Attribution is broken almost everywhere. Here's what we actually do about it

Every single attribution model available today is meaningfully wrong in some specific, identifiable way, and any agency claiming otherwise to a client is either not being fully honest with them, or genuinely hasn't looked closely enough at their own methodology yet. Last-click attribution systematically overweights whatever the final touchpoint happened to be. First-click attribution largely ignores everything that actually contributed to closing the eventual sale. The genuinely useful conversation here isn't about finding some mythical perfect model. It's about deliberately picking one that's honest and upfront about its own specific blind spots.

For a business with a genuinely short sales cycle and relatively few real touchpoints along the way, a single ad click leading fairly directly to a purchase, last-click attribution actually paints a reasonably honest overall picture of what's happening. The real problem arises when applying that same last-click logic to a business running a six-month, genuinely multi-channel B2B sales cycle, where it systematically and unfairly undervalues every single channel except whichever one happened to be clicked last in the sequence.

Data-driven or position-based multi-touch attribution models spread credit more realistically across the actual full customer journey, which is genuinely closer to reality for complex, considered sales processes. They also require a meaningful volume of conversions to be statistically reliable at all, and a lot of smaller clients simply don't generate enough raw conversion data yet for a multi-touch model to actually be more accurate than pure statistical noise.

Deliberately turning a specific channel completely off for a controlled period of time, then carefully measuring the actual real-world change in overall conversions across the business, answers a genuinely different and often considerably more useful question than any attribution model ever can: is this particular channel actually driving real, incremental results for the business, or is it simply claiming credit for conversions that would have happened anyway through some other channel regardless.

No single dashboard number represents the whole truth, and we say so directly and repeatedly. We pick the most reasonable available model for a given client's specific sales cycle and channel mix, and we're consistently upfront that it's meant to be directionally useful for decision-making, not a precise, exact accounting ledger. Clients who genuinely understand this limitation upfront consistently make noticeably better strategic decisions than ones who've been quietly sold on a false sense of precision.

We reassess the attribution approach for every client roughly twice a year, since a business's sales cycle and channel mix can shift meaningfully over that timeframe, and a model that made sense a year ago may no longer fit. This isn't a set-once decision. Treating it as one is itself a common mistake we see in accounts we inherit from other agencies.

iOS's App Tracking Transparency framework and the broader industry shift away from third-party cookies have made cross-device and cross-platform attribution meaningfully harder over the past couple of years, with a genuinely growing share of conversions now landing in a modeled or estimated bucket rather than being directly, deterministically observed and tracked. We explain this shift to clients honestly rather than letting an increasingly incomplete dashboard number pass by silently as if nothing had changed underneath it.

Where direct tracking has gotten measurably less reliable, we lean more heavily on aggregated platform-reported conversions and periodic incrementality testing to cross-check the picture, rather than trusting a single increasingly incomplete pixel-based number as the full and final word on what's actually working for that client's business.

One retail client's dashboard consistently showed a specific retargeting campaign as their top-performing channel by attributed conversions, month after month, which made it feel like an obvious place to keep increasing budget. Running a genuine geo-based incrementality test, turning the campaign off entirely in a matched set of comparable markets for three weeks, showed almost no measurable drop in overall conversions in those markets compared to the markets where it stayed active.

The retargeting campaign had mostly been claiming credit for purchases that would have happened anyway through organic and direct traffic, not genuinely driving new incremental ones. The client reallocated a meaningful share of that budget toward top-of-funnel channels the dashboard had been quietly undervaluing, and overall conversions rose within the following quarter, a result the original last-click dashboard view would never have surfaced on its own.

There's a real balance to strike here: being honest about a model's limitations without leaving a client feeling like every number in their report is essentially meaningless or arbitrary. We frame it as a range and a direction rather than a false, false-precision single figure wherever we reasonably can, this channel is very likely contributing meaningfully, this one is more uncertain and worth a dedicated incrementality test, rather than presenting every number with identical, unearned confidence regardless of how reliable it actually is.

Clients who've been through this more honest framing consistently tell us they trust our reporting more, not less, once they understand the real uncertainty behind the numbers, because it signals we're not simply telling them whatever story keeps a specific channel's budget renewed each month.

Moving pixel-based tracking to a server-side implementation, where the client's own server relays conversion events directly to the ad platform rather than relying entirely on a browser-based pixel that ad blockers and browser privacy settings can interfere with, has meaningfully improved data completeness for several client accounts over the past year. It's a real technical setup investment upfront, and it consistently pays for itself in more reliable, more complete conversion data feeding whatever attribution model we've chosen for that account.

We now recommend server-side tracking as close to a default for any client running a meaningful paid budget, treating browser-only pixel tracking as the fallback option for smaller accounts where the additional setup cost genuinely isn't justified by the budget being spent.