Most businesses start an AI engagement without a clear picture of what the process actually looks like. Here's a realistic, step-by-step view of what to expect, from first conversation to first deliverable.
The first conversation isn't a sales pitch
A good AI consultant's first call should feel more like a diagnostic than a presentation. They should be asking about your business, what you do, where the friction is, what you've already tried, not talking at you about their methodology or their past clients.
If you find yourself mostly listening in the first call, that's a flag. You should leave it feeling like someone understood your situation. The output of a good first conversation is a plain-language sense of whether AI is the right tool for the problem you described, and if so, what the right first step looks like.
What you'll need to provide early on
Clients sometimes assume the consultant will figure everything out on their own. In practice, the engagement goes faster when you come in with a few things ready, not polished, just available:
- A specific problem or workflow to start with. “Make us more efficient” is hard to scope. “Reduce the time our ops team spends manually matching invoices to purchase orders” is workable. The more specific, the faster the first phase moves.
- Access to the relevant data. Not everything, just the data that touches the problem you want to solve. This doesn't need to be clean or well-organized; assessing it is part of the work. But someone on your team needs to be able to pull it.
- A point of contact who knows the workflow. The people who actually do the process day-to-day are the source of truth. Access to one or two of them, even informally, is more valuable than any amount of documentation.
- Leadership clarity on what success looks like. “It works” isn't a success metric. “It cuts the time on this task from 4 hours to under 1” is. Getting alignment on the metric before the work starts saves a lot of friction later.
What a Readiness Assessment looks like in practice
If you start with a Readiness Assessment, which is the right first move for most businesses, the engagement typically runs two to four weeks. The consultant spends the first week doing interviews: key stakeholders, department heads, the people who manage the data, and the people who do the work being considered for AI.
The second and third weeks are analysis, mapping what AI can and can't practically do against what the business actually has (data quality, system access, team capacity). The deliverable is a written report: a clear picture of where you stand, a ranked list of AI opportunities by impact and effort, and a recommended first step.
You own the report outright. Whether you continue with the same firm or take it elsewhere or do nothing with it, it's yours to keep.
What a Pilot Project looks like
A Pilot takes one specific use case from the Assessment or Strategy phase and builds a working version of it on your real data. This typically runs six to twelve weeks and ends with three things: a working tool your team can actually use, a results report with measured numbers (time saved, accuracy rate, whatever the agreed metric was), and a clear recommendation on whether to expand, adjust, or stop.
The Pilot is designed to produce a decision, not just output. If the results aren't there, a good consultant will tell you plainly and recommend you stop rather than push toward a larger engagement that won't deliver.
Common surprises (and how to handle them)
The data is messier than expected. This is almost universal. Good engagements plan for a data-cleaning phase rather than assuming the data is ready. If a consultant's proposal assumes clean, structured data and yours isn't, push them on the assumption.
The most valuable use case isn't the one you started with. Assessments regularly surface better opportunities than the one that prompted the engagement in the first place. Stay open to it, the goal is the best return, not validating the original idea.
The team is more skeptical than leadership. The people whose workflow is being changed have legitimate concerns. A good engagement addresses them directly, involving the team early, explaining how their role changes (not disappears), and designing the tool so it assists rather than replaces human judgment.
The timeline slips when access slips. The most common reason engagements run long is waiting for access, to data, to systems, to the right person for an interview. Having a named internal point of contact who can unblock access questions usually cuts weeks off the schedule.
What “done” actually looks like
A well-run engagement ends with four things: a working tool or a written strategy (depending on the package), measured results that let you judge the value, documentation your team can actually use, and a clear picture of what to do next.
You should leave feeling more capable than when you started, not more dependent on the consultant. If the engagement increased reliance on external help without building internal knowledge, something went wrong. The right question to ask as you near the end is: can our team run this without you? If the answer is no, that's the conversation to have before you sign off.