales:call-prep` - **Sub-agents:** specialised helpers for parts of a task Everything is file-based, so plugins are easy to read, edit, version and share. Anthropic open-sourced eleven starter plugins (Sales, Finance, Legal, Marketing, Customer support, Product management, Data, Productivity, Enterprise search, Biology research, and a plugin for creating plugins) in its `knowledge-work-plugins` repository. **Copilot Cowork.** Microsoft's plugins contain **skills and connectors**, packaged as a standard **Microsoft 365 app package** (a `manifest.json` plus a `skills/` folder) and deployed by an admin through the Microsoft 365 admin center. See our [Copilot Cowork plugins guide](https://kaizenai.systems/blog/copilot-cowork-plugins-and-skills-explained/) for the details. ## Side by side | | Claude Cowork | Copilot Cowork | |---|---|---| | Skills (`SKILL.md`) | Yes | Yes, same format | | Connectors (MCP) | Yes | Yes (Streamable HTTP over HTTPS) | | Slash commands | Yes | Not yet supported | | Sub-agents | Yes | Not yet supported | | Hooks | Supported in Claude's plugin format | Not yet supported | | Packaging | Plugin folder | Microsoft 365 app package (`.zip` + manifest) | | Distribution | Install in Cowork, or upload your own | Microsoft 365 admin center, Microsoft 365 App Store | | Governance | Anthropic plans and admin settings | Microsoft 365 admin, Purview, Copilot Credit limits | | Data it sees best | What you connect | Your Microsoft 365 graph, plus what you connect | ## What carries over The good news is that **skills are portable**. A well-written `SKILL.md` with a clear `name`, a `description` full of trigger phrases and step-by-step instructions works in both products. MCP connectors are portable too, as long as they meet Microsoft's transport and authentication requirements. Microsoft documents a conversion path that imports a Claude plugin's skills and `.mcp.json`. What doesn't carry over today is slash commands, sub-agents and hooks. If your Claude plugin relies on them, move that logic into skill instructions, or into the connector itself, before targeting Copilot Cowork. ## Which should you build for? - **Your organisation runs on Microsoft 365** and wants central governance, Purview coverage and billing through Copilot Credits: **build for Copilot Cowork.** - **You want the richest plugin model today** (commands, sub-agents) or your team works mainly in Claude: **build for Claude Cowork.** - **You need both:** keep skills and MCP servers as the shared core, and package them twice. Writing the skill once and reusing it is the real advantage of a shared format. --- *We design portable skills and MCP connectors that work in both. [Start a conversation](https://kaizenai.systems/contact/?topic=Agentic%20AI).* **Sources:** [Customize Cowork with plugins](https://claude.com/blog/cowork-plugins) (Anthropic, January 2026) · [anthropics/knowledge-work-plugins](https://github.com/anthropics/knowledge-work-plugins) (GitHub) · [Build plugins for Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/cowork-plugin-development) (Microsoft Learn) --- # Copilot Chat, Cowork, Copilot Studio or Power Automate: which should you use? > Microsoft now offers four ways to put AI and automation to work. A simple guide to choosing the right one for each job, and when to combine them. By Abdul Basit · Published 2026-11-01 · Category: Copilot Source: https://kaizenai.systems/blog/copilot-chat-cowork-studio-power-automate-which-to-use/ Microsoft offers four overlapping ways to get work done with AI and automation. Pick the wrong one and you pay too much, get unpredictable results, or build something nobody uses. **The short answer: choose based on how predictable the work is and who starts it.** ## The four options at a glance | Tool | Best for | Who starts it | How predictable | |---|---|---|---| | **Copilot Chat** | Quick questions, summaries, drafts | A person, one question at a time | Varies; a person reviews every answer | | **Copilot Cowork** | Multi-step knowledge work with a finished deliverable | A person describes an outcome, or an event prompts a proposed task | Flexible; the AI plans the steps | | **Copilot Studio** | Custom agents for a specific audience or process | Users of that agent, in Teams, a website or Microsoft 365 | Guided; you design the topics, knowledge and actions | | **Power Automate** | Fixed, rules-based processes | Triggers such as a new record, a schedule or an approval | Fully predictable; same steps every time | ## How to choose **Ask: "Would two good employees do this task the same way every time?"** - **Yes, exactly the same way.** Use **Power Automate**. Approvals, data syncs and notifications don't need AI judgement, and deterministic flows are cheaper and easier to audit. - **Mostly, but it needs reading and judgement.** Use a **Copilot Studio agent**. You define the knowledge sources and actions, so behaviour stays within bounds. A good fit for an HR policy assistant or a customer-service triage agent. - **No, it depends on the situation and spans several systems.** Use **Copilot Cowork**. Account briefings, budget reviews and comparing large document sets are where its planning ability earns its cost. - **It's a one-off question.** Use **Copilot Chat**. It's quick and cheap. ## They work best together Real solutions often combine them. A Power Automate flow detects a new high-value case in Dynamics 365. Cowork prepares a briefing pack from the CRM, email history and contracts. A person reviews it and decides the next step. Each tool does what it is good at, and a person stays accountable for the decision. That matches our [human-in-the-loop principle](https://kaizenai.systems/blog/ai-agents-human-in-the-loop/). ## Three mistakes we see 1. **Using an AI agent for a fixed process.