Your attribution isn't broken. Your join keys are.

Buying a better attribution model to fix bad data is like buying a better camera to fix bad lighting.

Every marketing leader I know can tell you what their campaigns did. Almost none can tell you what their campaigns earned.

The reason is not a shortage of tools. Spend lives in the ad platforms. Engagement lives in LinkedIn and email. Revenue lives in a CRM that is usually half-clean. Three systems, three stories, and no reliable way to make them the same story. So budgets get defended with the metrics that are easy to count rather than the ones that matter, and marketing stays a cost center in the eyes of the people who sign the checks.

I have watched a lot of organizations respond to this by shopping for an attribution model. First-touch, last-touch, W-shaped, time-decay, data-driven. The debate gets sophisticated quickly and it is almost always the wrong debate.

The model is not where it breaks

Attribution projects fail because the data feeding the model is incomplete, inconsistent, overwritten, trapped in separate systems, or missing the timestamps required to reconstruct a sequence at all. Practitioners increasingly describe attribution as a data architecture problem rather than a reporting problem.

That framing matters, because the two problems have opposite remedies. If the model is wrong, you change the model. If the input is wrong, changing the model produces a differently shaped error rendered with the same confidence. Dashboards do not flag their own missing inputs. They just draw the chart.

There is a second, structural problem specific to B2B. Consumer attribution follows a person. B2B attribution tries to follow a committee, most of whose members never identify themselves at all. No model resolves an anonymous consensus into fractional credit. It only appears to.

The unglamorous layer

Underneath every attribution system sits a set of joins, and every join needs a key that actually matches. A campaign taxonomy applied consistently across every channel. UTMs that follow a convention rather than an individual’s habit. Deduplicated records, so one human being is one record. A real map from marketing activity to opportunity, with timestamps intact so velocity can be computed rather than guessed.

None of that is interesting. All of it is load-bearing.

And it degrades continuously. B2B contact data decays at roughly two to three percent per month, which is about thirty percent a year. A ten-thousand-record database sheds something like twenty-five hundred usable contacts annually. These figures come from vendors who sell data hygiene, and you should read them with that in mind. But the direction is not seriously disputed by anyone who has run a database, and neither is the consequence: an attribution layer built on records that rot is a layer that gets quietly less true every month while continuing to look identical.

Where healthcare makes it harder

Two complications land specifically on healthcare marketers.

The first is time. When your sales cycle runs eighteen months, the record that a touch occurred has to survive eighteen months of edits, migrations, ownership changes and enrichment passes to still be there when the deal closes. Most instrumentation is not built for that. It is built for a quarter.

The second is regulatory. Since 2022, and with revised guidance in 2024, the HHS Office for Civil Rights has been explicit that HIPAA-regulated entities may not deploy online tracking technologies in ways that disclose protected health information to vendors without a business associate agreement or individual authorization, including for marketing purposes.

Read plainly, that means some of the tracking your peers in other industries rely on is not available to you, and the gap has to be closed with first-party data and clean internal joins instead. Which makes the plumbing not merely important but the only route you have.

What I would do first

Not buy anything. Take one closed-won deal from last quarter and try to reconstruct its history end to end. Every touch, in order, with dates.

You will discover exactly where the chain breaks, and it will not be where you expected. Do it for five deals and you will have a prioritized repair list drawn from your own systems rather than from a vendor’s discovery call. Then establish your baseline honestly: what percentage of won deals can you currently trace at all? That number is your attribution coverage, and it is the only figure that tells you whether anything built on top is worth believing.

Fix that layer first. Everything smarter sits on top of it.

The question to bring to your next attribution conversation is not which model is most accurate.

It is what percentage of your closed business you can currently trace back to a first touch — and whether anyone in the room knows the answer.

Michael Carlson is the founder of Pertinacity, a growth agency that builds the AI and the systems underneath the marketing it runs.

Sources

• Openprise, Multi-touch attribution: the model isn’t the problem — https://www.openprisetech.com/blog/multi-touch-attribution
• Octane11, B2B marketing attribution: why it keeps failing — https://www.octane11.com/insights/insights-what-is-b2b-marketing-attribution/
• ZoomInfo, B2B data decay: rates, costs, and how to stop it — https://pipeline.zoominfo.com/marketing/b2b-data-decay
• Cognism, What is data decay — https://www.cognism.com/blog/data-decay
• HHS, Use of online tracking technologies by HIPAA covered entities — https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/hipaa-online-tracking/index.html
• Holland & Knight, OCR updates its website tracking tool guidance — https://www.hklaw.com/en/insights/publications/2024/03/ocr-updates-its-website-tracking-tool-guidance