Look first. Build one thing at a time. Keep what pays.
Most AI projects start with a tool. Ours start with a proper look at how your agency works. Then we build the most useful thing first, check it actually helps, and only then move on to the next.
01 / Why AI plans stall
A plan in a drawer changes nothing.
A consultancy runs a review, hands over a 60-slide deck and leaves. Three months later the deck is in a drawer and nothing has changed, because nobody inside the agency owned the work.
The other route is buying AI licences for everyone and hoping they get used. Mostly they don't, because the tool doesn't fit the way the team actually works.
A beautifully designed AI strategy that never gets implemented is worth less than a rough plan that ships in week one and improves every week after.
Alex Lockey, AI Strategy for Professional Services
Standing still has a cost too. Bullhorn's 2026 survey found that UK and Irish agencies using AI were four to eight times more likely to have grown revenue. It comes from a software company and shows a link, not proof, but it is hard to ignore.
02 / Step one: look
Before we build anything, we look at how your agency really works.
It takes three to four weeks, and we do it with you. It is the part most people want to skip, and the part that decides whether everything after it works.
- Talk to your people
- The owner, the managers and a few consultants. Where do the hours go? Where does work get stuck? What do your best people do differently?
- List what you pay for
- Every piece of software and add-on: what it costs, who actually uses it, and what information it holds.
- Look at your data
- With read-only access, we check how complete your records are and where your fees really come from. We change nothing.
- Measure today
- How long things take now, desk by desk, written down before anything changes. Without it, nobody can say later whether something helped.
- Write the plan
- Everything worth building, most valuable first, with the first few jobs described clearly enough for anyone to build them.
The plan is yours to keep, whether we build it or not. The home page shows an example of what we look at, using a made-up agency.
03 / How we know it's working
Every tool has to pay its way.
Before anything is built, we agree what it is for and how we will know it worked. For example, consultants spending two fewer hours a week on admin, or twice as many old clients getting a call. We write down today's number, the number we are aiming for, and when we will check.
Once a tool is running, we keep score: how often people use it, how often it gets things right, what it costs to run, and what it saves or earns. Using it means people actually work differently, not that they logged in once.
The ones that pay back stay and get better. The ones that don't are switched off, and the spend moves to the next workflow on the map.
Alex Lockey, Bench
Savings count too. If a tool we build does the job of software you currently rent, you can cancel that subscription, and the money counts towards the tool. That is how an agency ends up relying on fewer outside suppliers: one proven swap at a time, never all at once. The thinking page explains why this adds up.
04 / Why the order matters
Skip a step and the rest falls over.
AI on top of a messy business just makes the mess faster, and more expensive. So the dull parts come first.
One process at a time. Get the numbers visible before changing anything. Prove a change works on something small before asking anyone to trust it.
Alex Lockey, Singularity from a Company of One
The order is the same every time:
- 01Tidy the data the first job needs, and nothing more
- 02Write down how things work today
- 03Build the most valuable, easiest thing first
- 04Use it for a real week before starting the next
- 05Check the numbers, then keep it, fix it or switch it off
- 06Cancel any software it replaced, and spend the saving on the next build
- 07Update the plan with what you learned
Each new tool reuses what the last one set up, so it costs less. And if something is already working, we leave it alone.
05 / What gets built first
The most value for the least effort goes first.
Everything on the plan is judged on two things: how much it is worth to you, and how hard it is to build and get people using.
Worth a lot and easy to do goes first. It pays for what comes next and shows the team that the change helps them. Worth a lot but hard is planned carefully. Not worth much is left alone, however clever it sounds.
The order isn't fixed. When a tool beats its goal, similar ideas move up the list. When one disappoints, the money goes elsewhere.
06 / You stay in charge
Nothing goes out without a person saying yes.
Every new tool starts by preparing drafts. A consultant checks them and decides what is sent.
Grant trust per task, on evidence. Start with drafts you approve. Move something to automatic only when its track record earns it, individually.
Alex Lockey, Singularity from a Company of One
Only a job with a long record of getting things right gets to run on its own, and only that job. Chasing timesheets might get there in a few weeks. A note to a client might never need to, and that is fine.
Why your agency should own the system all of this runs on is on the thinking page.
Sources and further reading
Where this comes from
Passages on this page are taken from Alex Lockey's published writing. The rest is linked below.