AI-Enhanced Advertising: Where the Gains Are Real

AI-enhanced advertising strategies use machine learning, predictive modelling, and automated optimisation to improve how ads are planned, targeted, and measured. Done well, they reduce wasted spend, surface audiences that manual targeting misses, and free up strategists to focus on decisions that actually require human judgement. Done badly, they automate mediocrity at scale and give you a false sense of control.

The technology has matured enough that the question is no longer whether AI belongs in your advertising stack. It does. The question is where it genuinely earns its place, and where it just adds complexity without adding value.

Why Most AI Advertising Conversations Start in the Wrong Place

When I was running iProspect, we grew from around 20 people to over 100, and moved from loss-making to one of the top five agencies in our market. A lot of that growth came from getting sharper about where we were actually creating value for clients versus where we were just producing activity. Advertising technology, including the early versions of what we now call AI-driven tools, had a habit of creating the appearance of sophistication without the substance.

The conversations happening now about AI in advertising often repeat the same pattern. There is a lot of focus on the tools, the platforms, the automation features, and not enough focus on the strategic questions those tools are supposed to answer. What problem are you solving? Who are you trying to reach that you are not reaching now? What does growth actually require for this business?

If you are looking for a broader framework for how advertising fits into growth strategy, the Go-To-Market and Growth Strategy hub covers the commercial context that makes individual tactics worth pursuing.

AI does not answer those questions. It optimises toward whatever objective you set. If you set the wrong objective, it will optimise toward the wrong outcome with impressive efficiency.

Where AI Genuinely Improves Advertising Performance

There are areas where AI has moved from novelty to genuine operational advantage. These are not speculative. They are places where the technology has been tested at scale and where the evidence is consistent enough to act on.

Bidding and Budget Allocation

Automated bidding has been the clearest win. The volume of signals that feed into a real-time bid decision, device, time of day, location, browsing history, audience overlap, competitive pressure, is beyond what any human can process at the speed required. Smart bidding systems across Google, Meta, and programmatic platforms have consistently outperformed manual bidding in controlled tests, particularly when campaigns have sufficient conversion volume to give the model something to learn from.

The caveat is important: these systems need data. A campaign running 15 conversions a month does not give an AI model enough signal to optimise meaningfully. In those cases, you are often better off with simpler manual controls and a focus on building volume before handing the wheel to automation.

Audience Modelling and Lookalike Expansion

Lookalike modelling has been around long enough that we sometimes forget how significant it was. The ability to take a seed audience of your best customers and find statistically similar people at scale changed how acquisition campaigns were planned. AI has made that modelling more granular and more dynamic. Platforms can now update audience models in near real time as new conversion data comes in, rather than relying on static snapshots.

The strategic question this raises is whether lookalike expansion is actually reaching new audiences or just recirculating around the same pool of high-intent people who were going to find you anyway. I have seen this play out many times. A client celebrates strong ROAS from a lookalike campaign, but when you look at incrementality, you find that a large portion of those conversions were people already in the consideration set. The AI did its job. The strategy was the problem.

Creative Testing and Dynamic Optimisation

Creative has historically been the bottleneck in advertising experimentation. Running a proper multivariate test across headlines, images, calls to action, and audience segments required either a large budget or a long timeline. AI-driven creative tools have compressed that significantly. Dynamic creative optimisation platforms can now assemble and test combinations at a scale that would have been operationally impossible five years ago.

This is genuinely useful. Not because the AI has taste, it does not, but because it removes the mechanical constraint on testing. You can now run more experiments, learn faster, and retire underperforming creative without waiting for a quarterly review cycle.

The Lower-Funnel Trap AI Makes Worse

Earlier in my career I overvalued lower-funnel performance. I was not alone in that. The whole industry was moving toward last-click attribution and treating conversion volume as the primary signal of advertising effectiveness. It took years of looking at incrementality data, running brand-lift studies, and watching businesses plateau despite strong performance metrics to understand what was actually happening.

A lot of what performance marketing gets credited for was going to happen anyway. Someone who is already in your consideration set, who has visited your site twice and searched your brand name, is highly likely to convert. Serving them a retargeting ad and claiming credit for that conversion is not growth. It is expensive confirmation of an existing intention.

Think about how a clothes shop works. Someone who tries something on is far more likely to buy than someone who is just browsing. The fitting room is not the reason they buy. The brand, the product, the experience, the awareness that brought them into the shop in the first place, those are the reasons. Performance marketing often takes credit for being the fitting room when the real work was done much earlier.

AI accelerates this problem because it is very good at finding the people most likely to convert based on existing signals. Those people are disproportionately close to purchase already. If your AI-optimised campaigns are consistently delivering strong ROAS but your market share is flat, that is the pattern you are looking at. The algorithm is efficient. The strategy is not working.

This is not a criticism of AI. It is a criticism of how objectives are set. Market penetration requires reaching people who do not yet know they want what you sell. That requires deliberate investment in upper-funnel activity that AI models will not naturally gravitate toward, because the conversion signals are weaker and the attribution is messier.

How to Set AI Up to Work on the Right Problem

The practical question is how you structure AI-enhanced advertising so that it serves your actual growth objectives rather than just optimising toward the nearest measurable proxy.

Define Incrementality as a Success Metric

Before you hand a campaign objective to an AI system, ask what incremental outcome you are trying to drive. Not total conversions. Not ROAS. Incremental conversions, meaning conversions that would not have happened without the advertising. This requires holdout testing, geo-based experiments, or conversion lift studies. It is more work than reading a platform dashboard, but it is the only way to know whether your AI-optimised campaigns are actually growing the business.

