Module 6 · GTM Analytics & Instrumentation · Chapter 27 of 32
Attribution and Pipeline Analytics
Two dashboards can report the same month as $3.19M and $1.3M of pipeline and both be right. Learn to reconcile sourcing and conversion by definition, retire attribution theater, and read pipeline generation as the leading indicator it actually is.
10 min read
Our SVP of sales sent one question into the ops channel on a Tuesday: which July pipeline number is right, the $3.19M in the Salesforce report or the $1.3M in the Omni exhibit? He was not being difficult. He had two dashboards open, both built by people he trusted, both showing the same month, and they disagreed by more than the entire gap between making the quarter and missing it. The meeting to review pipeline generation had stalled on the arithmetic before anyone got to the actual problem, which was that generation had cratered. That is attribution theater in its purest form: a room spending its scarcest hour arguing about whose number is real instead of asking why the factory stopped producing.
Both numbers were correct. Neither was lying. They were different metrics wearing the same word, and the job of pipeline analytics is to make sure that never costs you a meeting again.
A pipeline number is a definition, not a fact
Pipeline analytics answers two questions and only two: where did pipeline come from, and how does it convert. The first is sourcing and generation, the second is funnel conversion. Attribution theater is what happens when you answer either one to feel good rather than to decide something. It is picking the touch model that flatters your channel, crediting a webinar for a deal the field had already worked for a quarter, or comparing a marketing-only numerator to a whole-company target and calling the result performance. The antidote is boring and it works: define every number as a query, reconcile the definitions before you reconcile the people, and govern the one definition you run the business on.
Start with the $3.19M and the $1.3M. The Salesforce report counted every opportunity with new ARR above zero, across all deal types, stamped with a discovery date this year: $3,190,594.50 across 72 opps for July. The Omni tile counted new-business ARR by SQO date with no floor on the amount: $1,875,160.50 across 67. The $1.3M our SVP quoted was neither of those; it was the marketing-sourced subset of the Omni figure, $1,267,046, being graded against a $15M marketing target. Three numbers, three definitions, one month. The bridge between them is exact and it is the whole lesson.
| Step | Amount |
|---|---|
| Salesforce report (all types, discovery date, new ARR > 0) | $3,190,594.50 |
| less Expansion, 13 opps (Omni counts new business only) | −$1,377,162.00 |
| plus one deal with an SQO date but no discovery date | +$61,728.00 |
| = Omni “Generated Pipeline by SQO Date” (new business) | $1,875,160.50 |
| less non-marketing-sourced, 27 opps | −$608,114.50 |
| = marketing-sourced (Source Type = Marketing MPA/MEA) | $1,267,046.00 |
Read the bridge and the argument evaporates. The gap is not a data error and it is not a date problem: 59 of the 72 opps appear in both populations, and discovery date versus SQO date moved exactly one record in nearly nine hundred, so the date basis is a rounding effect. Deal type is 96% of the gap, and a single Expansion deal (“Snowflake third-party redistribution”) is $900K of it. The $3.19M looked like the rescue number in the room. It was not new-business generation at all; it was mostly one large expansion opportunity dressed as pipeline gen. Once you can write the bridge, the honest headline writes itself, and it is not the flattering one.
- 1
Reduce each number to its query, not its dashboard
Write down the four filters that define it: deal type scope, date basis (discovery, SQO, created, close), the amount floor, and the source filter. Most disputes are two people quoting different values on one or two of these four and never saying so out loud.
- 2
Diff the populations at the opportunity level
Pull both opportunity lists and compare row by row. You are looking for which specific records are in one set and not the other. In our case 59 of 72 overlapped, so the story lived in the 13 that did not, and one of those 13 was almost the entire gap.
- 3
Build the bridge as a waterfall
Start from the larger number and subtract or add one definitional change at a time until you land on the smaller one to the dollar. If it does not reconcile to the cent, you have a data problem hiding behind the definitional one, and now you know to go find it.
- 4
Name which definition runs the business
The point of reconciling is not to declare a winner; it is to pick the one number that goes on the operating dashboard and hold everything else as a supporting cut. Both can exist. Only one gets to drive the meeting.
Sourcing without the theater
Attribution is where analytics goes to lie to itself. First-touch credits the channel that found the account and ignores the six months of field work that closed it. Last-touch does the reverse. Multi-touch models split the credit fractionally across every recorded interaction, which sounds fairer and mostly manufactures the illusion of precision from timestamps you do not trust. Marketing prefers the model that maximizes marketing’s number; sales prefers the one that maximizes sales’. Every model is internally defensible, which is exactly why letting the model float is a governance failure, not an analytics choice.
