Marketing Evolution: Why Fintech ROI Starts With Data

Stephen Williams sits at the intersection of two forces reshaping financial services: the rapid acceleration of AI and the growing complexity of data-driven marketing in regulated environments.

As CEO of Marketing Evolution, his path from engineer to CEO reflects a career grounded in building the systems behind modern marketing – experience that now informs how fintechs can better understand performance in an increasingly fragmented landscape.

At its core, Marketing Evolution is focused on helping organisations turn marketing data into actionable intelligence. The company's Marketing Performance Substrate unifies fragmented data sources into a single foundation for marketing intelligence. It reconstructs data gaps, embeds causal intelligence and continuously updates performance data to create a trusted system of record that both human teams and AI agents can reason and act on.

In a market where AI models are becoming more commoditised, Stephen argues that the real differentiator for fintechs is not the sophistication of the models themselves, but the strength of the underlying data infrastructure that powers them. That distinction is particularly critical in fintech, where privacy, governance and compliance requirements place strict limits on how data can be used. At the same time, customer journeys – from opening accounts to securing credit – are rarely linear, often spanning multiple channels, stakeholders and touchpoints over extended periods.

By anonymising data and reconstructing these journeys probabilistically, Marketing Evolution aims to give financial institutions a clearer, compliant view of how marketing drives outcomes – bringing much-needed transparency to one of the sector’s most complex measurement challenges.

Here, Stephen explains why data – not models – is becoming the defining factor in AI success and how fintechs can close the gap between ambition and execution.

If marketing is moving faster than infrastructure, how does this tension hit CMOs in regulated financial services? 

Let’s start with the broader reality facing organisations today: across every industry and every role, people are being told they need to use AI to work smarter, lower costs and find new sources of value. But for CMOs, the pressure is particularly intense because the role has evolved significantly over the years.

Today, CMOs are accountable not just for brand, but for revenue and performance. At the same time, they’re being asked to adopt AI quickly, prove ROI and make decisions across increasingly fragmented systems and data.

In regulated financial services, that challenge becomes even harder because of long customer journeys, multiple stakeholders, strict compliance requirements and significant data constraints.

Pulling that data together into a unified environment that properly reflects the business is what ultimately enables meaningful AI use cases. Once that foundation exists, organisations can layer models from OpenAI, Anthropic, Google or others on top of the data and begin extracting value from information they already have.

I think of it as every business – especially in financial services – is sitting on a mountain of gold. That gold is their data. But before it can create value, it has to be mined and refined. Or, in this case, structured and contextualised. 

The other thing happening right now is that many organisations believe the greater risk is being late to AI rather than being wrong. That’s creating enormous pressure to move quickly, even when the underlying data foundations aren’t fully in place.

AI may provide some value initially, but organisations will see diminishing returns if the infrastructure isn’t there to support it.

What’s the 'intermediary gap' costing marketers in banking and insurance attribution?

There’s significant cost, but it’s recoverable. At the end of the day, marketers optimise what they can measure. Often, that means they over invest in the wrong things and under invest in what they can’t properly attribute – especially upper-funnel and brand activity that happens well before conversion.

That challenge becomes especially pronounced in financial services because conversion is often owned by intermediaries: brokers, branches, advisors, platforms or other downstream stakeholders. Marketing may be driving demand, but the transaction is frequently captured elsewhere in the system.

As a result, attribution tends to credit the closer rather than the creator of demand. It’s an age-old measurement problem, but it becomes amplified in industries where customer journeys are longer, more fragmented and distributed across stakeholders.

To solve for this, that information must be pulled together holistically, into a unified data set that properly reflects the business. Once you can reconstruct those customer journeys, marketers can begin to understand how brand investment, media exposure and intermediary relationships influence downstream conversion outcomes.

For example, we work with one of the largest property insurers in the US. They invest heavily at the national brand level and have been very successful doing so. Historically, though, they struggled to connect that investment to performance at the policy-buy level, where local agents are also spending their own marketing dollars and ultimately closing the business. 

The agent relationship absolutely matters – it’s a critical part of the conversion process. But what we’ve helped them do is surface the impact of national brand spend on those downstream conversions.

