Lag-Aware Readouts

The program launched in February. The board deck is due in May. The median deal in this segment takes seven months to close.

So the May readout will report leads, or meetings booked, or opportunities created, because those are the things that have happened. Whichever one it reports will be treated as the answer, the budget will be set on it, and the question of whether any of it turned into revenue will be asked in about a year, by which time the program has been changed twice and nobody can attribute anything.

The decision it encodes

What do we measure now, when the outcome we care about lands in two quarters?

There is no arrangement that removes this tension. The readout either waits for the outcome the business cares about, and arrives after the decision, or it uses a proxy and inherits whatever the proxy gets wrong. What a system can do is make the choice explicit and price it.

Every proxy is a bet on a conversion rate

A meeting is not revenue. Treating a lift in meetings as a lift in revenue assumes the meeting-to-revenue rate is the same in both arms, and the whole point of a marketing program is often to change that rate.

The ladder of outcomes, and what each one costs to use

Outcome	What you gain and what you assume
Engagement	Available in days. Assumes engaged accounts buy at the same rate as before, which a program designed to reach new accounts violates by construction.
Meetings booked	Weeks. Assumes the meeting quality is unchanged, and reps are the ones booking them, so the measure moves when rep behavior moves.
Opportunity created	A month or two. The first outcome a sales organization recognizes, and the first that a rep can create at will when a number is due.
Stage 3 reached	A quarter. Much harder to create artificially and still well short of revenue.
Closed revenue	Quarters. The thing anybody wanted, arriving after the decision it should have informed, and lumpy enough that one deal moves a territory.

Three things a proxy readout can say, and only one is causal about revenue

A historical relationship between meetings and revenue does not make meetings a surrogate. Meetings can predict revenue perfectly across the account base while a particular program raises low-quality meetings that never convert, and a positive effect on the proxy can sit beside no effect on revenue. Historical predictiveness is about the population; surrogacy is about whether treatment effects on the proxy translate into treatment effects on the outcome, and the second is not implied by the first.

So the readout keeps three statements apart. The causal effect on the early outcome: the program raised qualified meetings by twelve percent, with its interval. The forecast of eventual revenue: on the historical conversion for this segment, those meetings are expected to produce a stated amount, with a wide interval, and labeled as a forecast that assumes the program did not change who is in the meetings. And the causal effect on revenue, which is unknown until revenue matures, and stays open in the readout until it does. A proxy earns the right to stand in for the third only after its treatment effects have predicted revenue treatment effects across several past programs of the same kind, which is the same standard the growth system holds activation to. Reporting a meetings lift as revenue at the point estimate, with no interval and no label, is the step that turns a defensible proxy into an indefensible claim.

Report the same effect at several horizons, and say when the cohort is mature

The single most useful discipline here. The same estimate at several points since assignment, reported together, distinguishes an effect that is still accruing from one that has already reversed, and it stops the horizon being chosen after the numbers are visible, which is the failure mode that makes proxy-based reporting untrustworthy in the first place.

The horizons come from the segment’s own sales cycle rather than from the calendar: the median and the eightieth-percentile time to close, so that the readout says what share of the cohort is mature at each point. Cumulative pipeline by time since assignment, time-to-stage curves by arm, and the mature share are what the readout shows. Several horizons do not by themselves prove that a lift was or was not pulled forward; only a terminal horizon long enough for the control arm to catch up can, and it is declared before launch.

Alongside the account-level lines sits the portfolio line: total pipeline across the territories or reps whose hours the program touched, which is where pipeline that was moved rather than created shows itself.

Say when nothing was learned

B2B sample sizes make this the normal outcome rather than the exception. A territory design on twelve territories, reading a lumpy outcome over two quarters, will frequently return an interval spanning zero and something large.

That is a result, and the useful form of it names what would resolve it: a longer window, a less lumpy outcome, more territories, or a larger program. The evidence review pattern from the lifecycle system applies unchanged, and it matters more here because the temptation to re-cut is stronger when the alternative is telling a board that a quarter of work cannot be evaluated.

Maturity, honestly staged

Stage 1: one declared outcome and horizon, fixed before launch. Ends the practice of reporting whichever metric moved.

Stage 2: several horizons reported together. Accrual, reversal, and pull-forward become distinguishable.

Stage 3: the revenue forecast calibrated and labeled as a forecast. A meetings lift can be converted to an expected revenue figure with an interval, and the readout says it is a forecast rather than a measured effect on revenue. Inconclusive is a permitted verdict, with what would resolve it named.

Stage 4: proxies validated at the treatment-effect level. Across past programs of the same kind, early-outcome effects are compared with the revenue effects they eventually produced, and a proxy decides a plan only where that translation has held; the final outcome is reconciled against the forecast every time it matures.

Why closed revenue is the noisiest thing you can measure Technical

Deal values in most B2B segments are heavily skewed, so the variance of total revenue in a territory is dominated by whether one large deal landed inside the window. The count of accounts reaching a stage has far lower variance than the value of what they reach, which is why a count-based outcome usually detects an effect that a revenue-based one cannot, on the same data.

The practical construction is to estimate the effect on counts, which is the part the design can see, and to report the value question separately rather than multiplying the count effect by a historical average deal value. Multiplying assumes the program did not change the mix of accounts, the deal size, or the close probability, and changing those is often what a program does. Report the incremental count, the distribution of deal value in each arm as it matures, and eventual revenue by arm; where a forecast is needed before maturity, a two-part model (whether a deal lands, and how large) with its assumptions labeled, or a predeclared capped revenue outcome, is cleaner than a conversion factor. That separates the thing that was measured from the thing that was assumed, and it stops a single outlier deal from becoming the finding.

Winsorizing or capping deal values is a defensible alternative and changes the estimand to the effect on typical deals, which has to be stated. What is not defensible is capping after seeing which arm the outlier landed in.

Context

Best suited for

Programs whose outcome arrives later than the planning cycle that funds them, which is nearly every B2B marketing program.

Data requirements

• Stage transition history with dates, so an outcome can be read at any horizon rather than only the current state
• Deal values joined to accounts, for converting count effects into value with an interval
• The sales-cycle distribution by segment, for the median and eightieth-percentile maturation horizons, and past programs’ early and final effects for validating proxies at the treatment-effect level
• The assignment and its date, so cohorts can be aligned on time since treatment rather than on calendar quarters

What changes by context

B2B SaaS

Product usage arrives in days and correlates with eventual expansion, which makes it the most useful early proxy available. It is also the one least under a rep’s control, so it is harder to manufacture than an opportunity record.

Sales-led

Most proxies above closed revenue are created by reps, so they move when rep behavior moves. That is not noise, it is part of the effect, and it has to be reported as program plus response rather than as program alone.

When it breaks down

• Sales cycles longer than the life of the program, where no horizon inside the decision window contains the outcome and the answer is that the question cannot be settled.
• Segments with a handful of enormous deals, where revenue variance swamps any plausible effect regardless of horizon.
• Organizations that will act on the shortest available proxy whatever it is accompanied by, for whom adding horizons adds only length.