AI for consultants: hand over prep, keep judgement

AI for consultants is most useful when it takes ownership of the work that has to be done before judgement can begin. Desk research, source collection, first drafts, meeting preparation and messy data can all be prepared by an agent. The consultant still decides what matters, what is missing, what can be defended and what should be said to the client.

That split matters because consulting is not a document production business. A client may receive a memo, a model or a set of recommendations, but the value sits in the choices behind them. The consultant has to decide which evidence deserves weight, which conclusion is too weak to carry into the room and which uncomfortable point should be raised rather than edited away.

Give the agent the work before the judgement

A useful delegation starts with a piece of work that has a clear input and a checkable output. Suppose a consultant is preparing for a market review. An agent can gather public material named in the brief, extract dates and figures, sort notes by company or topic, flag gaps and produce a source pack. It can also turn that pack into a first draft of the factual sections.

The same pattern works for internal material. Give an agent a folder of interview notes and ask it to group recurring issues, pull out statements that support each theme and create a clean working table. Give it a spreadsheet full of inconsistent labels and ask it to standardise names, identify blank fields and separate rows that need a person to inspect.

This is where AI can save a consultant from spending the first hour of a task getting the material into a usable shape. The output is not the answer. It is a better starting point for the person who has to form one.

Research can be delegated, relevance cannot

Desk research looks like an obvious AI task, but the brief has to be narrower than 'research this market'. A consultant usually knows what would change the recommendation. That knowledge should shape the delegation.

A stronger brief might ask for every announced factory opening by named competitors since a given year, with location, planned capacity, source date and any stated opening date. Another might ask for product pages that mention a specific feature, with the exact wording copied into a source table. The agent can do the collecting and sorting. The consultant decides whether the evidence is comparable, current enough and relevant to the client's question.

This also gives the review a clear shape. Instead of checking a long narrative for hidden assumptions, the consultant can inspect the source set, question odd entries and ask for another pass where the evidence is thin. Research becomes easier to audit because the preparation has been separated from the interpretation.

First drafts are useful when nobody mistakes them for conclusions

Consulting work contains plenty of writing that does not need to begin with an empty page. An agent can prepare a meeting brief, a first version of an interview guide, a factual company profile, a project update or the opening structure of a client memo.

The consultant should be suspicious of the places where a draft starts sounding certain. A sentence such as 'The company should enter the segment through a partnership' is not ordinary prose cleanup. It contains a recommendation, and recommendations depend on assumptions about cost, timing, control, risk and the client's tolerance for each.

A better handoff is to ask the agent to lay out the evidence for each option, list unresolved questions and draft the descriptive parts. The consultant then writes or rewrites the recommendation. That preserves the part of the work the client is actually paying to receive.

Data tidying is agent work until the meaning changes

The boundary is semantic. Cleaning a date column is preparation. Deciding that two revenue lines should be combined because they represent the same economic activity is analysis. An agent can propose the merge, but a consultant who understands the source material should decide whether it is valid.

This distinction becomes especially useful on recurring work. Once a team has decided how a field should be treated, that rule can be put into the next brief. What required judgement once may become routine preparation later. The key is that a person made the rule before the agent repeated it.

The client pays for calls that cannot be delegated cleanly

There are parts of consulting where delegation stops being attractive because the task is the judgement itself. Framing the client's problem is one. Choosing which tradeoff to accept is another. So is deciding how hard to push a conclusion when the evidence is incomplete.

A consultant also carries context that may never appear neatly in the source material. A recommendation that is financially sound may be unusable because the organisation cannot implement it this quarter. A technically correct answer may fail because it ignores who has authority, what has already been tried or what the client is unwilling to change. Those are not formatting issues.

The person should also own anything that creates a commitment outside the working process. Sending a client note, publishing a finding, changing a shared record, accepting terms or spending money deserves a deliberate sign-off. Preparation can move quickly. Commitment should stay visible.

Build the handoff around review points

The cleanest way to use AI in a consulting practice is to divide work by review point rather than by job title. Ask what an agent can prepare, what evidence a person needs to inspect and what action must wait for approval.

In Delagents, a person writes the brief, an agent carries out the work and stops at checkpoints before actions that matter. Results arrive as deliverables rather than disappearing into a chat thread. For consulting teams that want this pattern across client work, Delagents for agencies is the relevant starting point.

Pick one live engagement this week and mark up the workplan. Put a P beside every task an agent could prepare without changing the client's position. Put a J beside every task that depends on professional judgement. Put an A beside every external action that should wait for approval. Delegate one P task with a narrow brief, then review the result before widening the scope.