AI for business consultants and agencies: reducing work between meetings

Where capacity runs out

Consultants sell professional judgement but perform considerable supporting work between meetings: writing recaps, checking promised material, sending reminders, preparing reports and briefing colleagues.

Delegating some of this creates another task: transferring context. If the expert has to dictate the entire history again, little time is saved. Agencies encounter the same problem when replacing an account manager. The project is formally handed over, yet the client keeps reminding the new person about earlier agreements.

A handover needs conclusions and their reasons: why a hypothesis was abandoned, what the team decided to test instead and which data is missing. If AI preserves these connections with source links, a new participant can investigate without interrupting the expert at every step.

What belongs in a client brief?

A brief should prepare someone for the next conversation. Include the objective, current stage, approved decisions, hypotheses being tested, open tasks, expected material and questions for the next meeting.

Each task needs a responsible party: the consultant's team, the client or both. Otherwise, missing client material can look like the agency has failed to deliver. A promise and a confirmed deadline also need to remain distinct.

Do not copy every conversation into the brief. Keep it useful for the next decision, with details accessible through links. Readers should be able to check the reasoning when the summary is insufficient.

An example of preparation for a client call

Consider a hypothetical sales-funnel project. At the last meeting, the parties agreed to investigate two segments. The client would supply an export, after which the consultant would prepare an interview plan.

The next briefing could read:

• Current objective: choose a segment for the first experiment.
• Previous decision: compare segments A and B; no final choice has been made.
• Received: data for segment A.
• Outstanding: data for segment B, with no confirmed delivery date.
• Dependency: comparative analysis cannot yet proceed.
• Meeting question: wait for the second export or approve a limited experiment with A?

The conversation can begin with the next decision rather than working out which export arrived. If nothing is known about the second, AI should not invent a reason for its absence.

The same record helps hand preparation to a colleague. They can see which analysis is blocked and what to ask the client. Without that context, they may mistake partial data for a complete dataset and start work that has to be redone.

What should remain with the expert?

The expert judges a hypothesis, interprets the client's situation and takes responsibility for recommendations. AI can prepare material, draft reports and find outstanding agreements.

Sending a report, changing a promised scope or making a commitment on the consultant's behalf must be separated from drafting. Delegating an action does not remove contractual responsibility.

Anthropic's discussion of effective agent systems distinguishes predefined workflows from agents that choose their next actions. A recurring client report often benefits more from a predictable sequence with review than from broad autonomy without clear limits.

Running a pilot without reorganising the practice

Start with one type of engagement and a few clients whose material you are authorised to use. Executive AI addresses this scenario among others: retaining client history, preparing for the next meeting and following up on agreements between calls.

Define the work to test: process a meeting, draft tasks, prepare the next briefing and produce a report. Record the time each step currently takes and the errors that require correction.

Start with anonymised material from completed projects, then move to draft-only assistance in current work. Discuss external actions separately once limitations and approval rules are understood.

Does the time saved justify the subscription?

The following numbers are hypothetical, not Executive AI results or prices.

Suppose a consultant holds 24 client meetings a month. Preparation and follow-up take 30 minutes per meeting. With AI, including review, they take 15 minutes. The difference is 24 × 15 = 360 minutes, or six hours per month.

At a hypothetical internal hourly value of RUB 3,000, the freed-up time is valued at RUB 18,000. If subscription and support cost an assumed RUB 10,000 monthly, the difference is RUB 8,000 before setup and other expenses. This is a valuation of time, not cash in the bank. For a fixed-fee consultant, it becomes income only through additional paid work or a reduction in actual spending.

Include initial setup costs when estimating payback. If the recurring benefit after ongoing expenses is zero or negative, scaling does not fix the calculation by itself. Do not count extra sales until their connection to the time released is established.

Checking quality and separation between clients

Measure more than time: significant report corrections, missed commitments and mixed client context. A fast but misdirected report can outweigh several weeks of gains.

Give two fictional clients similar projects with different terms. Request separate briefings, then change an agreement in one and confirm that the other is unaffected. Check that an assistant sees only assigned clients and that revoked access cannot be bypassed through old summaries.

In January 2026, Anthropic published its guide to evaluating agents, addressing multi-step behaviour. For consulting work, evaluate the complete cycle, including corrections and the consequences of actions.