So when they invest in something like a major sports sponsorship or national advertising campaign, they can now see how those dollars ultimately flow through the system and contribute to measurable conversions happening at the local agent level – whether that’s in Memphis, Tennessee or anywhere else in the country.

Why does AI amplify bad data rather than fix it in marketing stacks?

This isn’t isolated to marketing stacks. AI is going to amplify bad data everywhere.

No matter how sentient AI may appear, it still fundamentally depends on the quality and structure of the underlying data to inform its analysis. And marketing data, specifically, is very, very bad data.

One major reason is that marketing data lives across multiple different providers. You don’t have one source of marketing data. You have data coming from Google Analytics, Meta, Nielsen, Lamar out-of-home and dozens of other places. And all of that data looks different.

The data has different levels of granularity and different levels of completeness. Then, all of it has to be joined with first-party data, which also looks different for every advertiser – email data, direct mail data, conversions data, web traffic data and so on.

So when we look at marketing data as a whole, we don’t see one giant data set that’s ready to go. We see dozens of fragmented data sets that are often incomplete and sparse in places. Some are at the national level, some at the ZIP code level and some at the DMA level.

Historically, that has forced data engineers to spend enormous amounts of time building systems, writing queries or creating Python scripts to process data on an ad hoc basis for whatever downstream analysis is needed.

But if you can fit that data into a uniform structure, erase those issues of granularity and unify across the silos, the picture changes completely.

We can bust through the silos. We can break down the clean room walls and bring the data into one location where we can see Amazon Ads next to Google next to Meta, right alongside channels like CTV, Linear TV and OOH, all in the same language and with the same structure. All of a sudden, you’re comparing apples to apples instead of apples to oranges and watermelons.

We have systems of record for customers in CDPs, for content in DAMs and for execution in MAPs, but there is no equivalent for marketing performance. That’s what we’ve stood up at Marketing Evolution.

The data is there, the tools are there, the analysis is there and the AI is there to provide an accurate picture of the market and the business’s universe, instead of amplifying misleading signals.

At the end of the day, we’ve got to maintain data centricity in how we model, analyse and use AI. The models are going to keep coming and going. They’re going to keep changing. They’re going to move fast.

This is ultimately a data infrastructure problem. It’s not a model problem.

It’s about getting the right data infrastructure in place, getting the right data into that infrastructure and contextualising it to match the business and how the business actually operates off that data.

Ultimately, if you get the data layer wrong, the CMO is locked into three years of bad answers and the CFO is looking at three years of unaccountable spend. It’s critical to get that data layer right as quickly as possible.

With 71% thinking they’re AI ready and only 37% truly are – what’s the real readiness gap?

I think many leaders – and honestly I’ve been guilty of this myself – apply a very simple proxy: if AI tools are in use, the organisation must be AI-ready. And that’s rarely the case.

When you actually dig into readiness and look at the conditions required to successfully deploy AI or an agentic workforce – things like data structure, cleanliness, integration, accessibility and governance – that assumption breaks down very quickly.

We conducted research to better understand that gap. We surveyed senior marketing leaders – CMOs, VPs, Directors and Senior Managers – and found that 71% believe they are AI-ready.

But when we evaluated whether the foundational data conditions required for AI readiness were actually in place, only 37% met the standard.

That’s a massive gap. And, even more striking, only 3% reported seeing consistent AI performance gains.

The reality is that many organisations hit a ceiling with AI very quickly because the underlying data foundation isn’t there. If AI is constantly trying to reconcile fragmented systems, disconnected sources and inconsistent structures, the complexity of the problem expands dramatically. That leads to weaker outputs, incorrect conclusions and more frequent hallucinations.

Organisations that meet the foundational data conditions for AI readiness are seeing roughly two times stronger AI performance outcomes than organisations that don’t. It’s critical to get the data layer right in order to extract real value from AI – especially as AI becomes more expensive and organisations lose the luxury of experimenting indefinitely at low cost.

How does a marketing performance substrate solve data unification for AI?

A Marketing Performance Substrate—or a system of record for marketing performance—solves this problem because the layer itself is intelligent. It’s able to recover missing data, synthetically filling observational gaps and reconstruct consumer-level journeys that simply wouldn’t be possible if the data were just sitting in a static warehouse or data lake somewhere. In that sense, it’s extracting meaning from the data before it ever enters more complex downstream models.

