AI
The bottleneck in strategy was never the writing. AI removes that in minutes and leaves the hard part untouched: which customers you prioritise, what you stop doing, and who is accountable for the call. For a UK SME leader, use AI for speed-to-options and keep the judgement.
AI can help you create a marketing strategy document without external expertise, but it cannot reliably create a strategy you can bet the business on. A winning strategy is a set of accountable choices: which customers you prioritise, what you want to be known for, where you invest, what you stop doing, and how you manage risk. AI is strong at speed-to-options — research synthesis, route drafting, scenario planning — and weak at accountability, evidence checks, and resolving trade-offs under real constraints. SMEs get the best results when AI generates multiple credible routes and a senior decision-maker applies judgement, validation and governance to select one coherent plan.
Strategic planning follows a structured process, so I thought: surely I can automate that.
I spent time testing how effective AI is at strategic decision-making for resource-constrained SMEs. The aim was to see whether an SME leader could press a “create my strategy” button, answer some prompted questions to guide the output, and get a unique, bespoke, AI-produced marketing strategy worth acting on.
There are areas where AI, and the stepped use of AI agents, really does help. Rapid deep research and extracting the important learning from large amounts of market, business, financial and competitive data are two obvious examples. But even with well-considered prompt engineering and extended thinking, I often found myself frustrated by the decision-making and the quality of the strategic output. It struggled to apply the “I” in AI.
It frequently based its reasoning on what it read on marketing websites — content it may well have helped to create in the first place.
So I kept revising the prompt, telling it more specifically what to do and what not to do, expecting the right strategic decisions to follow in time. They did not. After many attempts I paused the project. That may change in the months and years ahead, but it had not changed by early 2026.
Something was still missing. Maybe it is real-world experience and human intuition, or the years of building and executing strategies with the accountability of delivering results and learning what does and does not work. We may be underestimating how much that matters in guiding relevant, prioritised decisions.
AI can propose choices quickly, but even when prompted to prioritise it often picks the wrong ones. AI should never be allowed to own what it presents. When SME leaders are busy it is easy to accept its suggestions at face value, and you end up executing its work rather than yours — which destroys productivity, budget and results. If the plan points you at the wrong things, you spend the next 90 days executing and the quarter disappears.
So what is AI’s role in strategic marketing planning?
When SMEs say strategy, they usually mean one of three different things. Separating them makes AI use safer and more effective.
AI helps most with planning and execution, because those stages involve synthesis, drafting and iteration. It can assist strategy development by generating options, but it is not a decision-maker. Accountability stays with leadership.
Yes, it can create a document. That is not the same as creating a strategy you can rely on. A strategy that wins is defined by the hard choices you are willing to stand behind.
Here is the practical test. If your draft strategy does not clearly state what you will stop doing, who you will not prioritise, and what trade-offs you are accepting for the next 90 days, it is not a strategy. It is a set of ideas.
Do-it-yourself is realistic when you already have good customer insight, a clear ideal customer profile, reliable conversion and pipeline data, and a senior person who can make trade-offs under uncertainty.
External expertise pays for itself when you have internal disagreement, unclear positioning, long sales cycles, limited budget or capacity, a complex offer, or claims and compliance risk. In those cases the cost of fast-wrong decisions is usually higher than the cost of senior support.
AI helps you move faster. Humans must decide what is true, what matters, and what you will do next.
AI’s biggest strategic value is speed-to-options. It collapses the slow parts of planning so you can spend your time where it counts: evaluating decisions.
A practical rule: always feed AI both your business constraints and real customer or market evidence. Without both, it will produce something readable and generic.
McKinsey’s analysis of generative AI’s economic potential points to meaningful productivity impact across functions, including marketing work. For SMEs, the best use of the time saved is reinvesting it into better decisions, not producing more noise.
Most people still use these tools one question at a time. The bigger shift is that they no longer have to. The assistants will now run multi-step research on their own: they search, read dozens of sources, follow up on what they find, and come back twenty minutes later with a referenced document. What was a premium feature at the start of the year is now the ordinary way to use them.
For a small business this is the most useful development of the year. The evidence pack in Step 1 below used to be the part that never got done, because it was a week of someone’s time nobody had. It is now an afternoon.
Two things to hold on to. The output is a starting point for a conversation, not a finished analysis, and the quality of the question still decides the quality of the answer. And a longer document with citations feels more authoritative than a short one, which makes it easier to skip the checking. Pick the five claims the strategy actually rests on and verify those yourself.
A growing number of tools offer synthetic respondents, meaning AI personas that answer survey questions as your target customer would. For a business that cannot afford fieldwork the appeal is obvious.
The Market Research Society has looked at this properly and its position is worth borrowing. Synthetic responses are reasonable for generating hypotheses, testing early concepts and shaping questions. They are not a substitute for primary research when you are validating a decision, and they are weakest exactly where insight usually comes from, which is the outliers and the unexpected answers. MRS also notes that these models lean heavily on Western and particularly US data, and under-represent differences in ethnicity, disability and gender.