** If the steps never change, a flow is cheaper and more reliable. 2. **Building a custom agent when Cowork plus a skill would do.** A well-written skill file is often faster to deliver than a full Copilot Studio project. See [Cowork plugins and skills](https://kaizenai.systems/blog/copilot-cowork-plugins-and-skills-explained/). 3. **Skipping measurement.** Track time saved and cost per task for every option, especially usage-billed ones like Cowork. --- *Not sure which fits your process? [Describe it to us](https://kaizenai.systems/contact/?topic=Copilot%20%26%20Cowork) and we'll suggest the simplest option that works.* **Sources:** [Copilot Cowork is now generally available](https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/16/copilot-cowork-is-now-generally-available/) (Microsoft) · [Microsoft Copilot Studio documentation](https://learn.microsoft.com/en-us/microsoft-copilot-studio/) (Microsoft Learn) · [Power Automate documentation](https://learn.microsoft.com/en-us/power-automate/) (Microsoft Learn) --- # Rolling out Copilot Cowork safely: a 10-step checklist for IT admins > Copilot Cowork is off by default, which gives you time to prepare. A practical checklist covering permissions, pilots, spending limits, Purview and plugins. By Abdul Basit · Published 2026-11-01 · Category: Copilot Source: https://kaizenai.systems/blog/copilot-cowork-rollout-checklist-for-admins/ [Copilot Cowork](https://kaizenai.systems/blog/what-is-microsoft-365-copilot-cowork/) doesn't just answer questions. It acts: it edits files, runs multi-step tasks in the cloud and, with plugins, works inside your CRM and ERP. Microsoft ships it **off by default**, so admins decide when and how it arrives. Use that time well. Here is the checklist we work through with clients. ## Before you switch it on **1. Fix oversharing first.** Cowork works with whatever each user can access. Find SharePoint sites and Teams shared with "Everyone", and "anyone with the link" sharing on sensitive libraries, and fix them. This is the [single biggest Copilot readiness task](https://kaizenai.systems/blog/copilot-readiness-permissions-first/). **2. Check sensitivity labels.** Cowork carries labels through to everything it produces. If your HR, finance and legal content isn't labelled, the protections have nothing to apply to. **3. Confirm licensing.** Every Cowork user needs a Microsoft 365 Copilot licence. Usage is then billed separately in **Copilot Credits**, so agree in advance who owns that budget. ## The pilot **4. Choose a small, real pilot group.** Five to fifteen people with repetitive, multi-step knowledge work, such as finance close, account planning or service escalations. Avoid a pilot made up only of IT staff. **5. Turn Cowork on for that group only.** Use group-based enablement rather than enabling it for the whole tenant. **6. Set spending limits and alerts.** Cowork supports limits at tenant, group and user level, plus usage alerts. Start conservatively; it's easier to raise a limit than to explain an unexpected invoice. **7. Define three to five target tasks and measure them.** For each one, record the time it takes today, the time with Cowork, credits consumed and quality. That evidence is your business case. ## Governance **8. Confirm Purview coverage.** Audit, eDiscovery, Data Lifecycle Management, Insider Risk Management and Communication Compliance all extend to Cowork. Check that your retention and review policies cover the new content it creates. Microsoft has said data loss prevention (DLP) controls are coming to Cowork, so plan for them. **9. Control plugins deliberately.** Custom plugins are uploaded and approved through the Microsoft 365 admin center. Keep a short allow-list, review each plugin's connectors (what can it read? what can it write?), and make sure no credentials are embedded in skill files. Our [plugins guide](https://kaizenai.systems/blog/copilot-cowork-plugins-and-skills-explained/) covers what to look for. **10. Write a one-page acceptable-use note.** Explain what Cowork is for, what users must check before sharing its output, and how to report something unexpected. Keep it short enough that people actually read it. ## After the pilot Review the measurements, adjust the spending limits, add the plugins your pilot users asked for, and expand group by group. That's the kaizen approach: small, measured steps rather than a big-bang rollout. --- *Want a second pair of eyes on your Cowork rollout? [Book a free consultation](https://kaizenai.systems/contact/?topic=Copilot%20%26%20Cowork).* **Sources:** [Copilot Cowork is now generally available](https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/16/copilot-cowork-is-now-generally-available/) (Microsoft) · [What's new in Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/whats-new) (Microsoft Learn) · [Build plugins for Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/cowork-plugin-development) (Microsoft Learn) --- # CRM and telephony integration: a practical guide (and what we learned building one) > Connecting your CRM to your phone system means calls arrive with a name, and conversations end up on the customer record. The patterns, the pitfalls, and how to plan one. By Abdul Basit · Published 2026-11-01 · Category: Systems Integration Source: https://kaizenai.systems/blog/crm-telephony-integration-guide/ **CRM and telephony integration connects your customer database to your phone system.