[TL;DR: Using AI to write individual ads or emails gives you a speed boost, not a strategy advantage. The businesses pulling ahead are building compound marketing systems where every campaign teaches the system something, and the outputs get better over time, not just faster.]

Why AI for Marketing Isn't Paying Off for Most Small Businesses

Most businesses using AI for marketing follow the same arc. First month: genuinely exciting. The copy gets written in minutes. Research that used to take hours happens before lunch. Second month: the novelty fades. Third month: results look similar to before, just with less effort to produce them.

That's not a failure of AI. That's a failure of how it's being deployed.

There's a concept from the software development world called "compound engineering." When teams started applying it seriously in 2025, feature cycles that used to take weeks dropped to 1-3 days. Not because AI got smarter, but because the teams got smarter about how they built systems around it. The same principle applies to marketing, and almost no one is using it yet.

The Five Levels Nobody's Applying to Marketing

In early 2026, investor and technologist Dan Shapiro published a framework for AI adoption in software development, describing five levels from basic autocomplete to fully automated "dark factories" where no human ever reviews the code. The five-level model maps almost perfectly to marketing.

The reason most businesses plateau at Level 1-2 is not capability. It's investment. Reaching Level 3 and beyond requires what Shapiro calls investing in "the invisible work": the context, the patterns, the feedback loops that make the system smarter over time.

What Compound Engineering Actually Means for Your Campaigns

The compound engineering research, documented by Kieran Crowley at Every while building an AI email assistant called Cora, identified something counterintuitive: teams that spent 50% of their time investing in the system rather than doing the work got dramatically better results than teams that spent 100% doing the work.

The principle: every time something works, capture why it worked. Every time something fails, fix it at the source, not just the output.

Most marketing teams do the opposite. An ad hits 5x normal click-through rate, they note it and move on. A campaign falls flat, they blame the audience or the platform and try something different next week. The insight never gets captured. The system never learns. Six months later, they're facing the same problems with the same guesswork.

Applied to marketing, compound engineering looks like this:

This is why some businesses see AI as genuinely transformative and others see it as "meh." The first group built a system. The second group used AI for tasks.

Sutherland's Warning: You Might Be Optimising the Wrong Thing

Here's where it gets more interesting. Even if you do build a compound marketing system, Rory Sutherland, Vice Chairman of Ogilvy UK and one of the most challenging thinkers in applied behavioural economics, would ask you a difficult question: are you certain you're solving the right problem?

Sutherland's central argument in Alchemy is that humans systematically overvalue logical, measurable solutions and undervalue psychological ones. When a problem looks like it needs to be solved faster, we assume the solution is to do it faster. But often the real problem is something else entirely.

He uses the example of the London Underground. Engineers spent enormous sums building faster train routes. The single biggest improvement in passenger satisfaction per pound spent? Dot-matrix display boards showing when the next train would arrive. The wait itself did not change. The uncertainty did. That was the real problem all along.

"Not everything that makes sense works, and not everything that works makes sense."

Rory Sutherland, Alchemy

Apply this to AI marketing systems. You can build an efficient compound system that produces more content faster, optimising for reach, volume, and publishing frequency. But if the content is solving the wrong problem, if it's targeting the wrong moment in the buying journey, making the wrong claim, or resolving an anxiety the customer doesn't actually have, you've automated yourself into mediocrity at scale.

Sutherland calls this the "physical fallacy": the assumption that the measurable, operational version of a problem is the one worth solving. Often it is not. Often what matters is the perceived version of the problem, which is psychological, not operational.

Before you build a compound marketing system, answer these questions honestly:

• Is our bottleneck volume of content, or quality of our value proposition?
• Do our prospects understand what makes us different from the obvious alternatives?
• Are we reaching the right people and failing to resonate, or not reaching enough people at all?
• What anxiety does our customer have at the moment they're deciding whether to contact us?

The Two Investments That Actually Matter

The synthesis of Shapiro's five levels and Sutherland's warning gives you a clear framework. Building a compound marketing system that avoids the common traps requires two parallel investments, not one.

