Planning marketing with AI, in plain words

A doctor works out what is wrong before writing the prescription.

Mark Ritson, a marketing professor, says good marketing runs in that order. First work out what is happening in the market and in your numbers. Then decide where you will play and how you will win. Tactics come last.

He has a name for skipping straight to tactics: "tactification". It is the most common way a plan goes wrong.

Where AI helps

Reading your numbers, reviews and competitor pages quickly, and listing what might be going on.

Where you decide

Which explanation is true, and who you are really selling to.

From my own work

At the telepsychiatry company where I worked in 2023, only 1.8% of visitors booked a consultation. More budget would not have fixed that; the page was the problem.

Try this promptHere are our last three months of numbers. List five possible reasons conversion is low, and for each one, what data would prove it or rule it out.

A road trip: where are we, where are we going, which route, who drives, and how do we check the dashboard?

PR Smith built SOSTAC in the 1990s, and it is still one of the most used planning models. Six headings turn a vague idea into a plan that fits on a page.

It works for a whole company or for one campaign. Fill in the headings in order and you have a plan.

Where AI helps

Writing the situation summary and a first draft of the whole plan. A CIM article on SOSTAC in the AI age says to treat that output as a first draft and check it covers all six parts.

Where you decide

Whether the strategy really gives you an edge. AI can write a sensible strategy, and so can the AI at the company next door.

Try this promptWrite a one-page SOSTAC plan for [business] from these notes. Mark every guess as a guess.

A dartboard: throw at a lot of places, aim at the best few, then keep hitting the middle.

The book Traction lists nineteen ways to get customers, such as search, email, partnerships and events. Most teams pick the channel they already know. The Bullseye asks you to look at all of them first.

The book is usually summarised as three moves: brainstorm every channel, test the strongest ones for a small cost, then focus on the one that works.

Where AI helps

Brainstorming an idea for every channel, and designing a small test for each.

Where you decide

Which tests you can afford, and what result counts as "working".

Try this promptFor each of these channels, write a test I can run in two weeks for under [budget], and say what I would measure.

A quick vote on a long list of ideas.

Score each idea from 1 to 10 on three things: how big the win could be (Impact), how sure you are (Confidence) and how easy it is (Ease). Multiply the three, or average them. Sources differ, and either works if you do it the same way every time.

Intercom's version, RICE, adds Reach, how many people it touches, and divides by Effort. ICE is faster. RICE is stricter.

Where AI helps

Filling in a first score from past results, and listing what each idea needs.

Where you decide

Confidence. AI is happy to give every idea a nine.

From my own work

At a games publisher, studios, partners and deals each got a written score, and my time went to the few at the top. Fifty vetted studios became five signed ones.

Try this promptScore these ideas out of 10 for impact, confidence and ease, and back every confidence score with evidence we already have.

Planting and harvesting. You need both, and mostly planting.

In The Long and the Short of It (2016), Binet and Field studied close to a thousand campaigns from the IPA Effectiveness Databank. The best results came from putting about 60% of the budget into building the brand and 40% into activation.

Treat it as a starting point. Later work moved the best split further towards brand, and Binet himself says it changes by category.

Where AI helps

Sorting your spend into the two buckets, and showing what-if splits.

Where you decide

The split itself. It depends on your market and how long people take to decide.

From my own work

At that telepsychiatry company I moved 40% of the Meta brand spend into high-intent search, because people searching were already looking to book. The rule is a guide, not a law.

Try this promptHere is our spend by channel for six months. Sort it into brand building and activation, and show the split today.

A recipe or a chef. A recipe has fixed steps. A chef decides the next step.

Anthropic, the company behind Claude, splits AI systems in two. In a workflow, the path is set in advance. In an agent, the AI chooses its own next step. Its advice is to start with the simplest thing that works, because agents cost more and run slower.

It lists five common workflow patterns. Chain: one step feeds the next. Route: sort the input and send it to the right place. Parallel: do several things at once and combine them. Orchestrator: one AI splits the job and hands out the parts. Check-and-improve: one AI writes, another critiques.

Ethan Mollick, who writes about AI at work, gives four rules in his book Co-Intelligence: use AI for everyday tasks to learn what it is good at, stay the human in the loop, tell it what kind of helper to be, and assume today's AI is the worst you will ever use. His "jagged frontier" idea is that AI is brilliant at some hard things and poor at some easy ones, so you have to test it.

Where AI helps

Anything that repeats: research, first drafts, reports, sorting.

Where you decide

The check. Every workflow gets one point where a person reads the result before it goes anywhere.

From my own work

At a B2B agency, weekly competitor and account research went from four hours to one with Claude, n8n and Make. It was a workflow with fixed steps, and a person read each brief.

Try this promptTurn this weekly task into steps. For each step, say what goes in, what comes out and where a person checks.

A leaky bucket. Find the biggest hole and fix that first.

McClure, an investor, gave a talk in 2007 called "Startup Metrics for Pirates". The name comes from how the letters sound. The idea is to pick one number for each of five stages, look at where people drop off, and fix the biggest drop first.

The bars above are only a shape. Real drops differ from business to business.

Where AI helps

Pulling the numbers together, spotting the biggest drop and drafting what to test.

Where you decide

What counts as "activated", and whether a fix is worth the effort.

From my own work

The telepsychiatry work followed this order: fix the path to purchase first. That meant one landing page per condition before any new budget.

Try this promptHere are our counts for each stage. Find the biggest drop and list three tests that would address it.

A shop: people walk past, come in, buy, then come back.

Dave Chaffey created RACE at Smart Insights in 2010 for digital marketing. It is the same idea as the pirate metrics with plainer words, and it fits any business, not only start-ups.

Put a target on each stage, then check them in the same order every month.

Where AI helps

Writing the monthly check-in from your data, and flagging what moved.

Where you decide

The targets, and what to do about a stage that is behind.

Try this promptHere are this month's numbers for reach, act, convert and engage against target. Write a half-page summary, flag the stage that is furthest behind, and say what I should check first.

Several of these are summaries of books and talks, so wording differs between them. Where they disagreed, the text says so. Check the original before quoting anyone.