** Incoming calls show who is calling, staff can dial from the customer record, and calls are logged against the right account. Done well, it removes a surprising amount of daily friction: no "sorry, who am I speaking to?", no copying numbers by hand, no calls that never reach the CRM. This guide covers the main integration patterns, the decisions to make before you build, and the lessons from a CRM-to-VoIP integration we delivered recently. ## What "integration" can mean: four levels Not every business needs everything. Pick the level that solves your actual problem. | Level | What users get | Typical effort | |---|---|---| | **1. Contact sync** | The phone system knows your customers, so calls display a name instead of a number | Low to medium | | **2. Click-to-dial** | Call a customer directly from the CRM record | Low, often built into the phone vendor's app | | **3. Screen pop** | An incoming call opens the matching CRM record automatically | Medium | | **4. Call logging** | Calls (time, duration, direction, notes, sometimes recordings) are saved to the CRM record | Medium to high | **Level 1 is often the best first step.** It is quick to deliver, visible to every user from day one, and it produces the clean, matched data that levels 3 and 4 rely on later. ## Decisions to make before you build ### 1. Which system is the master? For customer contacts it is almost always the **CRM**. Syncing one way, from CRM to phone system, is simpler and avoids conflicts. Two-way sync sounds attractive, but it needs rules for every conflict: what happens when a number is edited in both places? ### 2. Which records should sync? Not every CRM contact belongs in the phone system. Agree a filter, such as active customers with a customer number or contacts with at least one phone number, so the phone directory stays useful rather than cluttered. ### 3. How will records be matched? This is where most integrations go wrong. You need a **stable key** linking each CRM record to its counterpart in the phone system, stored in an **ID mapping**. Matching on names or phone numbers alone produces duplicates as soon as anything changes. ### 4. How fresh does the data need to be? - **Scheduled sync** (for example, twice a day) is simple, robust and enough for most contact directories. - **Event-driven sync** (webhooks or change notifications) is near real-time but has more moving parts. - Whichever you choose, add a **manual "sync now" trigger** for urgent changes. ### 5. Where will it run, and who owns it? We usually build integrations as **Azure Functions** or **Logic Apps** in the client's own Azure tenant, or as **Power Automate** flows for simpler cases. Credentials stay in the client's environment, and they own the code. ## Pitfalls we see again and again - **Phone number formats.** "0300 1234567", "+92 300 1234567" and "00923001234567" are the same number. Normalise numbers to international (E.164) format before syncing and matching, or caller identification will silently fail. - **Duplicates on re-run.** Every sync must be safe to run twice. With an ID mapping, the second run updates records instead of creating new ones. - **Deletions.** Decide what happens when a customer is deactivated in the CRM: remove the contact, hide it, or leave it. Write that rule down. - **API limits and paging.** Phone and CRM APIs page their results and limit request rates. Fetch in pages, back off when throttled, and allow long-running jobs enough time to finish. - **Secrets in code.** API keys and OAuth client secrets belong in application settings or Azure Key Vault, never in source control. - **Silent failure.** If nobody notices when a sync stops, staff lose trust in the data. Add logging, alerts and a simple health check. ## A real example For a sales and service business, we built a one-way contact sync from their CRM into a cloud VoIP telephone system: - A **.NET Azure Function** runs **twice a day**, with an **HTTP endpoint for manual syncs**. - Only customers with a **customer number** are synced, which keeps thousands of relevant contacts and leaves out incomplete records. - An **ID mapping** (customer number → phone-system contact ID) makes every run idempotent: create once, update afterwards, never duplicate. - The phone system's API uses **OAuth 2.0**. Tokens are handled per request and never shared across clients. - **Connection-test endpoints** let support staff check both systems in seconds, and **OpenTelemetry** sends logs and metrics to Azure Monitor. The result for staff is simple: when the phone rings, they see the customer's name. That small change is what makes the integration worth building, and it sets up the next steps, screen pop and call logging. ## Integrating with Dynamics 365 If your CRM is Dynamics 365 or another Dataverse-based app, you have extra options: - **Microsoft's Channel Integration Framework** lets a telephony vendor's softphone run inside model-driven apps, enabling screen pop and click-to-dial within the CRM. - **Dataverse APIs and plugins** make it straightforward to create **phone call activities** on the right account or contact. - **Power Automate** can handle lighter, event-driven tasks, such as creating a follow-up task after a missed call. Check first whether your phone vendor already offers a Dynamics 365 connector. Sometimes the best integration is configuring one that exists, then filling the gaps with a small amount of custom code. ## A simple plan to get started 1. **List the systems and their APIs.** No API means no clean integration, so confirm this first. 2. **Agree the master system, the filter and the matching key.