For a UK SME the practical translation is short. Use AI to work out what to ask your customers. Then ask your customers.
The danger is not that AI produces nonsense. It is that it produces plausible writing that reads like a competent consultant until you check the assumptions.
This one is worth a worked example, because it happens on exactly the topic you are researching.
Ask an assistant how well traffic from AI search converts compared with ordinary search and you will get a confident multiple. The figures in circulation include 4x, 4.4x, 4.5x, 5x, 6x and 23x. Every one of them traces back to a vendor blog with no published methodology, and they are quoting each other. There is no underlying study.
A plan built on numbers like these looks rigorous and is not. The test is simple: can you click through to a named organisation, a stated sample size and a date? If not, the number is decoration. Use it as a hypothesis to check, never as a reason to move budget.
These models are trained overwhelmingly on US and Western material, a point MRS makes directly in its guidance for client-side teams. Ask about buying behaviour, channel norms, pricing conventions or regulation and you will often get a confident American answer wearing British spelling.
It matters most where you are least likely to notice: procurement habits, what a decent lead costs, how long a sales cycle runs, which channels your buyers actually use. Say “UK” in the prompt, say what sector and what size of business, and sense check anything that sounds like it came from a different market. It usually did.
OpenAI published research in September 2025 setting out why this happens, and the explanation is more useful than the symptom. Models are trained and scored in a way that rewards guessing. On a test where a wrong answer and a blank answer both score zero, guessing is the better strategy, so that is the behaviour that gets learned. As the paper puts it, if the main scoreboards keep rewarding lucky guesses, models will keep learning to guess.
That has a practical consequence for anyone using AI to build a plan. The model is not trying to mislead you and it is not broken. It is doing what it was rewarded for. So make the alternative available: tell it explicitly that “I do not know” is an acceptable answer, ask it to mark which claims it is confident about and which it is inferring, and treat a confident unsourced number as the least reliable thing in the document.
Fast drafting is helpful. Fast wrong is expensive.
Use this as a quick test. If several rows fall into “needs senior leadership” and you do not have it in-house, external expertise is usually the fastest route to a plan that performs.
| Task | AI helps most with | Humans must lead | Risk if AI-only | 80/20 safeguard |
|---|---|---|---|---|
| Market and competitor synthesis | Fast summarisation, drafting variants | Setting direction, judging quality | Wrong assumptions | Validate against real sources |
| Segmentation hypotheses | Proposing plausible segments | Validating with customer evidence | Targeting “everyone” | One primary plus one secondary segment |
| Positioning options | Generating distinct narratives | Choosing trade-offs, approving claims | Generic messaging | The competitor copy-paste test |
| Channel-mix scenarios | Drafting plans for constraints | Selecting the strategic path | Shiny-object plans | One capture plus one nurture channel |
| Budget logic | Turning notes into a coherent narrative | Owning the allocation | Unrealistic spend | A 90-day bet and a stop-doing list |
| Measurement | Drafting KPI ideas and reporting | Defining success and targets | Vanity metrics | Five KPIs maximum, with owners and a cadence |
| Risk and governance | Flagging potential issues | Ensuring ethical use and accountability | Reputation and legal risk | “Source it or remove it”, plus approvals |
Reality check: if you cannot explain why you chose this strategy in one minute, AI made the document but leadership did not make the plan.
AI raises the ceiling on what a marketing team can produce in a week. That is exactly why leadership matters more. When it becomes easy to generate ten strategies, the hard part becomes choosing the right one and holding the organisation to it.
A simple leadership model for AI-assisted strategy work:
Leadership is not about writing faster. It is about deciding better.
If you want AI to improve strategy work rather than decorate it, you need a process that treats AI as an option-generation engine with a human decision gate at every stage.
Do not try to win fifty prompts at once. Start by being the best answer to a small set of high-value questions.
Good inputs beat clever prompts. Gather commercial targets, ICP notes and sales call themes, win/loss learnings and objections, pipeline and conversion data, website and campaign analytics, a handful of customer interviews, and your real constraints: team capacity, budget, sales motion, delivery limits.
Add one line to the evidence pack this year: where does our demand actually come from now, and is that still true?
Pew Research tracked the real browsing of 900 US adults across nearly 69,000 Google searches and found that when an AI summary appeared, people clicked a result 8% of the time, against 15% when it did not. Only 1% clicked a link inside the summary itself. It is US data from March 2025, so treat the exact figures as indicative rather than your numbers, and the direction is not in doubt.
For planning, the point is not to panic about search. It is that an assumption you last examined two years ago may now be quietly wrong, and a plan is only as good as the assumption underneath it. Check your own analytics before you decide what to fund.
