Sourced, Influenced, and What Replaces Them

Marketing reports that it sourced 31% of pipeline and influenced 78%.

Sales reports that reps generated most of the same deals through outbound and their own relationships. The two figures are computed from the same CRM, both follow their stated rules exactly, and they sum to considerably more than the pipeline that exists. Everyone has known this for years and the meeting happens again next quarter.

The decision it encodes

Which number governs the pipeline conversation, and what are the others for?

Somebody has to answer that, because the alternative is what most organizations have: several numbers with no ranking, each owned by the team it flatters, and a quarterly argument that resolves by seniority.

Both conventions divide something that already happened

Sourced assigns a deal to whichever touch or process the CRM convention chose: the first touch, the lead source, or whatever created the record. It is a deterministic attribution rule, and it inherits the defect every such rule has: no counterfactual anywhere in it.

Influenced counts a deal if marketing touched the account at any point. It is generous by construction, it rises whenever marketing does more of anything, and on a mature account base it converges toward everything.

Neither is dishonest, and neither answers the question a budget owner is asking. They divide observed pipeline between contributors. Whether that pipeline would have existed without the program is not a quantity either one contains. Attribution is not incrementality makes the general argument; this is the same argument in a CRM.

What replaces them is an evidence ledger

The replacement is not a cleverer division of credit. It is a ledger of the measured effects from the designs that were run, each recorded with its scope, and an investment rule that reads the ledger rather than the conventions.

That is the same move the lifecycle system makes in evidence review: rank estimates by what they assumed rather than by how precise they look, and record each with its source, its interval, and its date. A territory rollout with a wide interval outranks an influenced-pipeline figure with no interval at all, because one of them has a comparison in it. What the ledger is not is one universal marketing-generated-pipeline number. A territory test of a webinar says something about webinars in those territories under that quarter’s sales policy, and it becomes the assumption for events, intent signals, and direct mail only by partial pooling across interventions similar enough to share an estimate, with the pooling stated.

What one entry in the ledger carries

Field	Why it is in the record
Intervention and population	What was added, and to which accounts by state at assignment, since creation and acceleration are different questions.
Assignment unit and design	Account, territory, partner, or rollout wave, and which rung of the ladder produced the comparison.
Outcome and horizon	The early outcome and the final one, each with the share of the cohort mature when it was read.
Sales policy in force	What reps were doing to both arms while the program ran, because the effect includes the response and does not transport to a different policy.
Effect, interval, and compliance	The intention-to-treat estimate with its interval, and how much of the assigned treatment was actually delivered.
Date and transport	When it was measured, and to which interventions and segments it may be applied as an assumption, and to which it may not.

Keep the conventions for what they are good at

Sourced and influenced should not be deleted, and saying so matters, because a proposal to delete them will be rejected and take the rest of the system with it.

Influenced pipeline is a coverage measure: which target accounts marketing has reached, and which have never been touched. That is useful for planning, it is what the metric measures, and it makes no causal claim. Sourced is an operational routing measure: which channel created a record, which matters for process and for finding where records come from.

What changes is the ranking. Neither one is permitted to be the answer to whether the program worked, and both carry a label saying which question they answer.

Maturity, honestly staged

Stage 1: both conventions labeled with the question they answer. Coverage and routing, not contribution. Costs nothing and stops the worst usage.

Stage 2: existing-pipeline accounts separated in the influenced figure. The share of influence that is accounts researching a purchase they had already decided to make becomes visible.

Stage 3: an evidence ledger, with scope on every entry. Each measured effect recorded with its intervention, population, unit, horizon, sales policy, interval, and where it may be transported, sitting above the conventions rather than beside them.

Stage 4: an investment rule that reads the ledger. Budget moves on measured effects, pooled only across interventions similar enough to share an estimate, and the conventions are reported as coverage without anybody mistaking them for evidence.

Why a better attribution model does not fix this Technical

The usual proposed fix is a multi-touch model: weight the touches, fit something on historical deals, and produce a fractional credit per channel. That improves the division and changes nothing about the underlying problem, because every input is a touch that happened on a deal that happened, and nothing in the data observes an account that was not touched.

In B2B the confound is worse than in consumer attribution rather than better, for two reasons. Buying groups mean the touches are spread across people whose relationship to the decision is unknown. And the reverse causation is stronger: research intensity rises as a purchase approaches, so the touch count is partly an outcome of deal progress rather than a cause of it. A model fitted on that will find touches predictive and will be reading the arrow backwards.

What a multi-touch model is legitimately good at is the same thing the conventions are good at, which is describing coverage and pathways. The thread on these methods sets out the boundary in general terms, and it holds here without modification.

Context

Best suited for

Organizations where marketing and sales each report a pipeline contribution figure and the two are never reconciled, which is most of them.

Data requirements

• Touch history joined to accounts through the account graph, with dates
• The existing-pipeline flag as at first touch, to separate research from influence
• Whatever measured effects exist from run designs, with intervals and dates
• Somewhere durable for the evidence ledger, with the scope fields, that planning reads from

What changes by context

B2B SaaS

Product-qualified accounts arrive through usage rather than through a touch, so they appear as unsourced and unattributed even when marketing drove the original signup. The conventions understate the motion most of these businesses rely on.

Sales-led

Reps create the records, so sourced credit follows CRM habits as much as reality. A change in how records are logged will move sourced share without anything real changing, which is worth knowing before anyone celebrates a trend.

Partner-led

A third party sits between marketing and the deal, so both conventions have to decide what a partner-registered deal counts as. Whatever is chosen, it should be the same choice in both directions rather than the flattering one in each report.

When it breaks down

• Organizations with no run designs at all, where there is no governing estimate to promote and the statement is that contribution is currently unmeasured.
• Programs evaluated purely on sourced pipeline by contract, such as some agency arrangements, where the metric is the deliverable regardless of what it means.
• Very small deal counts, where any split of a handful of deals is dominated by which ones happened to close.