** 3. **Start with one-way contact sync**, scheduled, with a manual trigger and monitoring. 4. **Run it in parallel for a week** and check a sample of records by hand. 5. **Then add screen pop or call logging**, one capability at a time. That's the kaizen approach to integration: a small, reliable first step, then steady improvement. **Next:** [Screen pop and call logging in Dynamics 365](https://kaizenai.systems/blog/screen-pop-and-call-logging-dynamics-365/): the options, and how to build them. --- *Want your CRM and phone system working together? See our [Systems Integration service](https://kaizenai.systems/services/systems-integration/) or [describe your setup](https://kaizenai.systems/contact/?topic=Systems%20Integration).* **Further reading:** [Channel Integration Framework overview](https://learn.microsoft.com/en-us/dynamics365/customer-service/channel-integration-framework/overview-channel-integration-framework) (Microsoft Learn) · [Azure Functions documentation](https://learn.microsoft.com/en-us/azure/azure-functions/) (Microsoft Learn) · [E.164 numbering plan](https://www.itu.int/rec/T-REC-E.164) (ITU) --- # Copilot Cowork plugins and skills, explained (and how to build your own) > Plugins connect Copilot Cowork to your business systems, and skills teach it your way of working. How they are packaged, deployed and secured. By Abdul Basit · Published 2026-11-01 · Category: Copilot Source: https://kaizenai.systems/blog/copilot-cowork-plugins-and-skills-explained/ Out of the box, [Copilot Cowork](https://kaizenai.systems/blog/what-is-microsoft-365-copilot-cowork/) knows your Microsoft 365 content. **Plugins and skills are how you teach it about everything else**: your CRM, your ERP, and the specific way your team prepares a board pack or handles a complaint. ## Skills and connectors: the two building blocks A Cowork plugin is made of one or both of these: - **Skills** are written instructions in a `SKILL.md` file. A skill tells Cowork how to do a particular job: the steps, the format and the checks. Cowork picks up a skill automatically when a request matches its description. - **Connectors** link Cowork to external systems through the [Model Context Protocol (MCP)](https://kaizenai.systems/blog/model-context-protocol-explained-for-business-leaders/). That is how Cowork can read a Salesforce opportunity or create a ServiceNow ticket. A skill without a connector suits workflows that are mostly prompting: "write our weekly status report in this format". A connector without a skill gives Cowork data access but no method. **The most useful plugins combine both.** ## Plugins you can use today As of November 2026, Microsoft offers first-party plugins for **Dynamics 365 Sales, Customer Service and ERP**, and for **Fabric**, plus Accounting, Finance and Capital Markets plugins. There are partner plugins from companies including Harvey, LSEG, Miro, monday.com, Moody's and Morningstar, with more announced. There are also connectors to Jira, Salesforce, ServiceNow, SAP, Workday and Zendesk. ## How a custom plugin is packaged Cowork plugins use the same **Microsoft 365 app package** format as Teams apps and Copilot agents: a `.zip` file that looks like this: ``` my-plugin.zip ├── manifest.json # Microsoft 365 unified app manifest ├── color.png # 192×192 icon ├── outline.png # 32×32 icon └── skills/ └── board-pack-review/ ├── SKILL.md └── references/ # optional supporting files ``` Each `SKILL.md` starts with a short header: ```markdown --- name: board-pack-review description: Review a draft board pack against our finance and governance checklist. Use when asked to check, review or QA a board pack. --- 1. Confirm every figure in the summary matches the appendix tables. 2. Flag any decision paper missing a risk section. 3. ... ``` Microsoft Learn sets a few rules worth knowing: - `name` must be kebab-case and **match the folder name exactly**. - `description` should **include the trigger phrases** people will actually use, because that is how Cowork decides when to apply the skill. - A skill can have up to **20 companion files** (5 MB each, 10 MB in total). - Plugin skills **cannot override Cowork's built-in skills**, so don't duplicate them. ## How plugins are deployed An admin uploads the package through **Microsoft 365 admin center → Manage apps → Upload custom app**. It then appears in Cowork under *Sources & Skills → Plugins*. Microsoft also provides a preview command-line tool for building and sharing test versions before a full rollout. ## Security: what to get right - **Never put secrets in `SKILL.md`.** Connectors authenticate through the plugin vault (OAuth) or dynamic client registration. - **Annotate your tools.** Cowork treats MCP tools without annotations as destructive and asks the user to confirm, which is the safe default. - **Scope connectors narrowly.** Expose read operations broadly and write operations sparingly. - **Plan around Information Barriers:** Microsoft notes that Purview Information Barriers aren't supported for plugin management. ## Already built plugins for Claude Cowork? Microsoft documents a way to convert Claude plugins, and the skill format is shared. Some parts don't carry over yet, such as slash commands, sub-agents and hooks. We compare the two in [Claude Cowork vs Copilot Cowork plugins](https://kaizenai.systems/blog/claude-cowork-vs-copilot-cowork-plugins/). --- *We build Cowork plugins and MCP connectors for Dynamics 365, Dataverse and line-of-business systems. [Tell us what you need](https://kaizenai.systems/contact/?topic=Copilot%20%26%20Cowork).