The operator’s move is not to find the true model. There isn’t one. It is to pick a single, opportunity-level source-of-record field, define it once in plain language (“what originated this opportunity”), and make that the only field that feeds pipeline-generation reporting. Ours is a Source Type on the opportunity with governed values (Marketing MPA, Marketing MEA, and the non-marketing origins). One field, one owner, one meaning. Multi-touch dashboards can still exist as a diagnostic for the marketing team to tune spend, but they do not get to be the number in the pipeline-gen review, because a fractional model that no one can reconcile to a headcount decision is theater with better graphics.
That field earns its authority only if it is governed. Revenue Architecture is blunt about this: once a team agrees on its data model, adherence has to be enforced through a change-request process, because changes to how you count ripple all the way up to how the company is valued. A source field that any admin can quietly repoint, or that reps free-text, is not a source of record; it is a rumor with a picklist. The stewardship mechanics belong to Data Governance and Quality and the deeper reference in the data-governance topic; the analytics point is narrower. If the definition of “sourced” can drift between two reports, your attribution is already theater, and no dashboard will save it.
Generation is a leading indicator: read the count first
Sourcing tells you where pipeline came from. Generation tells you whether there is any, and it is the earliest honest signal you have. It sits at the very top of the Bowtie, so it moves quarters before the revenue does, which makes it the number worth self-benchmarking on a trendline exactly the way the previous chapter taught: track your own new-business generation month over month and let the slope, not a single tile, tell the story.
Ours told an ugly one. New-business SQO pipeline generated in July ran $17.98M in 2024, $6.48M in 2025, and $1.88M in 2026. That is not a wobble, it is a curve bending down two years running, and the $3.19M “all-types” number existed largely to soften it. The sharper read was the count, not the dollars: 67 new-business SQOs in July 2026 against 94 the month before and 185 in the same month a year earlier. Dollars are noisy because one large deal swings them (see the $900K expansion). Opportunity count is closer to raw throughput, and it fell first. When you are diagnosing a generation problem, trust the count over the ARR, because the count is harder to rescue with a single fat opportunity and it degrades before the money does.
How it converts, and where the handoff leaks
Once pipeline exists, the second question is conversion, and Revenue Architecture gives it a spine: CR1 for lead-generation efficiency, CR2 for outreach, CR3 for prioritization (the share of discovery meetings that become accepted qualified opportunities), and CR4 for win rate. Marketing effectively owns CR1 and CR2, sales owns CR3 and CR4, and the handoff sits at the Prioritization checkpoint we defined as contracts back in Lead-to-Revenue Routing. Attribution analytics without conversion analytics is half a picture: it tells you a channel produced volume and never whether that volume was worth a rep’s calendar.
View as table
| Stage | Value |
|---|---|
| MQLs | 5,215 |
| Meeting (proxy) | 726 |
| New-business SAO | 89 |
| SQO | 69 |
Follow the cohort and the reliable seam declares itself. SAO to SQO holds at 74 to 77% every quarter across 2025 and 2026, which is a rate you can multiply against and forecast on. The steps above it are noisier: MQL to meeting swings between 8 and 21% because “meeting” is an account-level proxy stitched from the latest meeting date, not a true per-lead event, and the upstream capture is thin. End to end, MQL to SQO lands around 1 to 1.8% under strict same-record attribution. That number is lower than the 5 to 9% you get by crediting any new-business opportunity on an MQL’s account, and both are defensible; the governed same-record definition is the one that goes on the dashboard, and the looser one is the diagnostic. Same discipline as sourcing: pick the honest definition, keep the flattering one labeled as a cut.
The handoff analytics are where the two questions collide, and where our July meeting was quietly wrong before anyone opened a dashboard. The marketing-sourced generation number ($1.27M) was being compared to a $15M target, but no one had confirmed whether that $15M was a marketing-sourced goal or total company pipeline. Grade a marketing-only numerator against a whole-pipeline denominator and marketing will always look like it is failing by design, no matter what it produced. That is not a marketing problem or a sales problem; it is an analytics-hygiene problem, and it is the operator’s to catch. The rule is simple: the numerator and the target have to share a definition, or the ratio is theater.
You can now measure the top of the factory honestly: reduce any pipeline number to its query, bridge two conflicting numbers to the dollar, run sourcing off one governed field instead of a floating touch model, read generation as a leading indicator with the count ahead of the ARR, and watch conversion across the marketing-to-sales seam without grading a partial numerator against a whole-pipeline goal. Honest measurement is inert until it changes what a room decides on a fixed schedule. Turning these numbers into a weekly rhythm of decisions, the GTM Council and the operating cadence that reviews the exceptions rather than re-litigating the arithmetic, is The Operating Cadence and GTM Council, and it opens Module 7.