Microsoft 365 + AI: what's actually useful in 2026
What's working in M365 + AI
Honest list as of mid-2026:
Copilot for Microsoft 365 (commercial)
The general Copilot license. €30/user/month roughly. Genuinely useful for: drafting emails, summarising threads, summarising meetings, drafting PowerPoint decks from data, answering questions over your SharePoint / OneDrive / Teams.
Underwhelming for: heavily-customised workflows specific to your business. Generic Copilot doesn't know your business well enough to be the wow product the marketing suggests.
Our recommendation: pilot with 5-10% of seats for 90 days. Measure actual time saved (not anecdotes). Roll out to specific roles if ROI clears the per-seat cost.
Copilot Studio (custom copilots)
Microsoft's low-code platform for building custom copilots, available in your tenant with Dataverse integration. Better than 2024 — handles RAG over SharePoint, has reasonable tool calling, integrates with Power Automate.
Works well for: internal Q&A agents grounded in SharePoint / Dataverse content, simple workflow orchestration via natural language.
Hits ceiling on: complex multi-step agents, custom tooling, advanced evals, deep customization.
Our recommendation: prototype here. Move to custom when you outgrow it. The pattern is similar to PowerApps — see PowerApps to production.
Power Platform AI Builder
The AI features inside Power Apps and Power Automate. Form processing, prediction, document classification, sentiment.
Works for: form processing on standard forms (W-9s, invoices in standard layouts), basic sentiment analysis, low-volume document classification.
Hits ceiling on: custom schemas, varied layouts, volume past a few thousand documents per month, models other than what Microsoft offers.
Our recommendation: prototype if it fits cleanly. For production document processing especially, custom usually wins. See How AI invoice processing works.
Azure OpenAI Service
OpenAI's models hosted in your Azure tenant. Same API as openai.com, with data residency in your region.
Strong for: organisations that need EU data residency, want SOC 2 / HIPAA compliance documented at the cloud provider level, or run AI alongside other Azure workloads.
Caveat: model versions sometimes lag openai.com by a few weeks. Latest Claude isn't there (different provider). For Anthropic models in Azure look at the AWS Bedrock parallel (Anthropic models on AWS).
Our recommendation: default for Microsoft-tenant clients building custom AI. Trade-off is real (slight model lag) but operational benefits often dominate.
Microsoft Graph + Dataverse for custom AI
This is the underrated category. Microsoft Graph exposes Outlook, Teams, SharePoint, OneDrive, calendar, contacts, files via a unified API. Dataverse holds your structured business data. Both have well-documented APIs.
Custom AI built on top of these reads your actual business context. Examples we've shipped:
- A sales-prep agent that pulls calendar + email + CRM history before each call, summarises, and writes a brief.
- An internal knowledge agent that reads SharePoint documents the user has access to (respecting M365 permissions) and answers grounded questions.
- An ops agent that reads Dataverse data, identifies anomalies, drafts Teams messages to the right channel.
These are the highest-ROI AI builds we ship on M365. They are also the most differentiated — generic Copilot does generic things; custom AI on your Graph + Dataverse does your things.
What's overhyped
The marketing exceeds reality on these:
"Copilot will transform your business"
Copilot is useful. It is not transformative for the average user. Effective adoption requires investment in training, change management, and identification of specific high-leverage workflows. Treating it as a magic productivity wand wastes money.
"Copilot Studio replaces custom development"
Sometimes. For simple custom copilots, yes. For anything where you need real evals, complex tooling, custom UI, or model choice, it doesn't. Don't commit to it as the long-term answer until you've prototyped.
"Power Platform AI handles everything"
It doesn't. AI Builder is a tier of capability between "nothing" and "custom." Useful for some things; not the destination for production AI workloads.
"Microsoft + OpenAI is the only valid combination"
Azure OpenAI is great. So is direct Anthropic. So is direct Google. Pick per task. Mix and match. Don't lock in your model layer to one vendor for reasons of tenant alignment.
Patterns we ship into M365 environments
A short list of integration shapes that earn their keep:
1. Custom agents reading Microsoft Graph
User-permissioned reads of email, calendar, files, Teams. Agent acts within the user's existing permission scope. Output goes back into Teams as a card or to the user as a notification.
