What Does an AI Implementation Consultant Deliver in 90 Days?

The short version: in 90 days, a proper AI implementation consultant should give you one working process that runs without them in the room, documented enough that your team can rebuild or extend it, and a straight number showing what changed. Anything vaguer than that, a strategy deck, a roadmap, a list of "opportunities", is consulting theatre, not implementation. If nobody on your staff can run the thing on day 91, the 90 days didn't work.

Next step on this topic: AI Consultant for Accountancy Firms UK: Costs, What They Do, and Where.

Next step on this topic: Questions to Ask an AI Consultant Before You Hire One.

The shape of 90 days, and why it's never a straight line

Every consultant will show you a tidy three-phase graphic. Discover, build, deploy. In practice it's messier, and if yours claims otherwise I'd ask what they're smoothing over. I've run this exact arc more than a dozen times now, for logistics firms, architecture practices, a training company, a couple of solo consultants who needed their business to survive without them working eighty hours a week. The pattern that holds is this: the first month is slower than clients expect, the middle month is where the real friction shows up, and the final month is where most projects quietly fall apart if nobody's watching for adoption.

Days 1 to 30: diagnosis, not strategy

This is the bit people pay for and then resent, because it produces very little that looks like "AI". What it should produce is a shortlist of two or three processes, ranked by pain and by how measurable the fix is, plus a plain answer to "what data do you have and where does it live". I spend most of week one asking questions that sound almost insultingly basic: who does this task now, how long does it take them, what tool do they use, what happens when it goes wrong. Nobody enjoys being asked, but it's the only way to know if you're solving the right problem.

A real example. I worked with a distribution company in the Midlands, 38 staff, family-run, doing about £4m turnover. Their dispatch team was fielding roughly 60 "where's my order" calls a day. The founder wanted a chatbot. What the diagnosis showed was that most of those calls came from four big retail customers whose orders were split across two couriers with no shared tracking view. The fix wasn't a chatbot at all, it was an automated SMS and email update triggered off their existing dispatch software, with a simple AI-trained FAQ handling the remaining edge cases. If I'd built what they asked for in week one, they'd have got a shiny tool that solved the wrong problem. This is why anyone selling logistics AI work who skips straight to "let's build" without watching how dispatch and customer updates break down is guessing, not consulting.

By day 30 you should have, on paper: the one process being fixed, the current cost of the problem in hours or money, and a defined "done" that isn't "AI is implemented" but something like "70 percent of status update calls handled without a human".

Days 31 to 60: this is where projects earn or lose their fee

The middle stretch is build-and-break. You're setting up the actual workflow, whatever combination of tools that means for your business, testing it against real messy data, not the clean demo data everyone shows in sales calls, and finding out what breaks. In the distribution company's case, week five was when we discovered their courier's API only updated tracking status twice a day, which meant "real-time" updates were a lie we had to quietly stop promising and redesign around. That's normal. Any consultant who tells you the build phase went exactly to plan either did something trivially simple or isn't telling you the whole story.

By day 60 on that project we'd cut the daily call volume from around 60 to 18, roughly a 70 percent drop, freeing up close to 14 hours a week across two admin staff who'd previously done nothing but answer "where's my order". That's the kind of number that should exist by the two-thirds mark. If your consultant can't point to something measured, not projected, measured, by day 60, that's worth asking about directly.

This is also the phase where scope creep quietly kills timelines. The founder wanted to add invoice queries to the same system in week six. We didn't, and I said no plainly, because adding a second process before the first one was stable is how 90-day projects turn into 9-month projects that never quite finish either thing.

Days 61 to 90: handover, not just go-live

Here's the part almost nobody talks about honestly, because it's less exciting than the build. The last month should be spent making yourself replaceable. That means documented steps for the people who'll run this after you leave, at least one training session where your team does the task themselves while you watch and correct rather than doing it for them, and a written answer to "what happens when this breaks and I'm not on retainer". I use a version of the same list every time, close to what's in the AI consultant handover checklist I've written about separately, because without it the client's competence disappears the day the invoice is paid.

For the distribution company, day 90 deliverables looked like this: a working update system live on all four retail accounts, a one-page process doc for the two admin staff, a recorded 20-minute walkthrough video, a simple monthly report template they could fill in themselves, and a fixed monthly cost for the SMS and API usage so there were no surprise bills. Total spend across the 90 days, including my time and their software costs, came to just under £14,000. The saved admin time alone was worth roughly £18,000 a year at their pay rates, before you count fewer angry customer calls, which is the number that got signed off by the founder's business partner, who'd been sceptical from day one.

