Module 6 · GTM Analytics & Instrumentation · Chapter 25 of 32
Building the Bowtie Dashboard
A dashboard steers on the left of the Bowtie and gets graded on the right. Lay the tiles out that way, refresh each at the speed of the metric behind it, and close the loop so the board acts instead of just reporting.
11 min read
The first “GTM dashboard” I inherited had eleven tiles and every one of them was a corpse. ARR to date, GRR, NRR, a rolling NPS gauge, logo count, average deal size, a bookings-versus-plan bar. All true, all lagging, all reporting a quarter that had already been decided by the time the numbers hardened. A board member could read it and feel informed. An operator could read it and change nothing, because everything on it was an outcome, and you cannot steer a car by staring at the odometer. The week retention slipped, the GRR tile finally turned red in the following quarter’s board pack, long after the customers who left had stopped answering the phone. The dashboard was honest and useless at the same time.
The hierarchy of metrics explained why: those eleven tiles were investor-altitude and financial-altitude metrics, the ones that price the company, not the ones that run it. Run operations off the executive layer and you are always a lap behind. This chapter builds the layer underneath: an operating dashboard made from the metrics spine, volume and conversion and time by Bowtie stage, arranged so the metrics you can act on this week sit where your eye lands, and the metrics that grade you sit where they belong, off to the side as confirmation.
The board has a left and a right, and it matters which is which
The spine already sorted every metric onto a side of the Bowtie. Left-side metrics (leads, meetings, the response times, CR1 through CR4) lead, because they move first and you can touch them this week. Right-side metrics (onboarding, GRR, NRR, LTV, the delayed survey) lag, because they confirm months later what the leading signals already predicted. A dashboard is that same sorting made physical. Steer on the left, get graded on the right, and lay the tiles out so the layout itself teaches everyone reading it which is which.
The knot in the center is not decoration. It is where CR3 and CR4 sit, the two conversions pipeline management showed decide the quarter, plus the two ratios that read them in aggregate: coverage as a capacity question and velocity as dollars-per-day of throughput. Everything to the left feeds that knot. Everything to the right is what the knot produced, one cycle later. Read the board left to right and you are reading cause to effect, which is the only order in which a dashboard is any use to someone who has to do something on Tuesday.
Every tile earns its place or it comes off
A tile is not a fact you happen to have. It is a decision you intend to make, rendered as a number. If you cannot say what a person does differently when the tile moves, the tile is decoration, and decoration is worse than nothing because it fills the space a working instrument should occupy. Here is the tile set I actually keep, drawn straight off the spine, one motion at a time, each with the decision it is supposed to trigger.
| Tile | Reads (VM / CR / Δt) | Side | Refresh | The decision it should trigger |
|---|---|---|---|---|
| Pipeline coverage & velocity | VM4, CR3 x CR4, Δt4 | Knot | Daily | Add qualified pipeline or compress cycle before the gap is unrecoverable |
| CR3 x CR4 matrix, by rep | CR3, CR4 | Leading | Weekly | Route the fix: loose gate to ops, tight gate that still loses to enablement |
| Speed-to-lead | Δt1 | Leading | Real-time | Fix routing latency now; every minute past five costs qualification odds |
| New-business funnel | VM2 to VM5, CR1 to CR4 | Leading | Weekly | Open the one stage whose rate slipped, not “the funnel” as a group |
| Onboarding & first impact | CR6, Δt6 | Right (leads churn) | Weekly | Fire an onboarding play before the first renewal is ever at risk |
| GRR by motion | CR7 | Lagging | Monthly | Confirm durability; a drop here started one to two quarters upstream |
| NRR by motion | CR7 x CR8 | Lagging | Monthly | Check expansion is not masking churn; never comp on NRR alone |
| LTV : CAC by cohort | VM9 | Lagging | Quarterly | Altitude read for the board, not an operating lever you pull |
Two rows carry most of the teaching. The onboarding and first-impact tile is technically a right-side metric, but it is the earliest lagging signal you get, which makes it the closest thing the right side has to a leading indicator. Time to first impact (Δt6) predicts the first renewal months before GRR can confirm it, so it belongs high on the board even though it lives past the knot. And LTV:CAC sits at the bottom on purpose: it is a genuine, important number that an operator can do almost nothing about this week, so it earns a tile but never the headline. The hierarchy of metrics is the reason it is parked there instead of glowing at the top.