* **Sources:** [Build plugins for Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/cowork-plugin-development) (Microsoft Learn) · [What's new in Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/whats-new) (Microsoft Learn) · [Copilot Cowork is now generally available](https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/16/copilot-cowork-is-now-generally-available/) (Microsoft) --- # Model Context Protocol (MCP), explained for business leaders > MCP is the standard that lets AI assistants such as Copilot and Claude safely use your business systems. What it is, why it matters, and what to ask before you connect anything. By Abdul Basit · Published 2026-11-01 · Category: Agentic AI Source: https://kaizenai.systems/blog/model-context-protocol-explained-for-business-leaders/ **The Model Context Protocol (MCP) is an open standard for connecting AI assistants to tools and data.** It does for AI what USB did for peripherals: build one connector, and any assistant that supports MCP can use it. Anthropic introduced it in late 2024, and it is now supported across the industry, including in Claude, Microsoft Copilot Studio and both [Claude Cowork and Copilot Cowork](https://kaizenai.systems/blog/claude-cowork-vs-copilot-cowork-plugins/). ## Why it matters to your business Before MCP, every AI product needed its own custom integration with every system. Connecting three AI tools to five business systems meant fifteen integrations, each with its own security model. With MCP, each system is exposed once, as an **MCP server**, and every MCP-capable assistant can use it. In practice, that means: - **Less integration work:** one Dynamics 365 or ERP connector serves several AI tools. - **No lock-in:** if you switch AI vendors, your integrations stay put. - **One place to control access:** the MCP server decides what the AI can see and do. ## How it works, in one paragraph An MCP server publishes a list of **tools**, such as "search accounts", "get open cases" or "create a draft invoice". Each tool has a description and defined inputs. When the AI assistant needs information or wants to act, it calls the right tool. The server runs the request against your system, using the permissions of the person the AI is acting for, and returns the result. The AI never gets direct database access; it only gets the tools you chose to publish. ## Five questions to ask before connecting a system 1. **Whose permissions does it use?** The right answer is the signed-in user's permissions, never a shared admin account. 2. **Which tools can write data?** Keep read tools broad and write tools few. Clients such as Copilot Cowork ask the user to confirm actions on tools that aren't marked as safe. 3. **Where are secrets kept?** In a vault or an OAuth flow, never in prompts, skill files or config checked into source control. 4. **What is logged?** You should be able to answer "what did the AI do, and for whom?" weeks later. 5. **Who owns it?** An MCP server is production software. It needs an owner, monitoring and updates. ## Where to start Pick one system that people constantly look things up in, such as CRM, the service desk or project data, and expose a handful of **read-only** tools. Measure how often they are used and how much time they save. Add write tools later, one at a time, each with a confirmation step. This is the approach we follow in our [agentic AI work](https://kaizenai.systems/services/ai-agents/), and it is how [Kaizen PPM](https://kaizenai.systems/products/kaizen-ppm/) is being opened up to Copilot and Cowork. --- *Thinking about an MCP server for Dynamics 365 or Dataverse? [Let's talk](https://kaizenai.systems/contact/?topic=Agentic%20AI).* **Sources:** [Model Context Protocol](https://modelcontextprotocol.io/) (official site) · [Build plugins for Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/cowork-plugin-development) (Microsoft Learn) · [Customize Cowork with plugins](https://claude.com/blog/cowork-plugins) (Anthropic) --- # Screen pop and call logging in Dynamics 365: your options, and how to build them > Open the right customer record the moment the phone rings, and save every call to the timeline. The built-in options, the Channel Integration Framework, and the details that make it work. By Abdul Basit · Published 2026-11-01 · Category: Dynamics 365 Source: https://kaizenai.systems/blog/screen-pop-and-call-logging-dynamics-365/ In our [CRM and telephony integration guide](https://kaizenai.systems/blog/crm-telephony-integration-guide/) we described four levels of integration: contact sync, click-to-dial, screen pop and call logging. This post covers the last two for **Dynamics 365 and other model-driven apps**: - **Screen pop:** when a call comes in, Dynamics 365 opens the matching contact, account or case automatically. - **Call logging:** every call is saved as a **phone call activity** on the right record's timeline, with direction, duration and notes. Together they mean an agent never starts a conversation blind, and a manager never has to ask "did anyone call this customer?". ## Option 1: use what Microsoft already provides Before building anything, check whether a built-in option covers your needs. - **Microsoft Teams dialer in Dynamics 365 Sales.** Sellers can call from a record using Teams. After each call, Dynamics 365 creates a phone call activity on the record's timeline. For incoming calls, if the number matches a record, the *Call From* and *Regarding* fields are filled in automatically. Limitations to be aware of: the dialer doesn't support custom tables, and internal calls aren't tracked. - **Sales Hub dialer.