What should the first engagement cost and cover?

Scope the first piece of work so it is useful on its own: a diagnostic of a defined process, with a written finding and a recommendation you could hand to another firm. If the first step only makes sense as the entry to a larger program, it is a sales step, not a deliverable.

Should I hire a specialist or a large firm?

It depends on the failure you are trying to avoid. Large firms bring bench depth and procurement comfort. Specialists bring the operator who has done the work. For finance processes with audit exposure, the specific experience usually matters more than the size of the bench.

How do I evaluate a partner's AI claims?

Ask them to walk through one build end to end: what it does, who reviews it, what it got wrong in the first month, and what they changed. A partner who cannot describe a failure has not run anything long enough to learn from it.

How long before we see results?

Set a measure in the first week and check it in ninety days. Cycle time on a named process, error rate, or hours returned to a named person. If no one can state the measure at the start, no one will be able to claim the result at the end.

Putting AI to Work · Buyer’s guide

How to choose an AI strategy consulting partner.

Alicia Dahling, MBA, MACC · Dahling Consulting

Most AI strategy proposals are easy to write and hard to use. They describe a future state, rank your maturity, and leave the actual work untouched. The way to avoid that is to judge a partner on how they scope the first ninety days, not on how they describe year three. This page is the checklist I would use as the buyer.

Six things to look for

1
They start with your process, not their platform. A useful partner asks what work happens today, who does it, how long it takes, and where it breaks. A partner who leads with a named tool has already decided the answer before hearing the question.
2
They can name the work AI should not touch. Judgment calls, anything with legal exposure, anything where a wrong answer is expensive and hard to detect. A partner who cannot draw that line will hand you automation you have to babysit.
3
They own the controls, not only the build. Ask who reviews the output, on what cadence, and what the evidence trail looks like when someone asks how a number was produced. Automation without a control design is control debt with better marketing.
4
They are specific about data. Where your data goes, whether it trains a model, who at the vendor can read it, what the retention window is. These have concrete answers. Vague reassurance is a finding, not a comfort.
5
They have done the operating role, not only the advising. Someone who has closed books, defended a revenue policy, or sat through an audit knows what breaks in month three. That experience shows up in the questions they ask in the first hour.
6
They will scope something small first. A diagnostic or a single workflow, with a defined end date and a written deliverable. A partner who will only start with a transformation program is managing their revenue, not your risk.

Seven questions for the first call

Ask these in order. The answers separate operators from presenters within twenty minutes.

Which of my processes would you leave alone, and why?
What does the first deliverable look like, and can I use it if we stop after it?
Who does the work day to day, and what else are they staffed on?
How will we know in ninety days whether this worked? Name the measure.
Where does my data sit, and who can read it?
What happens to this after you leave, and who maintains it?
Show me something you built, not a deck about something you built.

Signals to walk away from

The proposal is a maturity model. Five stages, a spider chart, and your organization somewhere in the middle. It costs nothing to produce and tells you nothing you can act on Monday.
Savings are quoted before discovery. A percentage figure named in a first meeting is a sales anchor. Real numbers come after someone has looked at how the work is actually done.
The senior person disappears after the pitch. Ask, in writing, who is delivering. If the answer changes between the pitch and the kickoff, that is the pattern for the rest of the engagement.
No one mentions failure modes. Every automation has a way of being confidently wrong. A partner who has not raised one yet has not thought about yours.

Common questions

What should the first engagement cost and cover?: Scope the first piece of work so it is useful on its own: a diagnostic of a defined process, with a written finding and a recommendation you could hand to another firm. If the first step only makes sense as the entry to a larger program, it is a sales step, not a deliverable.
Should I hire a specialist or a large firm?: It depends on the failure you are trying to avoid. Large firms bring bench depth and procurement comfort. Specialists bring the operator who has done the work. For finance processes with audit exposure, the specific experience usually matters more than the size of the bench.
How do I evaluate a partner's AI claims?: Ask them to walk through one build end to end: what it does, who reviews it, what it got wrong in the first month, and what they changed. A partner who cannot describe a failure has not run anything long enough to learn from it.
How long before we see results?: Set a measure in the first week and check it in ninety days. Cycle time on a named process, error rate, or hours returned to a named person. If no one can state the measure at the start, no one will be able to claim the result at the end.

Start small, in writing

Bring one process. We will scope the first ninety days.

More on where AI belongs and where to keep it out is on the AI page.

Discuss a transformation