In July I spent 90 minutes in a room with twelve food and drink founders and operators, ranging from large national brands to smaller start-ups.

Before we started, most of them could name two or three things they used AI for and by the end there were 140 activities on the board.

The focus was on learning about their own businesses rather than AI specifically. Every one of those 140 things was already happening, taking somebody's time and an opportunity hiding in plain sight.

So why couldn't they see it beforehand?

Because the question is too big. If I sat you down and asked you to think of fifty ways to use AI in your business, you'd be there all afternoon and you'd get to about nine. That's not a knowledge problem. You know your business better than anyone. It's a search problem, and an open question gives your brain nowhere to start.

In consulting you learn about structured brainstorming. When you need a team to generate a lot of ideas fast and you can't afford to miss an entire category, you never ask the open question. You give people a structure and let them fill it in. It helps the brain to narrow the thinking down into bite-size chunks.

"How could we use AI?" is a wide question but "How could AI help us research our supply chain?" takes twenty seconds. The exercise is moving through the second type of question 42 times very quickly…

Every food and drink brand runs on the same seven types of work. £50k of turnover or £50m, it doesn't change.

Sales and commercial. Operations and fulfilment. Finance and admin. Marketing and brand. Procurement and supply chain. Product and NPD. Compliance, quality and food safety.

Tilda has a procurement department. If you're a one-person brand, procurement is a Tuesday afternoon and a WhatsApp thread with your suppliers and co-packer. Same work, different scale (and number of people doing it!)

That matters because the areas you cover badly are usually the ones nobody owns.

Write the seven down first. That's your one axis.

The second axis is RECIPE, which is the six types of work AI can take on.

Research. Finding and pulling together outside information. Backup suppliers, competitor pricing, getting your head round EPR.

Evaluate. Checking one thing against another. A health claim against the reg, a PO against your agreed price list, artwork against the last approved version.

Create. Generating content and first drafts. This is the one everybody's already doing.

Interpret. Finding patterns in your own data. True margin by channel once you've stripped out promotions and duty. Which SKUs are quietly costing you money.

Plan. AI as a thought partner. Buyer meeting prep, range strategy, scenario planning a factory shutdown.

Execute. Taking the action, not suggesting it. Processing the order, raising the invoice, updating the 3PL.

Nearly everyone starts at Create, because it's the most visible thing AI does and the least frightening. It's also, for most food and drink businesses, the least valuable one on the list.

Seven areas by six capabilities gives you 42 boxes. That's the board, and each box is one narrow question. Here's what an example looks like for each letter of RECIPE:

Research crossed with sales. Pull together your competitors' pricing across the category, what they're charging for a comparable product, across a range of different retailers.

Evaluate crossed with operations. Your customers' POs come in. Check the pricing on each one against what you actually agreed with them, before anybody raises an invoice.

Create crossed with marketing. Social captions, newsletters, first drafts of copy.

Interpret crossed with supply chain. Take your cost of goods over a period and work out which SKUs' costs are rising the most, why, and what that's doing to your gross margin.

Plan crossed with sales. You've got a big buyer meeting coming up. Give it your deck and the background on your brand, ask what questions the buyer's likely to come back with, then role play it until you've got the answers.

Execute crossed with operations. Processing sales orders end to end, without anybody retyping anything.

The rule I give people at this point is don't constrain yourself to what you think AI can do right now. Write down what you currently do in your actual business, then go and find out whether it's possible. The pace of change means something that isn't quite there today may well be in three months, and you'll never go back and check on an idea you never wrote down.

You don’t have to fill in every box either. Sometimes that means the box genuinely doesn't apply to you. But it could also mean nobody's ever thought about it.

Draw the grid. Seven rows for your business areas, six columns for RECIPE. A wall and some sticky notes is best because people can see each other's, but a shared spreadsheet works fine.

Give everyone fifteen minutes and one instruction: write down as many activities as you can, one per note, and don't judge any of them. Not "good AI ideas". Just things you actually do. Go box by box rather than letting people roam, because roaming puts you back into the open question.

Two rules: Nobody edits during the fifteen minutes, and nobody's allowed to say "AI can't do that yet". You're building the long list, not the plan. Something that wasn't possible in March may well be by September, and you'll never revisit an idea you never wrote down.

The team at Dr Will's got to 67 use cases in 25 minutes doing exactly this.

A hundred and forty opportunities is not a hundred and forty projects. You’re next step is to evaluate these for value and risk -which I’ll talk about in the next newletter.

Draw the grid on one sheet of A4, a spreadsheet or a slide. Then fill in one row only, whichever business area eats most of your week. Six boxes, fifteen minutes.

I run these as free group sessions with a handful of food and drink brands at a time, and one-to-one deep dives with individual teams. You can sign up to the next one here.

Next issue: the five types of value, and why hours saved is the laziest one.

Sanjay is co-founder of Pesto, AI-native ops software for UK food and drink.