Sales-Led Pipeline

A rep closes a deal in March. Marketing touched the account in January with a webinar and in February with a case study, and sales had been working it since November. The CRM records whichever of those facts its rules prefer, and both teams present the deal as theirs.

Everybody in the room knows this. The meeting happens again next quarter, because there is no third number with any better claim to being right.

The unit is owned by somebody with a quota

Every system on this site withholds something. Lifecycle withholds a message. Performance goes dark in a market. Organic withholds a change from half a template. Each is uncomfortable and each is possible.

Here the unit is a named account with an owner, a quota attached to it, and a list that exists because somebody believes those companies will buy. Asking for four hundred of them to be left alone for two quarters is asking a rep to miss a number for a measurement they did not request. That answer is no, and unlike the others it does not become yes when the case is argued better.

What is refused is withholding the selling. The marketing program is another matter. A control account can be sold to exactly as before and still not receive the webinar invitation, the direct mail, the intent alert, or the ABM advertising, and the comparison between accounts that got the program on top of normal selling and accounts that got normal selling alone is the effect of adding the program. That is a materially easier conversation than a do-not-touch list, and it is the design this system reaches for first. Where even that is refused, or where the program works through the rep and cannot be assigned per account, the comparison moves up to territories, partners, or the order of a staged rollout.

Two things make it harder than the rest

The outcome arrives after the decision. Deals close over quarters and plans are set in weeks, so every readout uses a proxy, and every proxy is a bet on a conversion rate the program may itself be changing.

The rep’s response is part of the effect. A campaign touches an account, a rep notices and works it harder, a deal appears. Effort the rep had already put in before the program is a confounder, and a design has to balance it. Effort the program caused is a mediator: it is the path the program works through, and adjusting for it changes the question from what the program did to something narrower that needs much stronger assumptions. The quantity this system estimates is the total effect of assigning the program under the normal sales response, with rep activity recorded as the mechanism and as compliance rather than adjusted away, and every readout says so.

A program can move pipeline without creating it. Rep hours are fixed inside a quarter. A strong marketing signal on one account buys it four extra hours, and those hours came from another account. The treated account progresses, the territory’s total pipeline may not, and an account-level readout cannot tell the difference. Every readout therefore reports two numbers: the effect among assigned accounts, and the effect on total pipeline across the rep or territory whose capacity the program redirected. Where the redirection is large, the territory is the unit to assign in the first place.

The four stages, in plain terms

Account graph is one row per buying decision, the contacts that resolve to it, and the state each was in when it was treated rather than today. The outcome lives at the buying decision and the treatment lands on people, so both grains have to exist and every reported number has to say which one it is. It also carries which question an account is eligible for: creating pipeline where there is none, accelerating an open opportunity, expanding a customer, or reactivating a dormant one, since those are different estimands on different populations.

Design is the ladder of what can be assigned. Blocked account-level randomization of the marketing treatment while sales keeps selling, where the treatment can be delivered per account. Territories or partners as clusters where it cannot, or where rep capacity would be redirected between arms. A randomized rollout order where everyone must eventually get it. A randomized encouragement where only a signal can be assigned. And a matched observational comparison, with its assumptions stated, only when nothing can be.

Readouts choose an early outcome that can be observed inside the decision window, report it at several horizons so accrual and reversal are distinguishable, keep the final outcome open until it matures, and demote sourced and influenced pipeline to what they measure, which is coverage and routing rather than contribution. Above them sits an evidence ledger: every measured effect with its intervention, population, unit, outcome, horizon, and the sales policy it was measured under, which is what the investment rule reads.

Prioritization allocates the scarcest thing in the building. Rep hours cannot be bought this quarter, so every account worked displaces one where a rep might have changed the outcome, and the ranking has to be by incremental value per hour rather than by likelihood to buy. It keeps a log of which accounts were assigned work, at what probability, and what was actually done, with a two-sided exploration band around the cutoff, so a new ranking can be scored on last quarter before a rep ever sees it.

Two loops, and they are not the same experiment

This system contains two learning loops that share the account graph and the outcome machinery and share nothing else: not the treatment, not the comparison, not the estimator, not the decision.

The program loop asks whether a webinar, an ABM program, an event, or an intent signal created or accelerated pipeline. Its treatment is the marketing program, its comparison is the accounts, territories, or rollout waves that did not get it, and its dashed edge runs from the evidence ledger back to design: every measured effect is recorded with its scope, and the next program is designed and funded on that record rather than on sourced pipeline.

The allocation loop asks which accounts should receive a scarce rep intervention. Its treatment is rep work, its comparison is the exploration band around the ranking cutoff, and its dashed edge runs from the prioritization log back to capacity: every allocation is logged with the odds each account had of being assigned, so a candidate ranking can be scored on the log before a rep sees it. Without that edge the model retrains each quarter on outcomes generated by its own previous recommendations, grows more confident about a shrinking set of accounts, and never discovers what it was wrong about.

The exploration band is the allocation loop’s instrument. It says what a rep hour does on accounts the ranking would have skipped. It says nothing about whether the webinar worked, and an earlier version of this page let it stand in for a standing control on the programs, which made the system look more closed than it is. The program loop’s standing comparison is a default protocol per eligible program, and it is negotiated, and the hub says so.

Where to start when none of it exists

First, the account graph, and specifically the open-opportunity state at assignment. Accounts already in a deal cycle belong to a different question than accounts with none, and until the two are separated every measurement of pipeline creation is flattered by accounts that were already buying.

Then one program assigned at the account, with sales left alone. Not a do-not-touch list, which will be refused, but a program half the eligible accounts receive on top of normal selling, blocked within rep and stage. Where that is refused, a rollout whose territory order was drawn rather than chosen. One assigned comparison is worth more than a year of sourced-pipeline reporting.

Then horizons on the readout, and the ledger. The same effect at the segment’s median and eightieth-percentile maturation times, which costs nothing once stage history exists and prevents the horizon being chosen after the numbers are visible; and a ledger entry with the effect’s scope, so the next plan has something to read.

Then capacity, and prioritization last. A ranked list changes nothing until somebody has written down how many accounts a rep can hold and what a worked account receives.

What this leaves out

Self-serve. Where the buyer is an individual and the account is a formality, the lifecycle system is the better fit and its machinery applies directly.

Sales enablement and rep coaching. Both plainly affect the outcome and both are outside what a marketing measurement system can claim, which is worth stating because the total effect this system reports includes the sales response they shape.