The hardest part of adopting AI inside an SME is not the technology. It is picking the right first project. Pick wrong and you waste twelve weeks and a budget for nothing. Pick right and a single build pays back inside a year, while quietly proving the model so the next build is easier to commission.
This playbook is the conversation we have on every scoping call. It is the framework we use to help founders go from "we know we need to do something with AI" to "we know exactly what to build first and what it should cost."
Start from the work, not the technology
Most failed AI projects start with "we want to use AI for X." The right question is the other way round: where in your business does manual work pile up week after week, the rules are clear enough to write down, and the cost of getting it wrong is recoverable?
That intersection is your first build. Anything else is a science experiment dressed up as a project.
Three traits to look for:
- Repeatable. The same kind of task happens many times a week. Quoting jobs, parsing supplier invoices, qualifying inbound leads, sending out booking confirmations, triaging support tickets, building monthly reports.
- Rules-based or pattern-based. A reasonably bright new hire could be trained to do the task with a checklist and a few hours of shadowing. If only your most senior person can do it, it is the wrong first build.
- Recoverable. If the system gets it wrong on a Tuesday, you can catch it on a Wednesday. You are not betting the business on the first build.
If a task scores high on all three, it is a candidate. List them and rank them.
Find the pain, not the prestige
Hours
× fully-loaded salary cost
Person-hours per week consumed by the task. The first currency. Multiply by fully-loaded salary and you have the floor of the payback.
Mistakes
× cost per error
How often someone fat-fingers a number, misses a deadline, or skips a step. Each mistake has a number behind it.
Lost opportunity
leads not chased, calls not returned
What gets dropped because the team is buried in this task. The hardest currency to count and the one that compounds the fastest.
When you rank candidates, do not pick the one that sounds most impressive in a board pack. Pick the one that hurts most.
Pain shows up in three currencies:
- Hours. How many person-hours per week does this task consume? Multiply by fully loaded salary cost.
- Mistakes. How often does someone fat-finger a number, miss a deadline, or skip a step? What does each mistake cost?
- Lost opportunity. What gets dropped because your team is buried in this task? Calls not returned, leads not chased, prospects ghosted because someone was reconciling a spreadsheet for two days.
The right first build sits on a task where at least two of those three currencies are visibly bleeding.
Write down what "good" looks like before any code is written
This is the step almost everyone skips. They jump from "we should automate X" straight to "let's get a quote."
Don't. Spend an hour writing down the answers to:
- What inputs go into this task today? (Emails, PDFs, spreadsheets, phone calls, CRM records.)
- What outputs come out? (Approved invoices, qualified leads, sent emails, updated rows.)
- What rules govern the decisions? List them in plain English.
- What are the edge cases your team handles instinctively? Write those down too. Most of the project risk lives here.
- What does "the system did the right thing" look like? How will you measure it?
If you cannot answer these in one sitting, the task is not as well-defined as you think, and that is a finding in itself. Either define it now, or pick a different task.
A good scoping document is one page. If yours is ten, you are over-thinking it. If yours is half a paragraph, you are under-thinking it.
Time-box the first build
The right first build for an SME is two to five weeks of work, not three months. There are three reasons:
- Learning compounds. You learn far more from one shipped tool that works than from a half-built bigger tool that doesn't. Ship small, learn, ship the next.
- Reversibility. A two-week build is easy to walk away from if the answer turns out to be "this should have been a spreadsheet macro." A three-month build is psychologically and financially harder to kill.
- Cash. Most SMEs are not in a position to commit £30k+ to a speculative first project. £1,500 to £4,500 to remove one painful task is a much easier yes.
If your shortlist of candidate tasks doesn't have anything that fits in five weeks, you have not broken the problem down small enough. There is almost always a single sharp slice of the broader problem that fits the time box.
Pick the tool, not the platform
A common scoping mistake is to start by choosing a platform: "we'll build on Salesforce", "we'll use n8n", "we should use OpenAI." This is backwards.
Pick the outcome first. Pick the task second. Let the platform fall out of the requirements. A good builder will choose between off-the-shelf SaaS, a low-code workflow tool, a custom-built app, or a hybrid based on what actually fits, not on what they happen to specialise in.
If your shortlisted builder leads with platform names before they have understood your task, that is a signal to keep looking.
What to bring to a scoping call
When you are ready to talk to someone (us, or anyone else), bring:
- A one-paragraph description of the task and why it hurts.
- A rough sense of the volume (how many times per week / month).
- One or two examples of the inputs (a redacted invoice, a sample lead, a typical email).
- The names of the systems involved (CRM, email, accounting tool, ATS, etc.).
- A range for the budget you are comfortable committing to a first build.
That is enough for a good builder to give you a written, fixed price within forty-eight hours, and to tell you honestly whether what you are asking for is a Build, a SaaS subscription, or a process change.
The honest answer
Most SMEs do not need their first AI project to be flashy. They need it to remove one repetitive task forever, without breaking when an edge case shows up. The framework above is how we help clients pick that task. It works whether you commission us, hire someone in-house, or do it yourself.
If you want a hand applying it to your own business, the scoping call is free, the written price is yours either way, and there is no obligation.