Platforms like Meta and Google have built-in lift measurement tools. They are imperfect, but they are better than attribution models that assume the last touchpoint caused the conversion.

Separate Prospecting From Retargeting in Your Architecture

One of the most common structural mistakes I see is allowing AI systems to blend prospecting and retargeting audiences in the same campaign. The algorithm will naturally shift budget toward retargeting because those audiences convert at higher rates. Your prospecting budget quietly disappears into audiences that were already warm, and you wonder why new customer acquisition has stalled.

Keep these separate. Set explicit budget floors for prospecting. Exclude existing customers and recent site visitors from upper-funnel campaigns. Give the AI a clearly defined job in each case, and do not let it blend the two because it is easier to manage.

Feed the Model Better Signals

AI models are only as good as the conversion signals you give them. If you are optimising toward purchases, you are training the model on a relatively small and potentially unrepresentative sample. Consider passing more granular signals: add-to-cart events, high-value page visits, video completion rates, email sign-ups. This gives the model more data to work with and can improve the quality of the audiences it builds.

First-party data is increasingly important here. As third-party cookie deprecation continues and signal loss increases across platforms, the brands with clean, well-structured first-party data will have a meaningful advantage in training their AI systems. This is not a future concern. It is already affecting campaign performance.

The Measurement Problem AI Cannot Solve

One of the things I took away from judging the Effie Awards is how rarely the campaigns that drove genuine business results were the ones with the most sophisticated measurement setups. Some of the most effective work was also the hardest to attribute precisely. Brand campaigns that shifted category perception over 18 months. Sponsorships that changed how a product was perceived in a market. These things moved the needle on revenue, but they did not show up cleanly in a performance dashboard.

AI optimises what it can measure. That is its fundamental constraint. The parts of advertising that are hardest to measure, brand salience, mental availability, category entry points, are often the parts that matter most for long-term growth. No amount of algorithmic sophistication changes that.

This does not mean you should abandon measurement. It means you should be honest about what your measurement setup can and cannot see. Go-to-market execution has become harder partly because the measurement environment is more fragmented, not less. Treating AI-generated attribution data as ground truth is a mistake. Treating it as one useful signal among several is the right approach.

Marketing mix modelling, even in simplified form, gives you a view of advertising effectiveness that platform attribution cannot. It is not perfect, but it is a different perspective, and having two imperfect perspectives is better than having one that you mistake for the whole picture.

Where Human Judgement Still Outperforms the Algorithm

There is a category error that happens when people talk about AI in advertising. They treat it as a replacement for strategic thinking rather than a tool that handles certain types of optimisation well. The distinction matters because it determines where you invest your team’s time.

I remember early in my agency career, being handed the whiteboard pen mid-brainstorm when the founder had to leave for a client meeting. The brief was for Guinness. My first thought was that this was going to be difficult. The second thought was to just get on with it. What that experience taught me, and what years of agency leadership reinforced, is that the quality of the thinking before the execution determines almost everything. No tool changes that.

AI cannot tell you whether your brand positioning is right for the market you are entering. It cannot tell you whether your pricing undermines your creative. It cannot tell you whether a short-term performance push will erode the brand equity you spent five years building. Those are judgement calls that require commercial context, category knowledge, and the kind of pattern recognition that comes from having seen similar situations play out before.

What AI can do is remove the mechanical friction from the parts of advertising that are genuinely mechanical. Testing creative combinations. Adjusting bids in real time. Identifying audience segments that manual analysis would miss. That is real value. But it is operational value, not strategic value.

The teams that get the most from AI-enhanced advertising are the ones that use the time it saves to do better strategic work, not the ones that use it as a reason to stop thinking. Sustainable growth still comes from understanding your market deeply, not from running a more automated version of the same campaign.

Practical Steps for Building an AI-Enhanced Advertising Approach

If you are looking to build AI more deliberately into your advertising strategy, the following sequence reflects what I have seen work across different business types and budget levels.

Start with your growth objective, not your technology stack. What does the business need advertising to do? Acquire new customers in a specific segment? Defend market share against a new entrant? Reactivate lapsed buyers? The answer shapes every other decision, including which AI capabilities are actually relevant.

Audit your conversion signal quality before you automate anything. If your conversion tracking is incomplete, your AI system will optimise toward an incomplete picture of your customers. Fix the data infrastructure first. This is unglamorous work, but it is the foundation everything else depends on.

Run incrementality tests early and regularly. Do not wait until you suspect a problem. Build holdout testing into your campaign architecture from the start so you have a baseline for what your advertising is actually driving versus what would have happened anyway.

Use AI for creative velocity, not creative strategy. Let the tools test combinations at scale. Keep the strategic creative decisions, the brand positioning, the tone, the key messages, in human hands. The algorithm will tell you which version of your idea performs better. It will not tell you whether the idea is right.

Review your campaign architecture for audience blending. Check whether your AI systems are quietly shifting budget from prospecting to retargeting. If they are, restructure before you optimise. The architecture problem will undermine any amount of algorithmic refinement.

There is a broader set of strategic decisions that sit above any individual advertising approach, including how you allocate budget across the funnel, how you sequence market entry, and how advertising connects to your wider commercial model. That context is covered in more depth across the Go-To-Market and Growth Strategy section of The Marketing Juice, which is worth reading alongside any specific channel or technology decisions.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions