Data-Driven Advertising: What the Numbers Still Can’t Tell You

Data-driven advertising means using audience data, behavioural signals, and performance metrics to inform where you spend, who you target, and what you say. Done well, it reduces waste and sharpens decision-making. Done poorly, it becomes a sophisticated way of optimising toward the wrong thing.

The promise is real. The execution is where most advertisers come unstuck, not because the data is bad, but because the questions being asked of it are wrong.

Why Most Advertisers Are Optimising the Wrong Thing

I spent several years managing large performance budgets across retail, financial services, and telecoms. One thing I noticed consistently: the better the attribution dashboard looked, the more nervous I became. Because clean attribution usually means you are measuring what is easy to measure, not what is actually driving growth.

Performance marketing, at its most data-driven, is extraordinarily good at capturing demand that already exists. Search campaigns intercept people who are already in-market. Retargeting reaches people who already visited your site. Dynamic product ads serve people who already added something to their cart. These are valuable activities. But they are not the same as creating demand, and when you over-index on them, you hollow out the top of the funnel without realising it until the numbers start declining 18 months later.

The data tells you these channels are working. The ROAS looks great. The CPA is efficient. What the data does not tell you is that the pool of people searching for your brand is shrinking because you stopped investing in brand-building three years ago. That is the measurement gap that trips up otherwise competent marketers.

Data-driven advertising is a tool for making better decisions within a strategy. It is not a substitute for having a strategy. That distinction matters more than any dashboard feature your ad platform is about to release.

If you want a broader framework for where advertising fits within commercial growth, the Go-To-Market and Growth Strategy hub covers the full picture, from positioning and channel selection through to measurement and scaling.

What Data-Driven Advertising Actually Involves

The term gets used loosely, so it is worth being precise. Data-driven advertising covers several distinct activities that are often conflated.

Audience targeting uses data signals, demographic, behavioural, contextual, or intent-based, to define who sees your ads. This ranges from basic demographic targeting on social platforms to sophisticated lookalike modelling built on first-party CRM data.

Creative optimisation uses performance data to determine which messages, formats, and visuals resonate with which audiences. This includes multivariate testing, dynamic creative optimisation, and the increasingly automated creative selection that platforms like Meta and Google now handle algorithmically.

Budget allocation uses channel performance data to decide where to concentrate spend. This is where attribution models become critical, and where most of the measurement problems live.

Bid management uses real-time data to adjust how much you pay for each impression or click, either manually or through automated bidding strategies. The platforms are now very good at this, which is both useful and a reason to be cautious about ceding too much control.

Each of these activities involves data. But they involve different types of data, different time horizons, and different risk profiles. Treating them as a single unified discipline is where the confusion starts.

The Attribution Problem Nobody Has Solved

When I was judging at the Effie Awards, one of the things that struck me about the strongest entries was how honest they were about measurement limitations. The campaigns that won were not the ones with the cleanest attribution. They were the ones where the teams had thought carefully about what they were trying to prove and had built measurement approaches that were genuinely fit for purpose, rather than just technically sophisticated.

Attribution is the central unsolved problem in data-driven advertising. Every model makes assumptions, and those assumptions systematically favour certain channels over others. Last-click attribution, which still dominates many accounts despite years of criticism, gives all credit to the final touchpoint before conversion. This makes search look like a genius and makes every channel that contributed earlier in the experience look like a passenger.

Data-driven attribution, the model that Google now defaults to, distributes credit across touchpoints based on their observed contribution to conversion. It is more sophisticated. It is also a black box, and it is built by a company that has a commercial interest in the model favouring Google-owned inventory.

Marketing mix modelling offers a more strong alternative for understanding the actual contribution of different channels, but it requires significant data volume, takes time to produce results, and is expensive to run properly. Most mid-market advertisers cannot justify the cost. They are left using platform-native attribution, knowing it is imperfect, and trying to triangulate with other signals.

The honest position is that no advertiser has a complete picture of how their media works. The goal is not perfect measurement. It is honest approximation, combined with enough intellectual humility to question what the numbers appear to be saying.

First-Party Data: The Actual Competitive Advantage

The deprecation of third-party cookies has been discussed so extensively that it has become background noise. But the underlying shift is real and it matters. Advertisers who built their targeting strategies on third-party data are now in a structurally weaker position than those who invested in first-party data infrastructure.

First-party data, the information you collect directly from customers and prospects through your own channels, is more accurate, more durable, and more strategically valuable than anything you can buy from a data broker. It is also harder to build. You need people to give you their information willingly, which means you need to offer something worth giving it for.

The advertisers doing this well are not treating data collection as a technical exercise. They are treating it as a value exchange. What can we offer that makes it worth someone sharing their email address, their preferences, their purchase history? That question sits at the intersection of product, marketing, and commercial strategy, which is why it rarely gets the attention it deserves in conversations that are dominated by platform tactics.

When I was growing the agency at iProspect, we pushed hard on this with clients in retail and financial services. The ones who had invested in CRM data and were willing to bring it into their media buying had a measurable edge on audience quality. The ones who were still relying on platform audiences were competing on the same terms as everyone else, which in practice means competing on budget.

Tools like those covered in Semrush’s breakdown of growth hacking tools are useful for understanding the broader ecosystem of data and analytics platforms. But the tool is never the strategy. It is what you do with the underlying data that determines whether any of this compounds into advantage.

How Algorithmic Buying Changed the Game (and What It Missed)

Programmatic advertising was supposed to make media buying more efficient by using data to match the right ad to the right person at the right moment. In many respects, it delivered on that promise. Targeting precision improved. Waste reduced. Buying became faster and more scalable.

What it also did was commoditise execution. When every advertiser is using the same platforms, the same audience segments, and the same bidding algorithms, the differentiation disappears. You end up in an auction with everyone else, paying market rate for attention that is increasingly fragmented and increasingly expensive.

The platforms are also optimising for their own objectives, which are not always aligned with yours. When you hand budget to a smart bidding strategy and tell it to maximise conversions, it will do exactly that. Whether those conversions represent genuine incremental business growth is a different question, one the algorithm is not equipped to answer.

I have seen this play out in accounts managing tens of millions in annual spend. The automated bidding looks clean on the surface. Conversion volume is up, CPA is within target, the platform dashboard is full of green numbers. Then you run an incrementality test and discover that a meaningful proportion of those conversions would have happened anyway. The algorithm was not creating demand. It was efficiently intercepting people who were already going to convert, and charging you for the privilege.

This is not an argument against automated bidding. It is an argument for understanding what it is actually doing, and building measurement approaches that can distinguish between efficiency and incrementality.

Segmentation That Works vs. Segmentation That Feels Sophisticated

Audience segmentation is one of the most discussed topics in data-driven advertising and one of the most frequently over-engineered. The temptation, especially when you have access to rich data sets, is to build increasingly granular segments on the assumption that more specificity equals more relevance.

Sometimes it does. Often, it creates segments that are too small to be statistically meaningful, too narrow to allow the algorithm to learn effectively, and too complex to manage without introducing errors.

The most effective segmentation I have seen in practice tends to be grounded in commercial logic rather than data availability. Not “what segments can we build from this data” but “what differences in customer behaviour or value justify treating people differently.” That is a much harder question to answer, and it requires collaboration between marketing, commercial, and product teams rather than a single analyst working in isolation.

The BCG framework on commercial transformation is worth reading in this context. It makes the point that commercial growth requires alignment across functions, not just better tools within marketing. Segmentation that is not connected to commercial reality, pricing, product, distribution, tends to produce targeting that is technically impressive and commercially inert.

Creative Is Still the Variable That Matters Most

There is a version of data-driven advertising that treats creative as a variable to be tested into oblivion. Run enough variants, measure enough signals, let the algorithm select the winner. The problem is that this approach optimises for short-term engagement metrics, click-through rate, conversion rate, video completion, rather than the harder-to-measure qualities that make advertising work over time.

Early in my career, I was handed a whiteboard pen at Cybercom during a Guinness brainstorm when the founder had to leave for a client meeting. The internal reaction, mine included, was somewhere between panic and determination. What that moment taught me is that creative thinking under pressure requires a point of view, not just a process. You cannot data-test your way to a point of view. You have to develop one, and then use data to pressure-test it.

The best data-driven creative work I have seen starts with a clear human insight, develops a genuine creative idea from it, and then uses data to optimise execution. Format, length, copy variations, call-to-action phrasing. These are legitimate things to test. The underlying idea is not a variable. If you are testing the idea, you do not have one.

Dynamic creative optimisation has genuine value when it is used to personalise execution for different audiences. Showing different product imagery to different segments, adjusting copy to reflect where someone is in the purchase experience. But it requires a coherent creative framework to work within, or you end up with a matrix of disconnected assets that share no consistent brand voice.

The Measurement Stack: What You Actually Need

Every few years, a new measurement technology arrives that promises to solve the attribution problem. Multi-touch attribution replaced last-click. Data-driven attribution replaced multi-touch. Marketing mix modelling is having a renaissance. Incrementality testing is increasingly accessible. Clean rooms are the current frontier.

None of these tools is wrong. All of them are partial. The mistake is treating any single measurement approach as definitive rather than as one perspective on a complex reality.

A practical measurement stack for most advertisers involves three things. Platform-native reporting for day-to-day optimisation decisions, with the understanding that it overstates platform contribution. Regular incrementality testing, even simple geo-based holdout tests, to calibrate how much of your conversion volume is genuinely incremental. And some form of longer-term brand tracking to monitor whether your advertising is building the mental availability that drives future demand.

The CrazyEgg breakdown of growth hacking makes a useful point about the relationship between experimentation and measurement. You cannot run meaningful experiments without a clear hypothesis and a measurement approach that can actually detect the effect you are looking for. Most advertising experiments fail on both counts.

The goal is not a perfect measurement stack. It is a coherent one, where you understand what each data source is telling you, what its limitations are, and how to weight it against the others.

Where Data-Driven Advertising Fits in a Growth Strategy

Data-driven advertising is not a growth strategy. It is a set of capabilities that can support one. The distinction matters because too many businesses treat media sophistication as a substitute for strategic clarity, and it rarely is.

I have worked with businesses that had genuinely impressive data infrastructure, clean CRM integration, sophisticated audience modelling, programmatic buying across a dozen channels, and were still not growing because the underlying commercial proposition was not compelling enough to justify the spend. More precise targeting of a weak message is still a weak message.

Growth comes from a combination of factors: a product or service that people actually want, a clear reason to choose it over alternatives, distribution that reaches the right people at the right moment, and communication that makes the value proposition legible. Data-driven advertising contributes to the last two. It cannot compensate for weakness in the first two.

The Semrush collection of growth hacking examples illustrates this well. The cases that hold up over time are ones where the data capabilities were in service of a genuine commercial insight, not ones where the data sophistication was the story in itself.

If you are thinking about how advertising fits within a broader commercial growth framework, the articles in the Go-To-Market and Growth Strategy section cover channel strategy, market entry, and how to build media investment decisions that connect to business outcomes rather than just media metrics.

The Questions Worth Asking Before You Spend Another Pound

After 20 years of managing advertising budgets across 30 industries, the questions I find most useful are not technical ones. They are strategic ones that the data alone cannot answer.

Are you measuring what you are optimising toward, or are you optimising toward what you can measure? These are different things, and the gap between them is where most advertising efficiency goes to die.

Is your advertising creating demand or capturing it? Both are valuable, but they require different channel mixes, different time horizons, and different measurement approaches. If you cannot answer this question, you probably have an imbalanced portfolio.

What would you do differently if your attribution model showed different results? If the answer is nothing, your strategy is not actually data-driven. It is data-decorated.

How much of your conversion volume is genuinely incremental? If you have not run an incrementality test in the past 12 months, you do not know the answer, and you should probably find out before you scale spend further.

What is the data not telling you? This is the most important question and the one least often asked. Data tells you what happened. It rarely tells you why. It tells you who converted. It does not tell you who decided not to, or why. Filling those gaps requires qualitative research, commercial intuition, and the willingness to sit with uncertainty rather than reaching for the nearest metric.

The Hotjar growth loop framework is a useful lens here. Understanding how users actually behave, rather than how your analytics platform records their behaviour, often reveals gaps between what the data shows and what is actually happening in the market.

Data-driven advertising, done with clear thinking and honest measurement, is one of the most powerful commercial tools available to a marketing team. The problem is not the data. It is the tendency to let the availability of data substitute for the harder work of thinking clearly about what you are trying to achieve and whether your advertising is actually helping you get there.

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