What you should get on paper

• One process, clearly named, that's fully working without the consultant present
• A before-and-after number: time saved, calls reduced, errors down, something measured twice, not once
• Written documentation your team could hand to a new hire
• At least one recorded training session, not just a live one nobody wrote down
• A named person on your team who now owns the process
• A plain cost for ongoing running, software, API fees, anything recurring, not just the consultant's fee

If any of those six is missing at the end of 90 days, ask why before you pay the final invoice.

Why some 90-day projects quietly stall by month four

Here's the bit that doesn't make the sales pitch. Most 90-day AI projects that fail don't fail during the 90 days at all, they fail in month four, five, or six, after the consultant has left, when the person who was trained on the new process goes on holiday, or leaves the company, and nobody else knows how to run it. I've seen this happen to a client who wasn't mine, an architecture studio that had a beautifully built proposal-automation system, similar in spirit to what I've done for firms handling architects and design studios drowning in proposals and admin, and within six months the one junior who understood the setup had moved on, and the whole thing quietly reverted to manual because nobody had written anything down. The consultant had delivered a working system. What they hadn't delivered was resilience to their own client's normal staff turnover. The uncomfortable truth is that a lot of what gets sold as "implementation" is really just "installation with a training call attached", and the difference only shows up three months after the invoice is paid, by which point the consultant is long gone and unaccountable. Ask your consultant directly what happens if the trained person leaves. If they don't have a clean answer, you're buying installation, not implementation.

What this should cost, roughly

UK-based independent AI implementation consultants generally charge somewhere between £800 and £1,500 a day, or a fixed 90-day package in the region of £8,000 to £25,000 depending on how many processes are in scope and how messy the starting systems are. Agencies tend to sit higher, sometimes considerably. If a number you're being quoted feels wildly out of that range in either direction, that's worth a direct question rather than an assumption, and it's covered in more depth on the page I wrote on what an AI consultant costs. Cheap isn't automatically bad and expensive isn't automatically good, but both extremes deserve a "why" before you sign anything.

It's also worth knowing what you're buying isn't always the same kind of person. Some businesses need a project-based consultant for exactly this 90-day arc. Others need someone embedded part-time for six to twelve months, which is a different role entirely, closer to what's described in the piece on the difference between a fractional AI officer and a consultant. Knowing which one you need before the first call saves everyone a lot of wasted time.

Before you hire anyone for this

Get specific on what "delivered" means before day one, not after day ninety. Ask for the metric that will be measured, ask who on your team will be trained to own it, and ask what the handover document will contain. If you're still working out how to hire well in the first place, the practical breakdown in how to hire an AI consultant for a small business covers the questions worth asking before you commit budget. And if you've got a discovery call booked already, it's worth reading through what to prepare for an AI consultant discovery call first, because the calls that go well are the ones where the business owner already knows roughly what problem they want solved, not just that they "should be doing something with AI".

For businesses that want proper hands-on support through exactly this kind of 90-day build, rather than a slide deck and a wave goodbye, that's the whole point of working with an AI implementation coach rather than a pure strategy consultant. The word "implementation" should mean something built, tested, handed over, and still standing after you stop paying for it.

If this is the kind of work you want done rather than read about, start with how to hire an AI consultant, then see what I do as an AI automation consultant.

Related: my ai consultant accountancy practices what to automate first page.

Frequently asked questions

What's a realistic first result to expect within 90 days?

One working process with a measured before-and-after number, not a full transformation. A single automated workflow that reliably saves hours or reduces errors is a realistic and honest 90-day result; anyone promising a company-wide AI overhaul in that timeframe is overselling.

How much should a 90-day AI implementation project cost?

For UK small businesses, expect somewhere between £8,000 and £25,000 for a fixed 90-day scope, or £800 to £1,500 a day if billed by time, depending on complexity and how many systems need connecting.

What happens after the 90 days end?

A good consultant leaves you with documentation, a trained internal owner, and a clear answer for what happens if that person leaves. Without those three things, the project usually reverts to manual within six months once staff or priorities shift.

Should I hire a consultant or a fractional AI officer instead?

If you have one or two clear processes to fix, a project-based consultant working a 90-day arc is usually cheaper and faster. If you need ongoing decisions made across the whole business, a fractional role embedded part-time over months is a better fit, and the two roles shouldn't be priced or judged the same way.

Official documentation

• Anthropic's own documentation
• the NIST AI Risk Management Framework