In forensic investigation, evidence isn't automatically trusted just because it's true. It has to survive an unbroken, documented chain from the point it was collected to the point it's presented, who handled it, when, and how it was stored at every step. A single undocumented gap, even a short one, is often enough for the evidence to be challenged and excluded, regardless of how solid the rest of the chain was.
Marketing attribution data runs through a strikingly similar chain: an ad platform, a landing page, an analytics tool, a CRM, and finally a reporting dashboard someone presents to the board. Most teams don't audit that chain the way a forensic team would audit theirs, which is exactly why attribution numbers quietly drift from reality without anyone noticing where it happened.
What chain of custody means, and why forensic teams treat it as absolute
The principle is simple and deliberately strict: every person who touches a piece of evidence has to document when they received it, what condition it was in, and who they passed it to next. The point isn't paperwork for its own sake. It's that evidence which can't demonstrate an unbroken history can't be trusted to still be what it claims to be, it could have been altered, mislabelled, or substituted at any undocumented point.
This is why forensic teams don't grade a chain of custody as "mostly intact." It's either unbroken, or it isn't, because a single gap is enough to introduce genuine doubt about everything downstream of it.
A chain of custody isn't judged on how many links held up. It's judged on whether any link broke, because one broken link is enough to compromise everything that came after it.
The handoffs where B2B attribution data actually breaks
Attribution data has its own chain of custody, even if nobody calls it that. A lead's original source and campaign information has to survive several handoffs intact, and each one is a genuine opportunity for it to be silently altered.
Why "mostly working" attribution isn't actually working
The forensic principle applies with uncomfortable precision here. A pipeline where three of four handoffs are solid and one silently overwrites data isn't 75% reliable, it's unreliable, because that one broken handoff can corrupt any lead that happens to pass through it, and there's usually no way to tell which reported numbers were affected and which weren't.
A dashboard showing "Paid Search: 34% of pipeline" gives no indication of whether that number survived an unbroken chain or absorbed several leads that were actually something else before a sync silently reclassified them.
A metric that's been reported the same way for years accumulates trust precisely because nobody questions it, which is also exactly why a silent break introduced at some point along the way can go undetected for years.
First-touch, last-touch, and multi-touch attribution models all depend on the same underlying chain of custody. A break at the analytics-to-CRM handoff corrupts the input to every one of them equally, regardless of which model gets chosen afterward.
Building a chain that holds up under scrutiny
None of this requires forensic-grade infrastructure. It requires treating the data pipeline with the same scepticism a forensic team applies by default.
- Walk a single test lead through every system in the chain: Click an ad, land on the page, check the analytics event, check the CRM record, check the dashboard. Confirm the same source and campaign survive unchanged at every step.
- Lock the "original source" field so it can't be silently overwritten: Most CRMs support this. A repeat visit or a later touchpoint should be recorded as its own field, not allowed to overwrite the first-touch data that took real budget to earn.
- Log every manual override, not just system-recorded data: If a rep reclassifies a lead source, capture who made the change, when, and why. It doesn't need to be forbidden, it needs to be documented, exactly like a forensic handoff.
- Audit the chain on a schedule, not just when a number looks wrong: Broken handoffs are silent by nature, they don't announce themselves with an obviously wrong number. A scheduled quarterly walk-through catches breaks a "does this look right" review will miss.
- Attribution data passes through a genuine chain of custody: ad platform, landing page, analytics, CRM, and reporting dashboard
- A single undocumented handoff, most often a CRM sync or a manual override, can silently corrupt every model built on top of that data
- "Mostly working" attribution isn't meaningfully different from broken attribution, because the break is invisible from the reporting layer
- Lock the original source field, log manual overrides, and audit the full chain on a schedule rather than waiting for a number to look wrong
Our attribution modelling work starts with auditing the full data chain, not just choosing a model, so the number a CFO sees can actually withstand the question "how do we know that's right."
Explore Attribution Modelling →