A metric is a promise your data has to keep
Most metrics advice hands you a list of forty things to track and leaves you to guess which five matter this quarter. The list is not the hard part. Every metric on it is defensible somewhere. The hard part is that each one silently assumes something about your data, and if that assumption is false the metric still returns a number. It just returns a wrong one, confidently, on a dashboard, in front of your board.
Win rate assumes your losses get marked lost. Sales cycle length assumes stage timestamps mean something. Pipeline coverage assumes your close dates are not all set to the last day of the quarter out of habit. Lead conversion rate assumes marketing and sales agree on what a lead is.
Break any of those assumptions and you do not get a missing metric. You get a plausible one, which is far more expensive, because people make decisions on it.
So the useful question is not "what should we track". It is "what have we earned the right to track". This is the ladder we use to answer that.
The readiness test
Before you add a metric to a dashboard, put it through four questions. If any answer is no, you are not ready for that metric, and adding it will cost you more than skipping it.
- Can two people compute this independently and get the same number?
- Is every field it depends on filled in more than 90% of the time, without someone being chased?
- If this number moves 20% next month, do you know which lever to pull?
- Is there a decision that changes based on this number? Name it.
Question four disqualifies more metrics than the other three combined. A number nobody acts on is not a metric, it is decoration, and it costs real attention to maintain.
Level 1, Counting
This is where everyone starts, and the mistake here is not tracking the wrong things. It is leaving too early.
Track new revenue booked, revenue lost, customer count, opportunity count by stage, and raw activity volume by channel. That is close to it. All of these are counts, all can be verified by hand, and none require your process to be mature.
The trap at Level 1 is the ratio. Ratios feel like real analytics, and a ratio computed on twelve data points is noise wearing a suit. If you closed three of thirteen deals last month your win rate is not 23%. Your win rate is unknown, and your sample is three.
You have earned Level 2 when your funnel stages have written definitions, your team follows them without being reminded, and you are closing enough deals per quarter that a percentage means something.
It is worth being precise about how few that is. At thirty closed opportunities and a 30% win rate, the 95% confidence interval around that rate runs roughly 14% to 46%. Thirty deals is the point where a win rate stops being noise, not the point where it becomes precise. Below thirty you are reading tea leaves, and the lumpier your deal sizes the longer that stays true.
What each metric actually requires
The table below is the short version of everything that follows. Every metric on it is defensible. The right-hand column is where they die.
| Metric | Level | What it actually needs | Fails when |
|---|---|---|---|
| New revenue booked | 1 | A closed-won date | Bookings and recognised revenue get mixed |
| Opportunity count by stage | 1 | Stages exist | Stages are subjective |
| Win rate | 2 | Losses are marked lost | Stale opportunities never resolve |
| Stage conversion | 2 | Exit criteria per stage | Reps batch-update stages |
| Lead to opportunity rate | 2 | One agreed lead definition | Marketing and sales define it differently |
| Sales cycle length | 3 | Trustworthy stage timestamps | Backdated or batched entry |
| CAC by channel | 3 | Fully loaded cost, source on the record | Salaries excluded, source overwritten on convert |
| CAC payback | 3 | Gross margin, not revenue | Revenue used in place of gross profit |
| Pipeline velocity | 3 | All four inputs trustworthy | Reported without its components |
| Net revenue retention | 3 | Clean expansion and contraction events | Renewals and upsells booked as new |
| Forecast accuracy | 4 | Forecasts recorded and kept | Nobody writes down the old forecast |
| Lead scoring performance | 4 | Precision and recall, not accuracy | Accuracy quoted against a low base rate |
| Multi-touch attribution | 4 | High deal volume, durable touch history | Volume too low, model fits noise |
Two metrics to stop reporting today
There is a longer list of these below. Two are worth acting on before you read another word.
Email open rates stopped being trustworthy in 2021, when Apple's Mail Privacy Protection began prefetching tracking pixels for Apple Mail users whether or not the message was ever read. Clicks and replies survived that change. Opens did not, and a rising open rate now tells you about your audience's mail client rather than your subject line.
Number of demos booked is a real metric that becomes a vanity metric the instant it becomes a quota, because demo count is the easiest number in the funnel to inflate.
And one metric almost nobody tracks that belongs at Level 1, because it costs nothing and predicts a great deal: median time from inbound form fill to first human contact. The long-running MIT and InsideSales work on this found the odds of qualifying a lead fall by roughly 80% after the first five minutes. The published cross-industry average response time is somewhere around 42 hours. Most teams have never measured their own, and the number is usually worse than the guess.
Get the full ladder
Enter your work email and we will unlock the rest right here: Levels 2 through 4 in full with the data prerequisite and failure mode for each, the complete list of advanced-looking metrics that are usually vanity, and a 30-day plan for installing the level you are actually ready for.
We will send the occasional useful thing and you can unsubscribe anytime. No spam.
Level 2, Converting
Counting tells you how much. Converting tells you where it stops, which is the first genuinely actionable thing a metric can do.
Track stage-to-stage conversion, overall win rate, average deal size by segment, and lead to opportunity rate. Now you can see that you are not short of pipeline, you are losing 70% of it between demo and proposal, which is a different problem with a different fix.
A word on what a win rate is supposed to look like, because this is the metric people benchmark themselves against most and understand least. Across roughly 655,000 opportunities analysed in 2025, the average B2B win rate came in around 19%, down sharply from the year before, and most healthy teams sit somewhere between 20% and 35%. Deal size moves it more than anything else: under $50K, 35% to 45% is normal, while above $100K it drops to 15% to 25% as the buying committee grows. If your reported win rate is well above the band for your deal size, the likeliest explanation is not that you are exceptional. It is that your losses are not being marked.