Require it to separate evidence from assumptions. If it introduces facts, it must provide sources or you delete them.
Each route needs a target segment focus and who you are not prioritising, a positioning claim with proof points, a channel bet with what you will stop doing, and 90-day priorities.
Kill or keep, on four questions. Is it different — could a competitor credibly say the same thing? Is it true — do we have evidence, or a fast plan to get it? Is it feasible against capacity, budget, sales motion and delivery? Is it safe on claims, compliance and data handling?
Objectives, segments, proposition, channel strategy, budget logic, a short roadmap, and a KPI model. Keep it simple enough to execute.
Let AI draft the reporting narrative and highlight anomalies. Humans still decide what changes, and why.
Governance sounds like paperwork until a draft plan includes sensitive data, unsubstantiated claims or risky recommendations. Then it becomes a growth risk. Four areas to lock down early:
Two changes since this article was written, and neither is the one people expect.
The Data (Use and Access) Act took effect on 5 February 2026, and the ICO updated its legitimate interests guidance on 23 March 2026. Direct marketing is now named in UK GDPR as a purpose that can constitute a legitimate interest. Useful, and not a free pass: the ICO is explicit that it does not apply automatically, so the three-part test still has to be done and recorded, and PECR still governs how you make contact.
The EU AI Act’s transparency rules apply from 2 August 2026, with a grace period into December for systems already on the market. They are narrower than the coverage suggests. The labelling duty covers deep fakes, meaning synthetic image, audio or video that would pass as real, and AI-generated text on matters of public interest published without human review. A marketing plan drafted with AI and reviewed by a person is neither.
The UK has no AI labelling law. What applies is the ordinary rule against misleading people, and the ASA expects disclosure where AI use is material to a decision. Invented case studies, synthetic testimonials and fabricated statistics fail that test in any jurisdiction.
One more worth adopting from the MRS guidance: be straight with your own people about where AI was used in the plan. It is easier to challenge an assumption when you know a machine made it.
You can start measuring impact without specialist platforms. The key is consistency: track the same questions monthly and improve the content that fails to show up.
For lead tracking, add a “how did you hear about us?” field including an AI assistant option, and encourage sales to tag leads who reference AI tools.
Because the expensive part of strategy is not writing. It is choosing the right bet, validating assumptions quickly, and aligning execution to constraints. External expertise is most valuable when it prevents wasted spend and reduces strategic risk.
Build a short evidence pack, generate three to five routes with forced trade-offs, run a decision gate, then commit to a 90-day plan with three priorities and a tight learning loop.
Start by understanding the cause. OpenAI’s own research found models guess because training and scoring reward guessing over admitting uncertainty. They are not malfunctioning when they invent a number, they are doing the thing that scored well.
So four habits. Tell the model that “I do not know” is an acceptable answer. Ask it to separate what it is confident about from what it is inferring. Insist on a named source, a sample size and a date for any statistic that will influence a decision. And keep a person accountable for the five claims your strategy actually rests on, by name, before the plan is signed off.
Yes, if someone senior can own the decisions. Many SMEs use focused strategic support to set direction, governance and a 90-day execution plan.
For working out what to ask, yes. For deciding anything, no.
The Market Research Society’s position is that synthetic respondents suit hypothesis generation and early concept testing, and do not replace primary research when you are validating a decision. They are weakest on the outliers and surprises, which is usually where the useful insight lives, and they carry a heavy US and Western data bias.
Ten real conversations with your own customers will beat a thousand synthetic ones, and for most small businesses ten conversations is a fortnight’s work. We often start there with clients, because it is the fastest route to a strategy anyone believes in.
AI planning tools are not replacing strategic thinking. They are raising the standard. The SMEs that win will use AI for speed while keeping humans accountable for direction, judgement and governance.
If you want help turning AI drafts into a strategy your team can execute, Epitomise provides senior, practical strategic marketing leadership: expert-led strategy development, mentoring, training and execution support, scoped by outcome rather than by hours. Explore the Transform Accelerator or book a discovery call.
External sources open in a new tab.
Who wrote this
Steve founded Epitomise in 2017 after UK, international and global marketing leadership roles, most recently as Global CMO of the Vitec Group’s Videocom Division. He works with SME and technology businesses on strategy, positioning and the execution that follows — more about Steve.
Read next
How UK SMEs can use AI across the whole buyer's journey, from discovery to decision, with 2026 evidence on what has actually changed. Includes where Claude Cowork and Claude Code fit for a small team.
ABM for UK SMEs: how to run account-based marketing with a small team, and use AI for account research and outreach without losing credibility. Updated for 2026.
Run marketing as a repeatable quarterly cycle — Review, Focus, Implement, Optimise — so strategy turns into shipped work and measurable pipeline.
A short discovery call is usually enough to tell. No deck, no pitch.
Book a discovery call