BoteamWhere to apply AI in your business: three places to look, and the order to do them in
Most owners ask “what can AI do?” That is the wrong question. The useful one is where in your business the work already repeats — and almost every opportunity sits in one of three places.
Three places to look, and why the list is short
Every AI opportunity in a small business lands in one of three places. If you are staring at your business wondering where to start, that is the whole map. Anything that does not fit one of those three is usually a tool looking for a problem.
| Where | What it covers | Typical first job |
|---|---|---|
| Internal operations | Admin, quoting, data entry, reporting | Turn a purchase order into a record |
| Customer experience | First enquiry, quote follow-up, support | Reply in minutes, follow up on a schedule |
| New revenue streams | Work you could not afford to offer before | A service that was too slow to sell |
Start inside, not outside
The order matters more than the choice. Start with internal work before you point AI at your customers. Internal processes are lower stakes — a mistake stays inside the business where you can catch it. Customer-facing work is higher stakes because a bad experience is public and hard to undo. Prove it internally, write down what good looks like, then let it touch a customer.
Internal first, customer-facing second. Not because customer work is harder, but because you want your first mistake to be private.
What this looks like in practice
Two real shapes, from small teams. Same idea underneath: the agent does the repeatable 90 per cent, the person approves.
What actually changes: the decision moves, the work does not
The common mistake is thinking AI replaces a person. In almost every small-business case, it replaces a step. The owner’s day does not shrink to nothing — it shifts from producing work to approving it and handling the exceptions. That shift is the whole change, and it is worth seeing side by side.
| Owner-led workflow (before) | AI-native workflow (after) | |
|---|---|---|
| Who does the repeatable work | The owner, in the gaps between jobs | An agent, continuously |
| When it gets done | Whenever the week quietens down | Within minutes, every day |
| What is dropped | Follow-ups, reminders, record-keeping | Almost nothing — it is on a schedule |
| Where the owner’s time goes | Producing the work | Approving the work |
| Where judgement sits | Every step (that is the bottleneck) | Only the decisions that matter |
| What the owner must do more of | Holding everything in their head | Writing down the standard, reviewing the exceptions |
| What breaks when they are away | Everything | The exceptions wait; the routine keeps running |
The point of AI-native operations is not a smaller role for the owner. It is a different one: more defining, less doing. The work that used to be invisible in their head has to be written down first — that is the part people skip.
How to tell if your business is actually getting better at this
A single working role is not the goal — it is evidence you can build the next one. Capability compounds, and it compounds in four stages. You can place your business on this ladder today and see exactly what the next move is.
| Stage | What it looks like | The next move |
|---|---|---|
| 1. Exploring | A few people use a chatbot. Nothing is written down. No role exists. | Write down one workflow, in the order it happens |
| 2. One role live | One defined job runs daily with approval. You can name its number. | Prove it for four to six weeks, then write the standard |
| 3. A system | Two or three roles feed each other. Your team works to written standards. | Hand over the next workflow; stop being the approver of everything |
| 4. AI-native | New work starts by asking “which role owns this?” The owner works on the business. | Reassess capacity each quarter; retire what no longer fits |
If you had to take a week off tomorrow, would the routine work still run? If the answer is no, you have not built capability — you have built a faster version of the bottleneck.
The part most businesses skip: redesign the workflow
Here is where the research gets uncomfortable. McKinsey surveyed 1,719 organisations across 97 countries in 2026 and found 80 per cent of people using AI report better personal productivity, but only 37 per cent of organisations can point to any impact on earnings — and just 6 per cent qualify as high performers, attributing at least 5 per cent of profit to AI. Everyone has access to the same tools. The difference was not the tool. Nearly three-quarters of high performers had fundamentally redesigned the workflow around AI; among everyone else, roughly a quarter had. Adding AI to an unchanged process makes one person faster and the business identical. Changing what the process is — who does what, in what order, with what approval — is what moves the number.
The tool is no longer the differentiator — everyone can buy the same one. What separates the businesses that benefit is the rebuild: the workflow redraw, and the knowledge base the agent runs on. That work is specific to how your business actually operates.
Why most rebuilds need two people in the room
This is the part that stalls quietly. Rebuilding a process is not a software setting you flip — it means deciding what the new workflow should be, then feeding the agent the right knowledge so it can do the work well. Two things have to come together. First, the raw material: your pricing rules, your standards, your exceptions, the judgement calls you have never written down. That only exists in the head of someone who knows the business from the inside. Second, the design: knowing how to structure that knowledge, where to put the approval line, what to measure, and which parts to leave alone. That is specialist work, and it is where most in-house attempts quietly run out of road.
| Business insider | AI specialist | |
|---|---|---|
| Knows | How the work really runs, and the exceptions | How to structure knowledge and where to draw the line |
| Provides | Pricing rules, standards, the judgement calls | The workflow design, the approval points, the measures |
| Alone | A process nobody knows how to build | A technically sound role that does not fit the business |
Not sure what your rebuild should look like?
That is exactly what the free 30-minute AI audit is for. We look at how your week actually runs, find the two or three changes worth making first, and tell you honestly what you should not automate yet. You get a written summary to keep — whether or not we ever work together.
Request your free AI auditNo obligation, no call required to start. Just tell us what you are trying to fix.
Build, buy, or leave it alone
Not every gap needs a build. McKinsey also found 32 per cent of organisations decided against buying at least one piece of software because they could build it themselves with AI coding tools. For a small business the rule is simpler: buy where the process is common (email, accounting, scheduling), and only build where the process is genuinely yours. The test before spending anything is one sentence — “this will work if [specific data] is reliable and [specific person] owns the result.” If you cannot fill in both blanks, you are not ready to buy. Gartner expects organisations to abandon 60 per cent of AI projects that are not supported by AI-ready data; the data problem is almost always the real one, and it costs nothing to check before you spend.
Name the number before you start
The most reliable predictor of whether an AI project delivers is whether anyone wrote down what success means before it began. MIT Sloan found 61 per cent of AI projects were approved on projected value that was never measured afterwards. Pick one number per role — response time, hours of admin returned, quotes sent per week, days-to-payment — and capture what it is today, this week. A small business can see its whole workflow at once, which makes it easier to measure than a large one, not harder.
Your action list
Ignore the sweep of all this and do these seven things, in order. Each one is small enough to finish this week. You do not need to buy anything to start.
- Pick one workflow that repeats every week and causes the most friction. Not five. One.
- Write it down exactly as it happens today, including the exceptions and the judgement calls.
- Capture today’s number: the hours, the delay, or the conversion rate you want to move.
- Mark clearly where a person must decide, and where the step is just typing.
- Set up one role for the typing part, with approval at the batch or checkpoint level — never silent.
- Decide what “good” looks like in writing, so you can check the output instead of re-doing it.
- Review after four to six weeks: did the number move? Then write the standard and pick the next workflow.
Start inside. Move your time from producing work to approving it. Write down the standard before you hand anything over. That is the whole game.