But even before that, the data has been fit to an ontology that reflects the business. It’s mapped to the right schema, taxonomy and business context so that the naming conventions match how the organisation actually talks about its creatives, media channels, campaigns and customers internally. That consistency becomes uniform across all of the underlying data sets.

By unifying both the structure of the data and the consumer journeys, we open up entirely new AI use cases for discovery, analysis and decision-making.  

And I don’t mean that in a hyperbolic way – the opportunity becomes near limitless once AI is operating against a truly unified and contextualised understanding of the business.

How would you advise a shift from backward looking measurement to forward looking proactive intelligence happens? 

Look, it’s tough, and it does require a real perspective shift. Until very recently, marketing measurement has been retrospective. We would execute campaigns, wait for the results and then analyse what happened. But now we have the data and the tools to simulate what we’re planning to do before we actually execute the plan.

That changes the role of marketing intelligence entirely. Instead of simply measuring outcomes after the fact, organisations can begin modeling likely outcomes in advance, using prior months, quarters and years of data to inform what is likely to happen next. That allows marketers to mitigate risk, make faster and smarter decisions and defend those decisions because they’re grounded in data.

It also creates much more complete full-funnel visibility. One way I think about this is through the way human vision actually works. Most people assume that what we see is simply the visual cortex processing light coming through the eyes. But in reality, the brain is constantly running a predictive model of what it expects to see.

That model – built over a lifetime of structured sensory data – is continuously predicting the world around us. The incoming visual information updates the model, but much of what we perceive is actually the brain anticipating what is about to happen.

That’s how we catch a baseball or react to a car braking suddenly in front of us. We’re not simply reacting to raw sensory input in real time. We’re operating against a predictive model informed by prior data and experience.

I think marketing is moving in a very similar direction.

Organisations with strong data foundations will be capable not just of explaining the past, but of predicting and simulating what comes next.

How would you advise a brand that says ‘I’m working with an MMM partner. Where do I go from here?’

If you’re already working with an MMM partner, that’s a great starting point because you already have a historical understanding of media performance. That knowledge doesn’t go away. In fact, it can be brought forward and applied to a more predictive intelligence layer. Your MMM can remain in place and continue informing the system as you move toward more adaptive decision-making.

But the next step does require a commitment to upgrading the underlying data infrastructure and moving toward more granular, integrated data.

It also means evolving toward measurement methodologies and models that can react faster to changing market conditions, shifting media spend and changing consumer behavior than traditional MMMs typically can.

If you think about something like COVID, we all lived through a period where consumer behaviour was changing dramatically month to month. MMMs eventually picked up on those shifts, but often after the fact.

A more dynamic intelligence layer – one operating on a weekly or even daily cadence – can move with the business in real time. It can identify those shifts as they’re happening and help organisations respond faster and with more confidence.

That’s really the broader shift happening right now. Historically, the martech stack has been made up of systems like CDPs, CRMs, DAMs, MAPs and measurement solutions – all serving different operational functions. Increasingly, though, organisations need an intelligence layer sitting across those systems that can make the data usable for AI, simulation, optimisation and decision-making. That’s where the industry is heading. 

How should brands be thinking about building their martech stack today?

In my mind, the martech stack today needs to start with understanding what data sources and systems you already have – whether that’s a CDP, a CRM system, DSPs or anything else relevant to your brand – and making sure those systems can work together rather than operate independently.

Traditionally, we’ve thought about the martech stack as a layered stack. Increasingly, though, I think it looks more like a spoke-and-wheel model, where you have a central data and intelligence layer feeding decision-making across the business.

That hub interacts in both directions with systems like the CDP. It’s taking data out of the CDP, but it’s also feeding information back into it to help better understand customer behaviour and what’s happening across the broader ecosystem.

As far as measurement goes, I think we’re now in a world where measurement can move as fast as the data itself. We don’t have to rely only on MMMs anymore. MMMs still have a place and can be useful when looking across longer time horizons, but marketers need to make decisions rapidly – sometimes even daily – and that’s where a more nimble measurement system becomes important.

So for me, the modern martech stack is centred around a unified data layer capable of supporting rapid measurement, optimisation, simulation and decision-making, while still interacting cleanly with the surrounding systems across the business.