I built myself an assistant in Microsoft Copilot Studio that brings together data from Dataverse, Azure DevOps with 7pace, Bexio and SharePoint. Every morning it sends me a briefing, it answers plain-language questions about projects, deals and invoices, and it reports delays, changes in the pipeline and overdue payments on its own initiative. It replaces neither my management responsibility nor any finance, HR or legal review, but it shortens the path to the information.
Why I built it
As CEO of CNEXT, I spend a large part of my day on recurring information requests, status updates and coordination. The answers sit in different systems. That gave me the idea: an agent that handles these requests and brings me what matters on its own initiative.
The set-up
The assistant is a custom agent in Copilot Studio. It reaches four systems through the Power Platform, that is, through connectors and dataflows. I use it in Microsoft 365 and Teams.
| System | Role | Contents |
|---|---|---|
| Dataverse | Central data platform | Work items, sales pipeline with opportunities, deals and forecasts, accounts and contacts, billing and resource planning, projects with milestones and progress |
| Azure DevOps with 7pace | Project management and time tracking | Work items, sprints, burndown, repos; with 7pace, time tracking per project and customer |
| Bexio | Source of truth for finance and invoicing | Accounting and financial reporting, invoices, contacts, orders from quote to invoice |
| SharePoint | Incoming post and documents | Digitised business correspondence, contracts with metadata, project documentation |
Contacts are synchronised between Bexio and Dataverse. None of these connections exists out of the box: every connection, query and write action has to be set up and tested deliberately.
What the assistant does for me
Morning briefing
Every morning I automatically receive:
- open deals and how they are developing
- critical project milestones for the week
- outstanding invoices and incoming payments
- important appointments and customer interactions
Plain-language questions
In between, I simply ask, for example:
- ‘What is the status of project XY?’
- ‘Which deals are in the closing phase this week?’
- ‘How many hours were booked on project Z last week?’
- ‘Show me the open invoices over CHF 10,000.’
- ‘Show me my booked hours this week by project.’
Proactive alerts
On its own initiative, the assistant reports:
- project delays and budget overruns
- important changes in the sales pipeline
- overdue payments and critical cash-flow situations
Time booking and billing
Through Azure DevOps and 7pace, it supports time tracking and billing:
- Booking suggestions: it suggests time entries based on the work items someone has worked on.
- Missing entries: it flags when team members still have bookings outstanding.
- Project analysis: it compares booked and planned hours per project.
- Customer billing: it compiles the billable hours for invoicing.
Preparation and communication
Before customer meetings it summarises the relevant customer information, and before team meetings the project progress. For new quotes it pulls out historical data and reference projects, it reminds me of customer replies still outstanding, and I can query incoming post in SharePoint directly in the chat.
If you build an assistant like this
- 1Start small: a daily overview of appointments and approved project status is a sensible first case. Decide which questions the assistant should answer, where the authoritative information lives and when it should refer the user to the person responsible.
- 2Record sources and permissions: for each source, keep track of the owner, refresh schedule and access model. Make sure the identity of the person asking is enforced and confidential content is excluded. Aggregated overviews can also expose sensitive information.
- 3Begin with read access: the assistant should only create tasks, send messages or change business records after a person has confirmed.
- 4Require evidence: when the assistant summarises status or figures, source links and timestamps belong with the answer.
- 5Test with real roles: ask typical, ambiguous and disallowed questions with realistic permissions. Look out for wrong attributions, stale data, access to other teams' information and unclear figures.
- 6Measure before expanding: compare answer quality, supporting sources, wrong answers, handling time and cost against a simple baseline.
Limits
- Not a record: a generated summary is not an accounting or HR record. Statements about revenue, invoices or staff must point to current records the user is authorised to see; for finance, in my case, that means Bexio.
- Data location: a Swiss Microsoft 365 data location is not a blanket promise that all processing, agent components, connectors and logs stay in Switzerland. Microsoft documents storage locations by service and data type. Map every service and data flow involved, which in a set-up like this includes the connections to Azure DevOps, 7pace and Bexio, and compare them with your contractual commitments and requirements.
- Logs: logging and retention must follow your internal policies.
- Time savings: a time saving can only be claimed once it has been measured for the specific workflow. If answers cannot be reliably supported or permissions do not hold, narrow or stop the pilot.
Sources: Microsoft Learn: guidance on generative orchestration in Copilot Studio · Microsoft Learn: using connectors as actions

