A deal closes after nine months. Marketing pulls the record and finds four touchpoints against the contact who signed.
Four touches, nine months. The number is absurd on its face, and everyone knows it. What actually happened is that an analyst read the documentation in March, a director sat in on a webinar in May, someone in procurement compared your pricing page against two competitors in July, and the person who signed showed up in August having been told by three colleagues that this was the option.
Your CRM recorded the last one. The other three were doing the work.
This is where lead-level touchpoint counting stops being useful in B2B, and it is a structural problem rather than a tracking bug.
The buying committee breaks the unit of analysis
Lead-level measurement assumes one person makes one decision. In B2B they frequently don’t: an evaluator, a user, a budget holder, a technical reviewer, someone in security or legal, and often someone whose only involvement is saying “we used them before and it was fine.”
Count touches per lead and you get a number per person, when the thing that converted was a group. It’s not a smaller version of the truth — it’s a measure of a different object. (If you haven’t set up per-lead counting yet, start there — the identity and taxonomy work below assumes it.)
The fix is to change the unit: count touches per account, not per lead.
Rolling up to the account
Mechanically this is a grouping operation, and mechanically it’s easy. The difficulty is entirely in the matching.
By CRM association. If contacts are properly linked to company records, sum the touches across every contact on the account. Cleanest option, and it only covers people who are already contacts. Whether it works at all depends on how well the contact-to-company relationship was set up in the first place — in HubSpot and most CRMs that is a configuration decision nobody revisits after launch.
By email domain. Group anonymous and known activity by @company.com. Simple, and it fails on the cases you care about: consultants using personal addresses, subsidiaries on different domains, and free-mail addresses from people at companies that use them.
By IP or reverse-IP vendor. Attempts to identify company-level visits with no form fill. Coverage varies a great deal by vendor and by whether the visitor is on a corporate network at all — which, several years into remote work, is often not the case. Treat the output as a directional signal, not a record.
Most teams end up combining the first two and treating the third as a bonus. That is a reasonable place to land.
What you lose in the roll-up
Two things, and both matter.
Sequence within the account disappears. “Eleven touches” tells you the account engaged. It doesn’t tell you the analyst read everything in March and the budget holder appeared in August — which is the part that would tell you what to send next.
Account size distorts the count. A 4,000-person enterprise generates more touches than a 40-person company for reasons that have nothing to do with buying intent. Comparing raw counts across segments produces a finding that is really just a headcount ranking. Normalise by segment, or compare accounts only against similar accounts.
The grid: which role touched what, and when
This is the version worth having, and it is the reason to do any of this.
The count tells you an account engaged. The grid tells you who engaged and when — which converts the analysis from a number into a decision about what to make next.
Roles down the side, stages across the top. For a typical B2B software or services purchase it looks something like this:
| Role | Problem aware | Evaluating | Validating | Approving |
|---|---|---|---|---|
| Champion / user | Blog, peer conversation | Comparisons, demo | Case studies | — |
| Technical evaluator | — | Docs, integration pages | Trial, API reference | — |
| Security / legal | — | — | Security page, DPA, SOC 2 | — |
| Economic buyer | — | Pricing page | ROI case, references | Contract |
| Procurement | — | — | Pricing, terms | Contract, MSA |
Fill your own version in from three sources, in this order:
- Closed-won deals. Take ten. For each, list everyone the CRM shows was involved, their role, and what they touched. Slow, manual, and by far the most accurate input you will get.
- Form fills with a job-title field. Gives you role data for the self-identified minority.
- Sales notes. “Looped in their head of security” is a data point, and it is usually sitting in a call summary nobody has read since.
The empty cells are the finding
This is the part people miss. A grid full of entries tells you the machine works. The blanks are where the money is.
An empty row means a role you have nothing for. If the security reviewer column is bare across ten deals, either they never look — unlikely, if you sell to companies with a security function — or they looked, found nothing addressed to them, and formed an opinion anyway. That opinion was formed by your competitor’s documentation.
An empty column means a stage where you go quiet. Accounts that engage heavily during evaluation and then show nothing during validation are not losing interest; they are validating somewhere you can’t see, usually on review sites and in peer conversations.
A cell where the same asset appears for every role means you are sending one piece of content to five people with five different questions. It works for whoever it was written for and does nothing for the other four.
The holes you cannot fill, and which ones matter
Most of the committee never identifies itself. Your role data covers people who filled in a form. The security reviewer who read your documentation for forty minutes and never gave a name is invisible, and no configuration fixes that.
So the honest grid has gaps, and the skill is knowing which gaps to care about:
- Gaps in roles you sell to directly — champion, economic buyer — are worth chasing, because those people do eventually identify themselves and the data is recoverable from sales notes.
- Gaps in gatekeeper roles — security, legal, procurement — are usually permanent. These people research anonymously and appear only at contract stage. Do not try to instrument them. Instead, ask your champion what their security team asked for, and build to that answer.
The second is the more useful discipline. It replaces a tracking problem you cannot solve with a discovery question you can ask on any call.
What to do about the dark funnel instead of pretending to measure it
A large share of B2B research now happens where you cannot see it: private Slack groups, peer communities, review sites, a text message to a former colleague.
The standard responses are bad. One is to buy a tool that claims to reveal it and treat its output as fact. The other is to ignore it and let the measurable channels take credit for work they didn’t do.
Three things work better, none of which involve tracking:
Ask. A single “how did you hear about us?” field on a form, unstructured and optional, will name sources your analytics has never recorded. The data is messy and self-reported and still more honest than an attribution model asserting confidence about it.
Ask better in sales conversations. “Who else has looked at this internally?” and “what did you read before this call?” are ordinary discovery questions. The answers describe the committee and the dark funnel at once, and they cost nothing to collect if someone writes them down.
Watch the gap. If your self-reported sources consistently name places your analytics doesn’t, that gap is the dark funnel, and its size is a number you can track over time even when its contents stay invisible.
When account-level analysis isn’t worth the setup
It is worth being direct about this, because the work is not free.
Skip it if your sales cycle is short and single-threaded. One person deciding in three weeks is a lead, and lead-level counting describes it correctly.
Skip it if your CRM associations are unreliable. Rolling up to accounts on top of bad contact-to-company matching produces confident numbers built on wrong groupings, which is worse than no numbers — nobody doubts a dashboard.
Skip it if you have fewer than a few dozen accounts in the segment. At that volume you can read the deals individually and learn more than the aggregate would tell you. Reading twenty account histories in an afternoon is genuinely better analysis than a chart built from twenty rows.
Do it when several people are demonstrably involved, cycles run over months, and there are enough accounts that patterns can exist. Professional services firms sit almost exactly on that line, which is why the pattern shows up there first.
Short version
- The buying committee, not the lead, is the unit that converts.
- Roll up by CRM association first, email domain second, reverse-IP as a hint.
- Expect to lose sequence and to gain a headcount bias; normalise by segment.
- Build the role-by-stage grid from ten closed-won deals, not from your analytics.
- Read the empty cells: a blank row is a role you ignore, a blank column is a stage where you go silent.
- Stop trying to instrument security, legal and procurement. Ask your champion what they asked for.
- Handle the dark funnel by asking, not by buying a tool that claims to see it.
- Below a few dozen accounts, read the deals instead.
Defining who the committee is in the first place — the roles, the order they appear in, what each one needs — is ICP and customer journey mapping.