** Microsoft has announced a built-in dialer in its 2026 release waves, with inbound and outbound calling and automatic call logging. Check availability in your region and licence before planning around it. - **Dynamics 365 Contact Center and Customer Service** include a native voice channel for service teams, with call routing and the customer record presented to the agent. **If your telephony is Teams Phone, or you can move to it, start here.** The rest of this post is for the common case where you already have another phone system and want to keep it. ## Option 2: your phone system, embedded with the Channel Integration Framework Microsoft's **Channel Integration Framework (CIF)** is the supported way to embed a third-party softphone inside model-driven apps. The softphone runs in a panel next to the CRM, and talks to Dynamics 365 through a JavaScript API. CIF 2.0 adds support for multi-session experiences such as Customer Service workspace. Many telephony vendors already ship a CIF-based connector. **Check your vendor's marketplace first.** If one exists, configuring it is far cheaper than building. If not, or if it doesn't do what you need, a custom widget is a well-trodden path. ### How screen pop works with CIF 1. The phone system notifies the softphone widget of an incoming call and the caller's number. 2. The widget asks Dynamics 365 to find and open the matching record using `searchAndOpenRecords`. 3. If exactly one record matches, it opens. If several match, the user picks one. If none match, the widget can open a new contact form with the number pre-filled. A simplified example, looking up a contact by a normalised phone number: ```js // e164 = caller number normalised to international format, e.g. +923001234567 const query = `?$select=fullname&$filter=mobilephone eq '${e164}' or telephone1 eq '${e164}'`; Microsoft.CIFramework.searchAndOpenRecords('contact', query, false); ``` ### How click-to-dial works with CIF The widget enables click-to-act and registers a handler. When a user clicks a phone number in Dynamics 365, the handler receives the number and record context, and asks the phone system to place the call. ## Call logging: getting the phone call activity right Whichever option you choose, calls should land in Dynamics 365 as standard **phone call activities** (`phonecall`), not as notes or a custom table. That way they appear on timelines, in activity reports and in Copilot summaries without extra work. A well-formed call log sets: | Field | What to store | |---|---| | Subject | e.g. "Inbound call from +92 300 123 4567" | | Direction | Incoming or outgoing | | Phone number | The number dialled or received | | Call from / Call to | Activity parties: the contact and the user | | Regarding | The account, contact, opportunity or case the call was about | | Duration | Actual duration in minutes | | Status | Completed, then *Made* or *Received* | Create the activity when the call **ends**, not when it starts. That way duration and outcome are known, and abandoned calls don't clutter the timeline. If agents add notes during the call, save them into the activity description as part of the same step. ## The details that decide whether it works - **Normalise phone numbers.** Dynamics 365 stores numbers however people typed them: "0300-1234567", "+92 300 1234567", "(0300) 123 4567". Screen pop only works if the search can match. Normalise numbers to international (E.164) format on save, or keep a normalised search column alongside the display value. - **Decide what "match" means.** Contacts only, or accounts and leads too? Which phone fields count? Agree the search order up front. - **Handle no match and multiple matches.** These are normal, not edge cases. Design the screens for them. - **Respect security.** Screen pop and logging run as the signed-in user. If they can't see a record, it shouldn't pop. That's correct behaviour, so make sure roles allow what agents genuinely need. - **Think about recordings carefully.** Store links to recordings in the phone platform rather than copying audio into Dataverse, and follow your consent and retention rules. - **Measure adoption.** Report on how many calls are logged, and with a Regarding record. That tells you whether the integration is being used as intended. ## A sensible order of work 1. **Clean and normalise phone numbers** in Dynamics 365. Everything else depends on it. 2. **Check built-in and vendor options** before writing code. 3. **Add call logging first.** It works even without screen pop and builds history immediately. 4. **Then add screen pop and click-to-dial**, starting with one team. 5. **Review a week of logged calls** with the team and adjust matching rules. --- *Need your phone system working inside Dynamics 365? See our [Systems Integration](https://kaizenai.systems/services/systems-integration/) and [Dynamics 365](https://kaizenai.systems/services/dynamics-365/) services, or [describe your setup](https://kaizenai.systems/contact/?topic=Systems%20Integration).* **Sources:** [What is Channel Integration Framework 2.0?](https://learn.microsoft.com/en-gb/dynamics365/channel-integration-framework/v2/administer/overview-channel-integration-framework) (Microsoft Learn) · [Microsoft.CIFramework methods, CIF 2.0](https://learn.microsoft.com/en-us/dynamics365/customer-service/channel-integration-framework/v2/reference/microsoft-ciframework-v2) (Microsoft Learn) · [Call using Microsoft Teams in Dynamics 365 Sales](https://learn.microsoft.com/en-us/dynamics365/sales/call-using-microsoft-teams) (Microsoft Learn) --- # What is Microsoft 365 Copilot Cowork? A practical guide for business leaders > Copilot Cowork turns Copilot from a chat assistant into something that finishes multi-step work. What it is, what it costs, how it is governed, and how to start. By Abdul Basit · Published 2026-11-01 · Category: Copilot Source: https://kaizenai.systems/blog/what-is-microsoft-365-copilot-cowork/ **Copilot Cowork is the part of Microsoft 365 Copilot that carries out long-running, multi-step work for you.