Use for: per-user assistants (sales prep, calendar triage, weekly summary digests).
2. Custom agents reading Dataverse
Tenant-permissioned reads/writes against your structured business data. Often invoked from PowerApps or Power Automate.
Use for: operations agents that work alongside your existing Power Platform builds — anomaly detection, exception handling, intelligent routing.
3. Power Automate orchestrating AI
Power Automate as the workflow engine, with one or two AI calls in the flow for judgment-required steps. AI calls go to Azure OpenAI or via HTTP to a custom endpoint.
Use for: workflow automation that needs occasional intelligence — see AI agents vs automation.
4. Custom front-ends on Dataverse data
Next.js applications consuming Dataverse via the Web API, with AI features layered on. Replaces or supplements PowerApps where PowerApps hits its ceiling. See PowerApps to production.
Use for: customer-facing portals, high-volume internal tools, performance-sensitive workloads.
5. SharePoint-grounded chatbots
Internal Q&A agents grounded in SharePoint content, deployed as Teams apps. RAG pipeline reads SharePoint via Graph; respects M365 permissions; cites sources.
Use for: HR Q&A, policy Q&A, internal knowledge bases. See Conversational agents.
Cost considerations
| Layer | Typical cost |
|---|---|
| Copilot for M365 | €30/user/month |
| Copilot Studio | €200/tenant/month + per-conversation costs |
| Power Platform AI Builder | bundled or capacity-priced |
| Azure OpenAI | same as openai.com + Azure infra |
| Custom AI on M365 data | one-time build cost + LLM pass-through |
For a 1,000-employee organisation, full Copilot rollout is €360k/year. Custom AI specific to your workflows usually delivers more ROI for less spend — though it requires the engineering investment up front.
The honest takeaway
Microsoft 365 + AI is a real platform in 2026. The right strategy is layered:
- Copilot for general productivity — for the roles where it pays back.
- Copilot Studio / AI Builder — for simple custom assistants and prototypes.
- Custom AI on M365 data — for the workflows that matter, where generic Copilot won't be deep enough.
Skip any layer based on your specific situation, but don't assume one layer replaces the others. They serve different jobs.
Where to go next
For broader AI development context see The state of AI development in 2026. For when Power Platform stops scaling see PowerApps to production. For our System Integration service page that covers M365 integration in depth.
If you're a Microsoft 365 organisation thinking about where AI fits, drop us a note. We have shipped many M365 + AI builds and can give you a fast take on what's realistic.
Frequently asked questions
Keep reading
From PowerApps to production: when low-code stops scaling
PowerApps is excellent for many Microsoft 365 organisations until it isn't. It hits ceilings around concurrent users, complex business logic, performance, and developer ergonomics. The migration path is rarely 'rewrite everything in Next.js' — it's hybrid: keep PowerApps for what it's good at, move the parts where it's failing to TypeScript on Cloud Run.
The state of AI development in 2026
In 2026, AI development is shipping production agents that earn their keep — document processing, voice, workflow orchestration — backed by Claude / GPT / Gemini and engineered with evals, observability, and guardrails. What's underrated: well-engineered automation with one or two LLM-judgment steps. What's overrated: 'autonomous AGI' marketing.
AI agents vs automation: which one do you actually need?
Use plain automation when the rules are deterministic — same inputs, same outputs, no judgment required. Use AI agents when inputs are unstructured (PDFs, emails, voice) or each instance needs a decision. Most production systems mix both: automation moves the predictable steps, an agent handles the messy ones.
System Integration
Plumb your stack. Microsoft 365, Salesforce, HubSpot, custom databases, ERP — speaking to each other reliably, with audit trails.
Custom Development
Web apps, mobile apps, dashboards, internal tools. React, Next.js, React Native, Power Apps — picked for the job, not the hype.
Workflow Orchestrator Agent
Cross-SaaS triggers — Microsoft 365, Slack, Sheets, HubSpot, Stripe — with idempotency and approvals
Want this delivered in your stack?
If the article describes a workflow you'd like to ship, drop us a note. We reply within one business day.