How to build an AI strategy from what you already have

Most advice on AI strategy assumes a blank page. You are not starting from a blank page, and the material you need is already in the building.

Think about the last significant system your business bought. A CRM, a warehouse system, a ticketing platform. Somebody wrote a paper. It had a number in it, a named owner and a go-live date, and there was a budget line, a rollout plan and a training day.

Now think about how AI got in. Somebody in marketing started using a writing tool. Someone in service tried a chatbot on the simpler tickets. Finance got a copilot licence on a renewal that never needed separate approval. Each one was cheap enough to stay below the level where anyone writes a paper, and each one was a sensible thing for that team to do.

So you have arrived somewhere real, and you arrived without a plan. Which means building the strategy is a different job from the one the guides describe.

Why is there no strategy behind your AI?

Because no single purchase was ever big enough to need one when the adoption started. A single team member's subscription costing less than the office coffee budget does not go on paper or into an H2 strategy planning session. Somebody signs up for Granola so their meetings write themselves up. Somebody else starts prompting by voice with Wispr Flow to get through a morning faster. Not much was thought about it at the time, and there was no reason to.

This was not a failure. The businesses that waited for a formal AI programme before letting anybody touch anything moved slower, learned less, and are still not getting the productivity these teams are getting. Your people solved their own problems with tools they could afford, which is what you would want them to do in a free market.

The cost turns up later, and the cost is fragmentation. Two years of choices made team by team, by capable people, for sensible reasons, and what you are left with is a set of tools that do not talk to each other. The same problem gets solved twice in two departments. What one team produces sits somewhere the next team cannot reach. Every individual decision was defensible and nobody has ever looked at the whole.

Who in your business already knows the answer?

The people who made the choices in the first place. They still work here, and in most businesses nobody has asked them.

Whoever picked that writing tool remembers why they picked it over the other one, what they hoped it would do and what it did instead. Whoever ran the chatbot trial remembers the thing they tried in March that had stopped by May. Between eight and ten people, your business already holds a detailed account of what AI has done for it, and that account has never been in one room.

There is one thing worth setting aside before you start. If you want a number for what your copilot licences did to productivity, that needs a controlled trial and somebody willing to fund it. Put that question down. Chasing it is the thing that stops most businesses from doing the useful version, which costs an afternoon and needs no budget at all.

Which questions do you ask?

Get the people who use it around a table for an afternoon and work through five things.

1. Start with what worked. Not what got adopted, but what changed something. Somebody will tell you the service team now closes a whole category of ticket without escalating it, and that is a real answer.

2. Move to what was chosen, and why. Every tool in the building beat an alternative in somebody's head. Those reasons are recoverable today and gone in two years.

3. Then what was tried and dropped. This is the most useful answer in the room and the slowest to arrive. A tool abandoned after six weeks taught you something exact about your data, your processes or your appetite.

4. Take a tally of what is still on the list. Everybody has a list. It tells you what your teams believe is possible, which is a better guide to where value sits than any vendor deck.

5. Close with which conversations have not happened. There is usually at least one, most often a few. Two teams solving the same problem separately, or a tool that would work across the business if one team stopped treating it as theirs.

What do you have at the end?

A strategy built from evidence rather than from a forecast. It says where AI already earns its place in your business, based on what has worked rather than on what a supplier claims. It says where the gaps are, and those are rarely where people expect. It also gives you an order to do things in, which is the part most AI strategies skip.

This is decision intelligence in its plainest form. Use what the business already knows to work out what it should do next. Technology makes that faster and connects it to your data, and the principle works with eight people and a whiteboard. Keeping the reasoning after that afternoon is a separate problem, and we wrote about it here.

What does this look like when we do it?

This is what Unloq was built for. We build decision intelligence, which means connecting the data a business already holds to the decisions its leaders have to make, and being able to say afterwards why a decision was taken and what happened next.

Most of that work starts exactly where this article starts. Not with the technology, but with working out where a business actually is and what it should do first.

We ran that exercise with Republic M. Kevin Justice, their CFO, described it afterwards:

"Unloq helped Republic M understand where we were on our AI journey and, more importantly, where we should start. What impressed us was the time they took to understand our business, our strategy, our systems, and the real pain points across the organisation. Their methodology gave us a clear view of our AI maturity, a practical roadmap for adoption, and a set of prioritised, fully scoped use cases aligned with our commercial and operational goals."

That was a discovery piece, so there are no outcome figures attached to it and we would be wary of anyone who offered you some at that stage.

Sometimes the answer is that the data underneath needs work before anything else is worth doing. On the XPENG UK launch with International Motors Group we reviewed more than 5,000 custom data fields in a full enterprise data audit and analysed 599 live support tickets. The part that took longest was working out which questions the business could not answer from what it already held.

Come and have the conversation

We have this conversation often enough that we started hosting it properly. A few times a year we put a table of leaders together over dinner and work through it, and the next one is for automotive.

Tuesday 22 September, 6:30pm to 10pm, at The Ivy on Temple Row in Birmingham. One table, sixteen people who run automotive businesses, hosted with S&W and DRPG. There is no charge. Yiannis Maos opens with a short welcome and then hands over, because the guests are the point.

The evening runs on five questions, close cousins of the ones above: what has AI actually changed about a decision you make, who owns AI in your business and is it working, what is on the list you have decided not to do, what would break if all of it stopped this afternoon, and what would you need to know to grow the business five times.

Ask for a seat. There are sixteen and every one is confirmed by hand. The list closes on Monday 14 September. If automotive is not your sector, send us a note and we will keep a seat for the next one.