** You describe the outcome you want. Cowork makes a plan, works across your files, email, calendar and connected business systems, and comes back with a finished result rather than a draft. You can watch its progress and steer it as it goes. Microsoft opened Cowork to its Frontier early-access programme on 30 March 2026 and made it generally available on 16 June 2026. It was built with Anthropic, on the technology behind Claude Cowork. This guide sums up what matters as of November 2026. ## How is Cowork different from Copilot Chat? Copilot Chat is a conversation: you ask, it answers, and you do the work. Cowork is closer to delegating a task to a colleague. | | Copilot Chat | Copilot Cowork | |---|---|---| | Interaction | Question and answer | Describe an outcome and let it work | | Length of work | One turn at a time | Long-running and multi-step; it keeps running in the cloud when your laptop is off | | Output | Answers and drafts | Completed artefacts: edited spreadsheets, documents, briefings, analyses | | Tools | Mostly Microsoft 365 content | Microsoft 365, plus plugins and connectors to Dynamics 365, Salesforce, ServiceNow, SAP and more | | Triggers | You start it | You start it, or it can propose tasks when an email, Teams message or CRM event arrives | In the Microsoft 365 Copilot app, a toggle switches between Chat and Cowork. Cowork is also available on iPhone and Android through the Microsoft Copilot mobile app. ## What kind of work suits Cowork? Microsoft's own examples are a good guide. They include comparing thousands of files across two product versions, editing batch spreadsheets, ranking at-risk opportunities in a sales pipeline, preparing a monthly budget review and putting together a daily briefing. The pattern is the same each time: **work that needs several sources, several steps and a finished deliverable.** It is less suited to a quick factual question, where Chat is faster and cheaper, or to a fixed process that runs the same way every time, where Power Automate or a Copilot Studio agent is more predictable. ## What does it cost? Two things are required: 1. **A Microsoft 365 Copilot licence** for each user. 2. **Usage-based billing in Copilot Credits.** Each task's cost depends on model use, how much context is retrieved, tool calls and run time. At general availability, pay-as-you-go was listed at $0.01 per Copilot Credit, with volume discounts on a commitment plan. Cost per task varies a lot, so measure it during a pilot before you budget for a full rollout. ## How is it governed? This is where Cowork differs most from consumer AI tools: - **Cowork is off by default.** Admins choose when to turn it on and for which users or groups. - **Spending limits** can be set at tenant, group and user level, with usage alerts and reporting. - **It runs inside your Microsoft 365 trust boundary.** Sensitivity labels carry through end to end. Audit, eDiscovery, Data Lifecycle Management, Insider Risk Management and Communication Compliance all apply. Microsoft has also said data loss prevention (DLP) controls are coming to Cowork. - **Plugins are admin-controlled.** Custom plugins are uploaded and approved through the Microsoft 365 admin center. One caution: Cowork can only see what each user can see. Overshared SharePoint sites become very easy to find. That is why [Copilot readiness starts with permissions](https://kaizenai.systems/blog/copilot-readiness-permissions-first/). ## Which AI models does it use? Cowork lets you choose between models from Anthropic and OpenAI, and Microsoft adds new versions often. Treat the model list as something that changes month to month, and test your own tasks rather than relying on benchmarks. ## How to get started 1. **Tidy permissions** on the sites and teams the pilot group uses. 2. **Pick 5–10 people** with real, repetitive knowledge work, such as finance close, account planning or service escalations. 3. **Turn Cowork on for that group only**, with a modest spending limit. 4. **Write down three or four target tasks** and measure time saved against credits spent. 5. **Add plugins** for the systems those tasks touch. See our guide to [Copilot Cowork plugins and skills](https://kaizenai.systems/blog/copilot-cowork-plugins-and-skills-explained/). --- *Want help planning a Cowork pilot? [Talk to us](https://kaizenai.systems/contact/?topic=Copilot%20%26%20Cowork).* **Sources:** [Copilot Cowork is now generally available](https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/16/copilot-cowork-is-now-generally-available/) (Microsoft, June 2026) · [Copilot Cowork: now available in Frontier](https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/30/copilot-cowork-now-available-in-frontier/) (Microsoft, March 2026) · [What's new in Copilot Cowork](https://learn.microsoft.com/en-us/microsoft-365/copilot/cowork/whats-new) (Microsoft Learn) --- # Leaving Project Online: a practical migration checklist > What to inventory, what to clean up and what to move first when you take your project portfolio off Project Online and onto Power Platform. By Abdul Basit · Published 2026-10-28 · Category: Project Management Source: https://kaizenai.systems/blog/leaving-project-online-checklist/ Microsoft has announced the retirement of **Project Online**, and many PMOs are now asking the same question: *where does our portfolio go next, and how do we move it without losing years of history?