Data-Driven Marketing: What It Gets Right and Where It Fails

6 min read · 1,241 words

Data-driven marketing became the default posture in most B2B marketing functions over the last decade, and it’s mostly made marketing better. It’s also created a specific, predictable failure mode that the phrase itself never warns you about: optimising hard for what’s measurable while quietly starving what isn’t, until the whole function is excellent at proving short-term wins and unable to explain why long-term brand health is flat or declining.

What data-driven marketing actually got right

The shift away from gut-feel budget decisions toward measurable, testable marketing has been a genuine improvement. Attribution modelling, even the imperfect versions most teams run, forced a level of accountability that didn’t exist when campaigns were approved on the strength of a good creative pitch alone. Marketing teams that adopted rigorous testing discipline — proper A/B tests, held-out control groups, incrementality testing rather than correlation — generally allocate budget better than teams that don’t. That part of the shift is real and worth keeping. It’s the same discipline I’ve argued for in the CX score trap: what NPS never tells you — a single tidy metric is comforting precisely because it’s simple, and simplicity is not the same thing as accuracy. The uncomfortable truth in both cases is that the metric everyone already trusts is usually measuring something narrower than the decision it’s being used to justify.

Where it goes wrong: the over-attribution trap

The specific failure mode is over-attribution to last-click and other easily measurable, near-term signals, at the expense of activity whose payoff arrives months or quarters later and resists clean attribution. A prospect who converts after seeing a brand campaign eight months earlier, three retargeting ads, and a sales call gets attributed entirely to whichever touchpoint the model can see clearly — usually the last one. Forrester’s research on B2B campaign performance has repeatedly found that high-performing campaigns work across the full buyer journey, not just the measurable final step — but budget allocation models built purely on last-click data systematically under-fund everything except that final step.

This is the mechanism by which “data-driven” marketing quietly becomes short-term marketing. Nobody decides to defund brand-building; the attribution model simply can’t see brand-building’s contribution, so it looks like waste next to a channel that can show a clean, immediate conversion number. Binet and Field’s long-run effectiveness research — the same body of work behind the 60:40 brand-to-activation guidance — has found precisely this pattern across hundreds of case studies: over-indexing on short-term, measurable activation produces short-term results that decay, while businesses protecting brand investment sustain effectiveness for years longer.

What good data-driven marketing looks like instead

The fix isn’t abandoning data discipline — it’s being honest about what your measurement stack can and can’t see, and deliberately protecting investment in the things it can’t. That means running actual incrementality tests (geo holdouts, matched-market tests) rather than relying purely on attribution models for anything meant to build long-term brand equity, and reporting brand-building spend against leading indicators — category entry point coverage, branded search volume, share of voice — rather than forcing it into the same last-click framework as a paid search campaign, where it will always look like it’s losing. Google’s own push into open-source marketing mix modelling with Meridian is itself a signal that even the platforms most associated with last-click measurement now acknowledge attribution alone understates what long-run channels like brand actually contribute.

A 2026 addition to the original argument: AI has made this worse, not better

Generative AI tools have made it cheaper and faster to produce content optimised for exactly the measurable, short-term channels attribution models already over-reward — more paid social variants, more retargeting creative, more landing page tests. Gartner’s CMO Spend Survey research has flagged marketing budgets under sustained pressure through this period, which only sharpens the incentive to chase the channels with the cleanest, fastest-to-produce attribution story. That’s genuine value, but it compounds the imbalance this piece is describing unless a team is deliberately protecting brand-building budget as a separate, protected line rather than letting it compete quarter to quarter against channels with cleaner, faster-to-produce attribution stories.

The practical fix isn’t a policy memo — it’s a governance change. Ring-fence a specific percentage of budget for brand-building activity at the start of the planning cycle, before any quarter-by-quarter reallocation conversation happens, and require a genuine incrementality test before that ring-fenced spend can be cut, not just a weaker attribution number. Most of the erosion I’ve watched happen didn’t come from a deliberate decision to defund brand — it came from nobody having pre-committed to protecting it before the pressure showed up.

What this means for you

Audit your last two quarters of budget reallocation decisions and ask specifically what got cut and why. If everything that got cut was brand-building, awareness, or anything with a measurement lag longer than a month, that’s the over-attribution trap operating in your own function, not a genuine data-driven insight about what’s working. I’ve written about the specific budget-defence mechanics for protecting that spend in how much a B2B company should spend on brand. LinkedIn’s B2B Institute has published a related argument for treating brand and demand generation as complementary rather than competing line items in the same budget review — worth reading alongside this if the reallocation conversation is a recurring fight in your organisation.

Frequently asked questions

What is the biggest risk of purely data-driven marketing?

Over-attribution to easily measurable, short-term touchpoints (usually last-click), which systematically defunds brand-building and other long-lag activity that a measurement model can’t cleanly observe, even when that activity is working.

How can a marketing team measure brand-building if attribution can’t capture it?

Incrementality testing (geo or matched-market holdouts), marketing mix modelling, and leading indicators like category entry point coverage and branded search volume — reported separately from last-click attribution, not forced into the same framework.

Has AI made the attribution problem better or worse?

Worse, in practice. AI makes it cheaper to produce more content for the channels attribution already over-rewards, compounding the imbalance unless brand-building budget is deliberately protected as a separate line.

Should marketing teams stop using attribution models entirely?

No — they’re genuinely useful for optimising the channels they can observe well. The fix is recognising their blind spots and using a different measurement approach specifically for brand-building rather than forcing everything into one model.

What’s a practical first step to fix over-attribution in a marketing function?

Run one incrementality test (a geo holdout is the simplest version) on a channel currently judged only by attribution. Comparing the incrementality result against the attributed result usually reveals the gap immediately.

Where has your own team’s attribution model quietly starved something that was actually working? I’d like to hear what the incrementality test revealed when you ran it.

Related reading: winning the internal argument for brand.