Straight answer: An AI implementation project spends roughly 70% of its time sorting out your data, your processes and your people, and only about 30% of it on the AI itself. It takes longer than the sales page said, it will expose problems in your business that have nothing to do with AI, and the tool working is the easy part compared to getting your team to use it.
I've run these projects with accountancy firms, recruitment agencies, a manufacturer in the Midlands and more solo founders than I can count. Every single one of them expected something different to what they got. So here's what happens, in order, with real numbers attached.
The first two weeks look nothing like AI
If you've paid someone to help you implement AI and the first two weeks involve spreadsheets, a shared document titled "current process" and a lot of questions about how your invoicing works, that's not a sign the project is going badly. That's the project working correctly.
Before any tool gets touched, a competent implementation looks at three things: where your data lives, how messy it is, and how your team currently does the task you want to automate. This is the part most people skip when they try to do it themselves with a free trial of ChatGPT, and it's the part that decides whether the whole thing succeeds or fails six months later.
I usually see clients want to jump straight to "which tool do we buy." Wrong order. Tool choice is maybe the fourth decision you make, not the first.
A real one: the bookkeeping practice that nearly gave up in week three
A bookkeeping practice in Leeds, fourteen staff, came to me wanting to automate client onboarding and first-pass categorisation of transactions. Reasonable goal. Bank feeds, receipts, the usual mess.
Week one, we mapped the current process. Week two, we found that four different staff members were categorising the same type of transaction four different ways, none of them wrong exactly, just inconsistent. There was no written standard. It existed only in people's heads.
By week three, the owner was frustrated. She'd been told this was an "AI project" and instead her team was in workshops arguing about categorisation rules. She nearly pulled the plug and went back to doing things the old way.
We pushed through it. Once the rules were written down and agreed, the AI layer took about nine days to build and test. The categorisation accuracy went from roughly 60% consistent (their own estimate, from a manual audit we ran) to 94% within the first month of live use. Staff time on that specific task dropped from around eleven hours a week across the team to under three.
The AI didn't create that result. The clarity did. The AI just executed it faster than a human could.
Week by week: what a realistic timeline looks like
For a small business, here's the shape of a project that works, not the shape that gets promised in a sales call:
- Weeks 1 to 2: discovery, process mapping, data audit. No AI touched yet.
- Weeks 3 to 4: tool selection and a small pilot, usually one task, one team, not the whole business.
- Weeks 5 to 8: testing, fixing, retesting. This is where accuracy gets checked against real work, not demo data.
- Weeks 9 to 12: rollout to the wider team, training, and the first real measurement of whether it stuck.
That's roughly a 90-day arc, which is exactly why I built my 90-day roadmap around it rather than pretending it happens in three weeks. If someone tells you they can implement AI across a team in ten days flat, ask them what they're skipping. Usually it's the bit that stops it falling apart in month four.
If you want the fuller breakdown of how long each stage takes depending on the size of your team, I've laid that out in a separate realistic timeline piece, because "how long will this take" is the single most common question I get asked before anyone signs anything.
The bit that catches people out: adoption, not accuracy
Here's the uncomfortable part that doesn't show up in most write-ups of this topic. The tool almost always works. Vendors have got good at this. What kills these projects isn't the AI getting things wrong, it's your own staff quietly not using it.
I've watched adoption sit at 90% in the first week of a rollout, everyone excited, everyone testing it, and then drop to under 40% by week six once the novelty wears off and the old habit reasserts itself. Not because the tool broke. Because nobody made using it easier than not using it, and nobody was accountable for checking.
This is a people problem dressed up as a technology problem, and it's the number one reason implementation budgets get written off as wasted money. The business owner blames the software. The real cause is usually that no single person owned adoption, there was no follow-up training in week four, and the tool got left to compete against twenty years of muscle memory with no support.
If you're going into a project, ask upfront: who owns adoption after week eight? If the answer is "everyone," that means nobody, and you should expect the same drop-off I've seen dozens of times.
What it costs, in real terms
For a small business, a run implementation project, from discovery through to a working, adopted process, typically runs somewhere between £3,000 and £15,000 depending on scope, whether you need custom integration work, and how many processes you're touching. A single-task pilot at the low end. A multi-department rollout with custom workflows at the top end.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Cheaper quotes usually mean less discovery time and less training support, which is exactly where the adoption problem I just described tends to come from. More expensive doesn't automatically mean better either. I've seen consultants charge five figures for what was, once you stripped away the jargon, a chatbot with a logo on it.
If you're weighing up whether to bring someone in versus doing it in-house, it's worth understanding what a coach or consultant does differently to a tool vendor. A vendor sells you software. A decent coach or consultant sits with your actual process first and only recommends software second, which is the order that produces the Leeds result rather than an expensive shelf-ware result.
What good looks like at day 90
By the end of a project that's gone well, you should be able to point to three specific things:
- A measurable time or cost saving on the specific task you targeted, not a vague "it's helped."
- At least 70% of the intended team using the tool without being chased, six weeks after launch.
- One person internally who understands the process well enough to train a new hire on it without calling the consultant back.
If none of those three things are true after ninety days, the project hasn't finished, whatever the invoice says.
Signs a project is about to go sideways
A few things I watch for now, because I've seen them precede a failed rollout every time:
- The kickoff conversation is all about the tool and none about the current process. Backwards order, guaranteed rework later.
- Nobody from the actual team doing the task is in the room during scoping, only management.
- There's no plan for what happens in week six, when the initial enthusiasm has worn off.
- The data hasn't been looked at yet and everyone's confident it's "pretty clean." It never is.
If your project has more than one of these, pause and fix it before spending another pound on tooling.
Where to start if you haven't begun yet
If you're a founder without a technical background wondering where to even start, the honest starting point isn't picking software, it's getting a clear, unbiased read on which single process to fix first. That's all a good strategy session should give you: one prioritised target, not a forty-tool wishlist.
And if the idea of doing this without a developer on staff is what's putting you off, that's a solvable problem, not a reason to wait another year. There's a clear path for non-technical founders to get this done without learning to code, and most of my clients arrive exactly there, capable business owners who just haven't had someone map the process out for them yet.
Frequently asked questions
How long does an AI implementation project usually take?
For a small business tackling one process, expect roughly 8 to 12 weeks from first discovery call to a working, adopted result. Bigger rollouts across multiple teams or departments can run 4 to 6 months, mostly because training and adoption take longer than the technical build.
What's the biggest reason AI implementation projects fail?
Not the technology. It's almost always adoption. The tool works fine in testing, then usage quietly drops off within six weeks because nobody owned the follow-up training or held the team accountable for using it once the novelty wore off.
How much should a small business budget for an AI implementation project?
Realistically between £3,000 and £15,000 for a scoped project, depending on how many processes you're touching and whether custom integration work is needed. Cheaper quotes often mean less discovery and training time, which is exactly where projects fall down later.
Do I need to fix my data before starting an AI project?
Yes, almost always. Most projects spend their first two to three weeks on process mapping and data cleanup before any AI tool is even chosen. Skipping this step is the single most common reason results disappoint six months in.
Want this done for you? See how a fractional AI officer can run this inside your business.
Related reading: AI Phone Receptionists for Small Business: What They Cost and Where They Quietly Lose You Money and AI Phone Agents for Small Business: What Happens When You Let AI Answer Your Calls.