* Here is the checklist we use before any migration to a Dataverse-based PPM solution such as [Kaizen PPM](https://kaizenai.systems/products/kaizen-ppm/). ## 1. Inventory what you actually have - **Projects**: active, closed and on-hold. How many are you really going to move? - **Enterprise custom fields and lookup tables**: these often carry the most business logic, and the most clutter. - **Resource pool**: named resources, generic roles, cost rates and calendars. - **Timesheets and actuals**: decide how far back you really need them. - **Reports**: list every Power BI or Excel report that reads from Project Online, because each one needs a new data source. ## 2. Decide what *not* to migrate A migration is the cheapest moment you will ever have to clean up. Archive closed projects to a read-only export, drop custom fields nobody has filled in for a year, and merge duplicate resources. > Rule of thumb: if nobody can tell you what a field is for, it does not move. ## 3. Keep the lineage Every migrated record should remember **where it came from**: its source system, original ID and when it last synced. That lets you reconcile counts, re-run a migration safely and answer audit questions later. Kaizen PPM's data model carries these lineage fields on every migratable table for exactly this reason. ## 4. Move in waves 1. Reference data first: resources, calendars and lookups. 2. Then a **pilot portfolio** of 5–10 live projects with their schedules. 3. Then the remaining active projects. 4. Finally, history and actuals, if you still need them. ## 5. Re-point reporting and train people The data being moved is only half the job. Rebuild your key reports on Dataverse, then run short role-based sessions for PMs, resource managers and executives. --- *Planning a move? [Talk to us](https://kaizenai.systems/contact/?topic=Kaizen%20PPM%20demo). We will review your current setup and tell you honestly how big the job is.* --- # Copilot readiness starts with permissions, not prompts > Microsoft 365 Copilot can see whatever each user can see. Fix oversharing before rollout, or Copilot will surface it for you. By Abdul Basit · Published 2026-10-20 · Category: Copilot Source: https://kaizenai.systems/blog/copilot-readiness-permissions-first/ The most common surprise in a Microsoft 365 Copilot rollout has nothing to do with AI. It is **permissions**. Copilot answers using the content each user already has access to. That is the right design, but it means every overshared SharePoint site, "Everyone except external users" link and forgotten Teams channel suddenly becomes *very easy to find*. ## A short readiness checklist 1. **Find overshared content.** Start with sites shared to large groups or the whole organisation, and with "anyone" links. 2. **Fix ownership.** Every site and team needs an accountable owner. Orphaned sites are where oversharing lives. 3. **Label what is sensitive.** Use Microsoft Purview sensitivity labels on HR, finance and legal content so protection travels with the file. 4. **Pilot with a small group.** Choose people with real, repetitive knowledge work and track what they use Copilot for. 5. **Measure adoption, not licences.** A licence that nobody uses is the most expensive kind. ## Then build on it Once the foundations are sound, the high-value work begins: **Copilot Studio agents** grounded in your own processes, and **Cowork** for multi-step tasks such as preparing a client briefing from CRM, email and meeting history. --- *Want a readiness review? [Get in touch](https://kaizenai.systems/contact/?topic=Copilot%20%26%20Cowork).* --- # Useful AI agents keep a human in the loop > A simple four-rung ladder for deciding what an AI agent should be allowed to do on its own, and what it must leave to a person. By Abdul Basit · Published 2026-10-12 · Category: Agentic AI Source: https://kaizenai.systems/blog/ai-agents-human-in-the-loop/ "Agentic AI" is everywhere, and so are demos of agents doing everything on their own. In real businesses, the agents that last are the ones people **trust**. Trust comes from being clear about what the agent may and may not do. We use a simple four-rung ladder. | Rung | What the agent does | Who is accountable | |---|---|---| | **1. Capture** | Gets data in with less typing: reads an email, fills a record | The person, assisted | | **2. Surface** | Finds and summarises on request | The person asks, the agent retrieves | | **3. Interpret** | Explains a result that rules or calculations produced | The agent narrates and never sets the value | | **4. Decide** | Chooses a course of action | **Always a human** | Most of the value sits on rungs 1–3. Rung 4 is where an agent can *propose* and *rank* options, but a person approves. ## Guardrails we build into every agent - **Security-trimmed**: an agent only sees what the user it acts for can see. - **Explainable**: every output cites the records or documents it used. - **Approval on writes**: nothing important is committed without a confirmation step. - **Auditable**: AI-assisted content is marked as such, with the confirming user. - **An off switch**: every feature can be disabled per environment. This is the same approach we take with the AI features in [Kaizen PPM](https://kaizenai.systems/products/kaizen-ppm/), and with the agents we build for clients. --- *Have a process you think an agent could take over? [Tell us about it](https://kaizenai.systems/contact/?topic=Agentic%20AI).*