Refresh at the speed of the metric, not the speed of the tool
Modern BI will refresh anything as often as you ask, which is exactly the trap. A leading tile is worth watching often because you can act on it often: speed-to-lead in near real time, coverage and the CR3-by-CR4 matrix daily to weekly. A lagging tile refreshed hourly just manufactures noise, because GRR does not meaningfully move between Tuesday morning and Tuesday afternoon, and a number that jitters teaches people to ignore it. Match the cadence to the metric. Δt1 and coverage want a short pulse; GRR, NRR, and LTV want a monthly or quarterly beat where a real move actually means something. The dashboard’s refresh schedule is itself a statement about which tiles you expect people to steer on and which you expect them to check.
The closed loop beats the delayed survey
The most expensive gap on most right-side dashboards is the one between when a customer decides to leave and when your instrument tells you. The classic instrument there is the post-sale NPS survey, and it is a lagging confirmation dressed up as a signal. It arrives weeks after onboarding, often after the customer has already formed the judgment that decides the renewal, and it lands as a score on a slide rather than a task in someone’s queue. Revenue Architecture is blunt about the alternative: instrument a continuous closed feedback loop off leading signals instead of waiting on the delayed survey.
| Closed feedback loop (leading) | Delayed post-sale NPS (lagging) | |
|---|---|---|
| What it reads | An adoption or time-to-first-impact signal, on the record, the moment it moves | A survey score sent weeks after onboarding, once the churn judgment has usually formed |
| When it fires | Inside the same customer cycle, while you can still change the outcome | A quarter later, confirming what already happened |
| Who acts | A trigger play routes to the owning CSM automatically | A QBR slide notes the score fell |
| What it changes | The outcome, because the intervention lands before the renewal | The record, because the renewal has usually already been decided |
| Best use | Steering the right side of the Bowtie in flight | Grading the quarter and catching slow drift |
Closing the loop is what turns a dashboard from a mirror into a machine. The same CR3-by-CR4 matrix at the knot is the template: each cell prescribes an action, so a bad quarter routes to the desk that owns the weak cell instead of landing on the whole team as a group failing. Do the same on the right side. Wire the adoption drop, the stalled onboarding, the slipping Δt6 to a play that fires on its own, and the board stops being something you read on Monday and becomes something that acted on Thursday while you were in another meeting.
Wiring it, in order
The build has a sequence, and skipping steps is how you end up back at eleven corpses. The order matters because each step depends on the one before it holding.
- 1
Pull every tile from the governed definitions
Each tile reads a VM, CR, or Δt exactly as the metrics spine and the field guide define it, one motion at a time. A tile that means something different from the field guide is a second source of truth, which is the thing you were trying to kill.
- 2
Split the board leading versus lagging
Put the steerable left-side metrics and the knot where the eye lands first. Park the lagging outcomes to the side as confirmation, not as the headline. The layout should teach the reader which numbers to act on without anyone explaining it.
- 3
Match refresh to the metric behind the tile
Δt1 and coverage on a short pulse; GRR, NRR, LTV on a monthly or quarterly beat. Refreshing a lagging tile constantly only adds jitter that trains people to ignore it.
- 4
Give every tile an owner and a written decision
Next to each tile, name who acts and what they do when it moves. Any tile that fails this comes off the operating board and goes to a monthly review instead.
- 5
Close at least one loop
Wire one leading signal (a stalled onboarding, a slipping Δt6, an adoption drop) to an automatic play so the dashboard fires an action, not just a color change. Then add the next loop once the first one is trusted.
Handing off
You now have an operating dashboard, not a board pack: tiles pulled from the governed spine, laid out so leading metrics steer and lagging metrics grade, refreshed at the speed of the metric behind each one, every tile tied to a decision, and at least one closed loop firing a play before the delayed survey could ever land. What the board shows you, though, is a single point in time. A tile that reads 22% this week is not good or bad until you know whether it was 26% last quarter and heading down, or 18% and climbing. A snapshot cannot tell you which. For that you have to watch the tiles as a line across cohorts and read the shape of the curve, which is Cohort and Trendline Analysis, and it is next.