How to prepare for an AI consultant

Almost every overrun in this kind of work is caused by the client, not the supplier, and not through any fault of character. Questions go unanswered for a week, credentials belong to somebody on vacation, and a four-week engagement becomes seven.

All of it is preventable in an afternoon. Here is the list.

Name one decision-maker who will actually be reachable

The single highest-value item. One person with authority to answer questions and approve direction, available within a day throughout.

Not a committee. Committees are how a two-day question becomes a two-week question, and the cost of that lands on your invoice one way or another.

If nobody senior can commit that attention right now, delay the start. An engagement run around an absent decision-maker produces a document nobody can act on, which is the most expensive possible outcome.

Sort out access before day one

Make a list of every system that matters and check, specifically, that somebody available can grant access to each one. Your CRM, your finance tool, your email platform, your phone system, whatever your team works in.

Then check for the usual traps: an account held by one person who is away, a platform where only an external administrator can add users, a security review nobody has scheduled, a tool where the original owner has left the company.

This is free to fix in advance and expensive to fix mid-project, and it is the second most common cause of slippage.

Write down what you already know

Any process documentation, volume figures, or notes on where things stick. Bring it to the first meeting.

Two reasons. It compresses discovery, which should reduce your scope and your fee. And a supplier who does not adjust in response to genuinely useful material has a fixed template rather than a real method, which is worth learning early. Where overpaying happens.

Do not build anything new for this. If it does not already exist, that absence is itself a finding.

Capture a baseline

Pick the two or three numbers the work is supposed to move and record them now, before anything changes.

Whatever is relevant: how many inquiries you get and how many convert, how long a quote takes to go out, how many hours a week go into a particular task, how often something has to be redone.

Without a baseline, success becomes a matter of opinion in six months, and opinion favors whoever is presenting. This takes an afternoon and it is the difference between evidence and a feeling. Measurement design.

Tell your team, honestly, before the consultant arrives

The failure nobody plans for is quiet resistance, and it is completely rational when people think a project might eliminate their job.

Say what is happening and why. If the honest answer is that roles will change, say that instead of pretending otherwise, because the pretense collapses in the first interview and then nobody tells the consultant anything true.

What you need from the people who do the work is candor about the annoying parts of their day. You will not get it from anybody who suspects the exercise is aimed at them. This one determines the quality of the entire diagnosis.

One group to bring in early rather than late: whoever looks after your systems. They are the most able to stall this and the most likely to be excluded from the decision, usually for reasons that turn out to be correct. Why IT resists and what to do about it.

Decide what is off the table

Write down the constraints before you are asked. Data that cannot leave your environment. Systems nobody is allowed to touch. Processes that must stay human because the contact is the relationship. A regulatory boundary. A tool you are contractually stuck with for two more years.

Stating these upfront costs you nothing and prevents a set of recommendations you cannot act on.

What you do not need to do

Worth saying, because clients over-prepare in the wrong direction.

You do not need to clean your data first. How messy it is happens to be one of the most useful findings, and tidying it in advance can hide the problem while leaving it in place.

You do not need to have an AI strategy. That is the thing you are buying.

You do not need to pick tools. If you arrive having already chosen a platform, you have narrowed the diagnosis before it started.

You do not need to know your own answer. If you do already know, buy a build instead of a diagnosis.

The realistic time commitment

For a diagnostic engagement: a couple of hours for the decision-maker, an hour each with a handful of people who do the work, and someone reachable for questions. A few hours total across a few weeks.

Any supplier telling you it requires almost nothing from you is either not doing much or setting up an argument about delays later. The honest version is that your availability is the main variable in the schedule. Where the time goes.

Our own first session is built around this: how the engagement runs. Turning up with the baseline and the access sorted will shorten it.