For professional services firms with 10 to 50 people

Which of these is actually an agent problem?

The same process questions route to the same person, and the document that answers them is not where people look first. Before building anything, the useful question is whether this needs an agent, an automation, or a better form.

Three patterns look the same from the outside and are not interchangeable. Choosing wrong means rebuilding, so the choice is worth making deliberately and in writing.

Book a readiness assessment

What we see in professional services

One person is the interface to the process

Everyone knows who to ask. That person answers the same questions between doing their actual work, and nothing about it is written down in a form anyone else can use.

A demo that never becomes a rollout

Someone builds something that works on the happy path. Real permissions, missing information and exceptions arrive later, and it quietly becomes a thing one person maintains by hand.

The tool was chosen before the problem was described

A pattern gets picked because it was the one in the demo, not because it matches what the work needs. The mismatch shows up at the point it is expensive to change.

What gets built

Assess

We review the Microsoft 365 tenant, licensing, the documents involved and the work people repeat, then say which pattern fits and which does not. You get a written recommendation, including the case for not building anything yet.

Fixed scope and fixed fee. The report can recommend not building anything yet.

Build

One agent or automation in production against the agreed scope, with acceptance checks covering the exceptions rather than only the happy path. What it should refuse is specified alongside what it should do.

The build scope and delivery window are agreed after the assessment.

Extend and govern

The first agent is a project. The tenth is an estate, so environment strategy, DLP and an inventory of what exists are worth settling before there are ten. Handover documents how it works, its limits and how to maintain it.

Each extension is scoped after the assessment, with its delivery window agreed before work starts.

Apps for professional services

Policy and Handbook Assistant architectureAn employee asks a question. The Copilot Studio agent searches a SharePoint policy library. When a supporting passage is found it answers with a citation. When none is found it refuses, names the owning team, and does not answer from general knowledge.EmployeeTeams or webCopilot Studio agentGenerative answersAuthenticates as the userso SharePoint permissions applySharePoint libraryscoped to the library, not the sitePaid time offExpense reimbursementRemote and hybrid workOnboardingIT and equipmentGrounded answerwith a citation back to the source documentNo grounded source — and the model must not answer from general knowledgeRefuse and redirectStates the answer is not in these documentsand names the team that owns itRefusal paths are logged — questions that hit them show which policies the handbook does not cover.
Answers only from the corpus and refuses rather than guessing

Questions from professional services firms

We are not sure we need an agent at all. Is the assessment still useful?

That is the most useful case for it. A written answer explaining that a simpler process, form or automation would do the job is a decision you can act on, and it costs less than discovering it after a build.

How do we avoid ending up with a demo nobody can run?

By specifying the exceptions before the build, not after. Real permissions, missing information and conflicting sources are acceptance checks, and handover includes the documentation needed to maintain it without us.

Start with a readiness assessment

A review of what is worth building on your tenant and what isn’t. There is no obligation to proceed to a build.