What this level demands is a funnel your team follows. Not a documented funnel. A followed one. Stages have to be written as exit criteria, meaning a condition someone else can check, rather than as feelings. "Interested" is not a stage. "Has confirmed budget and named an economic buyer" is a stage.
The failure mode here is subtle and common. When stages are subjective, reps batch their updates. A deal sits in stage two for six weeks and then moves through three, four and five in the same afternoon because the rep is tidying up before a forecast call. Your conversion rates still calculate. They are describing a data entry habit rather than a sales process, and no amount of dashboard work will reveal that.
Fixing this is field-level work rather than reporting work, and it is the same job as cleaning up the fields the metric depends on. Our CRM Data Hygiene Playbook covers the audit and the validation rules that make a stage field trustworthy in the first place.
Also at this level, decide what a loss is. Companies that do not mark losses cleanly develop a category of deal that is neither won nor lost nor open, and it grows until win rate becomes meaningless. If you have opportunities with close dates more than two quarters in the past and no outcome, you have this problem right now.
You have earned Level 3 when your stage timestamps reflect when things actually happened, your losses are marked with a reason from a picklist, and your marketing spend can be tied to something downstream of the click.
Level 3, Diagnosing
Here the questions become expensive to answer and worth answering. What does an acquired customer cost? How long does the cycle take, and where does the time go? Which segment actually pays back?
Track customer acquisition cost by channel and segment, sales cycle length by segment, pipeline velocity, CAC payback period, net revenue retention, and cost per qualified opportunity.
Two of those need care.
Pipeline velocity multiplies opportunity count by win rate by average deal size, then divides by cycle length. It is useful precisely because it is a compound metric: it moves when any of four inputs move, which makes it an excellent early warning and a terrible diagnosis. Never report it without the four components beside it.
CAC payback is the number most often computed wrong in early-stage SaaS, usually by taking revenue instead of gross profit, or by counting only paid media as acquisition cost. If your CAC excludes salaries it is not CAC, it is a media efficiency ratio, and calling it CAC will make your unit economics look roughly twice as good as they are.
The prerequisite for this whole level is a closed loop. Spend has to be attributable to source, source has to survive onto the opportunity, and the opportunity has to survive to revenue with the source still attached. Most CRMs lose this at the second step.
The failure mode at Level 3 is precision theatre. A CAC figure quoted to the dollar, built on a source field filled in 60% of the time, is not more accurate than a range. It is less honest than one.
You have earned Level 4 when you have at least four quarters of clean history, enough volume that a cohort is more than a handful of accounts, and a team that has already acted on Level 3 numbers at least once.
Level 4, Predicting
Prediction is where most of the interesting work is, and it is also where every unpaid debt from the levels below comes due at once.
Track forecast accuracy against actuals, cohort retention curves, propensity or lead scoring, multi-touch attribution, and expansion revenue by cohort.
The rule that governs this entire level: a model is a compression of your history. If the history is inconsistent the model learns the inconsistency and repeats it back to you with a confidence score attached, which is worse than having no model, because the confidence score buys it trust it has not earned.
Two specific cautions.
Model accuracy is not a meaningful number on its own. If 5% of your leads convert, a model that predicts "no" for every single lead scores 95% accurate and is worth nothing. Ask for precision and recall at the threshold you will actually use, or ask what it beats. Accuracy alone is a number people quote when the real numbers are unflattering.
Multi-touch attribution is oversold at every company size below several hundred deals a year. Below that volume the models are fitting noise, and the output tends to confirm whatever the person who configured it already believed. Single-touch attribution plus a self-reported "how did you hear about us" field is less sophisticated and considerably more truthful.
Forecast accuracy is the one metric at this level almost every company should track, and almost nobody does. Record the forecast, wait, compare it to what happened, and keep the history. It costs nothing, it needs no model, and it is the fastest way to find out whether any of the rest of this is working.
Metrics that look advanced and usually are not
Some numbers signal maturity without producing decisions. They are not always wrong, but they earn their place far less often than they appear on dashboards. Open rates and demo counts are covered above. Four more.
Marketing qualified leads as a headline metric rewards volume over fit, and the moment it becomes a target the definition starts drifting downward. This is Goodhart's law arriving on a dashboard: a measure that becomes a target stops being a good measure.
Website sessions and follower counts move for reasons that have nothing to do with revenue, and no one has ever changed a go-to-market decision because sessions rose 8%.
Average response time, unless it is tied to an outcome. Fast routing to the wrong owner is still the wrong owner.
Any metric reported without a denominator or a comparison. "We generated 340 leads" answers nothing on its own, and the absence of the denominator is usually deliberate.
Installing the right level in 30 days
Week one, find your level honestly
Run the readiness test against every metric currently on a dashboard. Most teams discover they are reporting Level 3 metrics on Level 1 data, which explains why nobody trusts the dashboard.
Week two, fix the prerequisite rather than the metric
If win rate is unreliable, the work is resolving stale opportunities and adding a loss reason, not rebuilding the report. The report was never the problem.
Week three, cut
Delete every metric that failed question four of the readiness test. A dashboard with six trustworthy numbers beats one with twenty-five contested ones, and the deletion is what makes the remaining six get looked at.
Week four, add one thing
Add exactly one metric from the next level up and give it an owner by name. One per month is a realistic pace. Teams that add six at once end up trusting none of them.
The mindset that makes it stick: a dashboard is not a trophy cabinet. Every number on it should be one somebody would defend in a room, and anything nobody would defend should come off.
Want us to run this with you?
The fastest version of this exercise takes about a week with someone who has done it before. Take the free RevOps Health Check for a scored read on where you are, or book a call and we will run the readiness test against your live dashboard.
Take the RevOps Health Check Book a Free Assessment