Stop Hunting the Perfect Attribution Model, Start Building a Stack

Attribution has always been treated like a courtroom trial. Gather the evidence, put each channel on the stand, and let the judge decide which one gets the credit for the sale. The problem is that the trial now has forty witnesses, half of them are anonymous, and the judge can't see through the privacy glass. Dan Harris, writing for MarTech, argues the whole verdict-based framing needs to go.

My read: he's right, and the engineering teams building measurement pipelines have known it for a while. The marketers are finally catching up to what the data platform folks have been muttering about since the cookie started dying.

Key Details

The piece, published by MarTech on August 6, 2026, reframes attribution as a decision-support system. Its job isn't to declare a winner. It's to estimate which marketing and sales interactions are associated with revenue outcomes so teams can make sharper budget and pipeline calls. That reframing does more work than it looks like on paper.

Harris points at three shifts. The first is structural. B2B buying journeys, once described as six or seven touches, now commonly run 20 to 40 touches across multiple channels and multiple stakeholders. Last-click and first-touch still give directional value, but they can't narrate the whole story anymore.

The second shift is technical, and this is the guts of it. Signal loss is coming from every direction: browser privacy controls, cookie restrictions, ad blockers, AI-driven search, and identity fragmentation. Even properly implemented event tracking depends on recognizing a person across sessions, devices, and systems. That recognition problem is what's driven organizations toward server-side tracking, conversion APIs, identity stitching, CRM-connected measurement architectures, and first-party data strategies. Anyone who has stood up a server-side tag manager and then tried to reconcile it with Salesforce knows the boring bit is where it all falls over.

The third shift is organizational. Channel managers want tactical optimization. Marketing leaders want investment logic. Executives want confidence that spend is moving the business. No single model can serve all three audiences.

Harris then walks through the toolkit. Position-based attribution suits leadership narratives about demand creation and conversion. Time-decay fits long buying cycles where later engagement carries more weight. Data-driven attribution weights interactions based on patterns across historical opportunities, but it demands clean data and enough volume to be statistically meaningful. Marketing mix modeling operates at the aggregate level and is better suited for budget allocation across channels, particularly for events, sponsorships, and brand marketing that digital attribution routinely undercounts. Multi-touch attribution remains an effective tactical optimization tool. The strongest measurement stacks combine attribution, MMM, pipeline analytics, win-loss research, qualitative sales feedback, experimentation, and account engagement data.

Why This Matters for Performance Marketing

If you run paid acquisition for a B2B product, the courtroom analogy has been quietly poisoning your budget meetings for years. Someone in the room asks "what did LinkedIn drive last quarter?" and the answer depends entirely on which model the analyst opened first. Position-based says one thing, time-decay says another, and the CMO picks the one that flatters the plan they already wanted to fund.

The reason this matters for traffic teams is that the inputs to those models have degraded faster than most attribution vendors have updated their marketing decks. The Privacy Sandbox work, Apple's ITP, the Meta Conversions API era, all of it means the click-to-conversion chain is now stitched together from partial signals rather than observed end to end. If your entire budget defense rests on a multi-touch report generated from browser-side pixels, you're defending a bridge with half the cables missing.

Harris's argument that MMM complements rather than replaces attribution is the practical bit. MMM is aggregate, so it doesn't care that a third of your users blocked the pixel. It cares about spend curves and revenue curves and what happened when you turned the taps up or down. For anyone spending real money on events or brand, and getting told by their attribution dashboard that those channels drove almost nothing, MMM is the corrective lens. It won't tell you which creative won on Tuesday. It'll tell you whether the sponsorship deck was worth renewing.

The tactical optimization job, which creative, which keyword, which audience, still sits with multi-touch attribution and platform-native reporting like the Google Ads API and the Meta Marketing API. Use the right tool for the right question. Stop asking one dashboard to do both jobs.

Industry Impact

For engineering and data teams inside iGaming, fintech, and B2B SaaS, the operational fallout is real. A measurement stack of the kind Harris describes isn't a product you buy. It's a set of pipelines you own. Server-side tracking needs infrastructure, conversion APIs need reliable server-to-server delivery, identity stitching needs a canonical customer graph, and CRM-connected measurement needs the CRM to be the source of truth rather than a downstream reporting layer.

That's a lot of plumbing. Most of it lives with data engineering, not marketing ops. The teams that get this right treat the attribution layer as a product with SLAs, versioning, and clear ownership. The teams that get it wrong let the marketing agency spin up a Google Tag Manager container and then wonder why finance and marketing can't agree on last quarter's numbers.

There's a knock-on effect for vertical specifics. In iGaming, the 20 to 40 touch reality plays out across affiliate channels, retention email, push, and app reinstalls, most of which don't sit cleanly in a browser-based attribution model. In fintech, long sales cycles for B2B products push you firmly into time-decay territory. In crypto and DeFi, wallet-based identity fragments the picture further because the "user" often isn't a person you can stitch to a CRM record at all. Each vertical has its own signal-loss profile, and each needs a slightly different combination of the tools in Harris's stack.

The organizations that thrive here will treat measurement architecture as a first-class engineering concern, not a quarterly project the analytics team runs when someone complains about the numbers.

The Road Ahead

My prediction: within the next two years, the "attribution vendor" category as a standalone SaaS purchase will collapse into the broader customer data infrastructure space. The reason is straightforward. Once you accept that no single model wins, the value moves to whoever owns the clean, stitched, first-party dataset underneath. The model on top is almost commodity at that point.

Watch three signals. First, how quickly MMM tooling gets democratized. The classic knock on MMM was that it required consultants and months of work. Cheaper, faster MMM changes the math for mid-market teams. Second, how vendors handle the identity stitching problem as third-party cookies keep degrading and AI-driven search chews further into the discovery funnel. Third, whether CFOs start demanding measurement stacks that reconcile with finance data, because the moment attribution numbers have to tie back to booked revenue in the general ledger, half the vendor claims in the market stop surviving contact with reality.

Back to the courtroom. The verdict model is dead. What replaces it looks more like a panel of expert witnesses, each answering a specific question, with a judge who knows better than to ask any one of them to solve the whole case.

Key Takeaways

• B2B buying journeys have expanded from 6 to 7 touches to 20 to 40 touches, and no single attribution model can narrate that story.
• Signal loss from browser controls, cookies, ad blockers, AI search, and identity fragmentation has pushed measurement toward server-side tracking, conversion APIs, and first-party data.
• Match the model to the question: position-based for leadership narratives, time-decay for long cycles, data-driven when volume and data quality allow, MMM for budget allocation and brand.
• MMM complements attribution rather than replacing it, particularly for events, sponsorships, and brand marketing that digital attribution undercounts.
• The strongest measurement stacks combine attribution, MMM, pipeline analytics, win-loss research, qualitative sales feedback, experimentation, and account engagement data.