Build a context library that includes:

• 20-30 examples of your best-performing ads, each annotated with why they worked


• Your customers' actual language, drawn from reviews, testimonials, and sales conversations, not what you think they say


• Your documented positioning relative to each key competitor


• Your content rules: what you never say, what always converts, what reliably falls flat

Before building the system, audit what you're actually solving:

• What triggers a buyer to search for what you offer? This determines your messaging hierarchy.


• What anxiety stops them from converting once they find you? This determines your trust signals.


• What does "good" look like to them before they have enough experience to judge you properly? This determines your credibility signals.

What This Means for Your Business

If you're using AI to produce marketing content faster and wondering why it's not moving the needle, you're likely at Level 1 or 2 of the adoption curve. That's where most businesses are. It's a reasonable starting point, not a permanent destination.

The move to Level 3 starts with one decision: the next time something works, document why before moving on. The next time something fails, diagnose the root cause before changing tack. Build a feedback loop, not just a production line.

The move past Level 3 requires honesty about what problem you're actually solving. Faster content is not always the answer. Sometimes the bottleneck is psychological, not operational. Sometimes the real problem is the anxiety your customer feels at the moment of decision, and a better-worded page about your guarantee matters more than fifty more ads.

The businesses that will look back at 2026 as the year they got ahead were not necessarily the ones who adopted AI first. They were the ones who built the better system, and made sure that system was solving the right problem to begin with.

FAQ

How long does it take to build a compound marketing system?

Expect three months of consistent effort before you notice the compounding effect. The first month is context investment: documenting your brand voice, capturing your audience's real language, mapping your competitive positioning, and building your content rules. This feels slow and invisible, but it's the foundation everything else builds on. The second month is pattern capture: adding structured review into your campaign workflow so that every test teaches the system something. By the third month, briefing your AI produces dramatically better first drafts because the accumulated context does the heavy lifting. Most businesses expect AI to improve results immediately. Compound systems improve results gradually and then suddenly, in a way that feels disproportionate to the effort.

Can a small business with one marketer actually build this kind of system?

Yes, and often a small business builds it more effectively than a larger team. Large marketing departments have institutional habits and approval workflows that make it hard to invest in invisible infrastructure. A solo operator or a tight team can dedicate 30 minutes per week to pattern capture and context building without fighting internal bureaucracy. The system does not need to be technically complex. A well-organised document containing 50 examples of great-performing ads, each annotated with why they worked, is worth more than an expensive AI marketing platform with no contextual intelligence behind it. Start simple, add to it consistently, and the value compounds.

What's the biggest mistake businesses make when adding AI to their marketing?

Using AI as a replacement for strategy rather than as an amplifier of it. AI produces content at scale and speed. If your strategy is unclear, your positioning is weak, or your value proposition is not solving the right problem, AI will produce more of that same mediocrity at speed and scale. The Sutherland warning applies directly: build a faster machine only after you are confident it is pointed in the right direction. The best AI marketing results we see consistently come from businesses that had a clear, differentiated position before they introduced AI, and used AI to execute that position more efficiently. Clarity of strategy before capability of tools.

How do I know if I'm solving the wrong problem?

A useful test: if doubling your ad budget would not meaningfully change your results, the bottleneck is not reach or volume. It's conversion, positioning, or trust. Signs you may be solving the wrong problem include: high click-through rates paired with low conversion rates (people are interested but something stops them); strong brand recognition but poor consideration (they know you but don't think of you first when the need arises); low repeat business despite positive customer feedback (you're not staying mentally available). In each case, producing more content faster will not fix the underlying issue. The audit comes first.

Further Reading

• Alchemy by Rory Sutherland - The strongest case available for solving psychological problems rather than operational ones
• Co-Intelligence by Ethan Mollick - The most grounded research on how humans and AI actually collaborate effectively, including why most people plateau at surface-level use
• How Brands Grow by Byron Sharp - The empirical foundation for understanding what actually drives marketing effectiveness at scale
• The Long and the Short of It by Les Binet and Peter Field - IPA effectiveness research showing how investment in systems (brand) vs tasks (activation) produces different compounding returns
• Compound Engineering: The Definitive Guide - Kieran Crowley's framework for investing in AI systems rather than individual AI tasks, directly applicable to marketing workflows