Measuring what matters: the few numbers that should change what you do

You open your analytics and find 40,000 impressions, 611 website visits, 18 sign-ups, three trial users and one payment. The 40,000 feels GREAT - but is this the number to care about the most?

The payment is actually closer to the thing you care about - tangible cash - although the smaller numbers around it explain how somebody reached that point. And that's because our metrics, no matter how small they appear, all work together to create the story of our customer and our growth.

The decision

Early measurement should begin with one commercial objective: what are you trying to learn or improve over the next few weeks? The objective decides which customer action deserves the most attention that month or that quarter.

Revenue belongs in the plan from the beginning, because every route eventually leads to the money. It may be too rare to guide your next weekly decision when you're pre-revenue or working with your first five users, so in these cases a qualified conversation or completed first customer result can give you an earlier read.

One main measurement and three to five supporting figures are enough at this time, because if there's one thing a solo founder is good at, it's building a dashboard the way an anxious parent packs for a day trip: six contingency outfits, three kinds of wipes and no idea how there's chocolate all over the car.

Three different facts

The previous chapters asked you to record your activity and what customers did afterwards. Keep those facts separate when you review them.

A high activity number can sit beside a weak customer response: the gap between the two might send you back to the audience, message, or channel, and the gap alone won't tell you exactly what's happened.

Attention figures deserve the same restraint; views, impressions, followers, and email opens can describe how many people had a chance to notice you, and they become more informative when you can see the next action.

The customer action needs a definition

You need a short route from first attention to the result your customer came for. It won't describe every real buying journey, especially when somebody reads an article, sees you on LinkedIn and later arrives through a colleague. It will give you a consistent set of events to count.

A sign-up proves that an account exists. Your activationThe first useful result a customer receives from the product or service. Creating an account is a sign-up; activation happens when the customer reaches the outcome the product was built to provide. event should describe what the customer achieved after that. For a reporting product, it could be sharing the first client report with a colleague. For a booking product, it could be receiving the first completed booking. The product determines the event.

What the number means

A label such as “conversion rateThe share of people or accounts that moved from one defined action to another during a stated period. The starting action, ending action, count and denominator all belong beside the rate.” leaves plenty of room for interpretation. It could mean website visitor to sign-up, trial to payment or sales call to signed contract. The definition needs the context of the result:

The definition needs to stay stable while you compare results. If “qualified conversation” means a booked call one week and a buyer with confirmed need and budget the next, the trend belongs to two different measures.

A baselineThe result recorded before you set a target or change the activity. It gives you something real to compare with later results. should come before a firm target. Two to four weeks may be enough for a fast activity such as outreach. A considered sale needs a complete buying cycle; otherwise, the target is being judged before the customer could reasonably decide.

Small numbers need their clothes on

Suppose two of your three trial users activate. The calculation is 66.7%, and the decimal point makes it look as though several clever people in glasses were involved. Write two of three trial users (66.7%). The count shows how easily the rate could change when a person who prefers words over numbers arrives.

Early rates are provisional, but they still help you spot where people stop, as long as the count and denominatorThe total number of eligible people or accounts used to calculate a rate. Showing it beside an early percentage reveals how much one additional result could change the figure. remain visible. You don't need to wait for a statistically impressive sample before learning that none of five trial users completed setup; you do need to avoid presenting that result as a permanent conversion rate.

For an external benchmarkA result from another company, dataset or period used as a point of comparison. It is only useful when the customer, product, price, action, channel maturity and sample are similar enough to yours. to guide your target, the comparison should cover:

• whether it covers a similar customer and product;
• whether the price and buying process are comparable;
• whether it counts the same action;
• how mature the channel was; and
• how many people or accounts were measured.

A benchmark with several mismatches can still give you a question to investigate. It shouldn't become your target.

Each channel needs enough time

Cold outreach can produce replies within days, while search content and Pinterest can take longer to build distribution. Their reporting windows should reflect those facts. This prevents a new compounding channel from being declared a failure after one week, and it also prevents a fast channel from receiving months of activity without a relevant response. Record when the test began, what customer action you expected and when you agreed to review it. Set reminders so that timescales don't get out of hand.

The later chapter on deciding what to continue, change or stop will handle the verdict - this chapter makes sure the verdict has something more substantial than vibes underneath it.

Attribution can contain an unknown

A customer might first discover you on Reddit, join your email list and pay after a referral from a colleague. Attributing the whole process to the last email alone removes most of the route that helped them decide.

An early system only needs:

• First known sourceThe earliest recorded route through which somebody found you. Use `Unknown` when the source cannot be supported by the record.: where they first found you, when you know it.
• Latest promptThe most recent message, conversation or recommendation behind a customer action. It can differ from the channel through which the customer first found you.: the message, conversation or recommendation behind their latest action.
• Unknown: use this when the route can't be supported by the record.

This is enough to show repeated routes without inventing exact credit for every interaction. A direct question about how somebody found you can improve the record, although some journeys will stay incomplete.

Your time belongs in the result

Cash spend and founder time affect different decisions in different ways, so record them separately. A community test might cost $0 and take eight hours. A search ad test might cost $300 and take one hour. Combining both into one figure requires an hourly rate and several assumptions that may distract from the comparison.

Founder time is often entered as $0, which makes every marketing experiment profitable and every weekend unavailable. Customer acquisition cost (CAC)The cost of gaining a new paying customer under a stated cost definition. It becomes more informative when the definition is stable and there are enough customers for one extra sale not to transform the result. becomes more informative once you have stable cost definitions and enough customers for the result to survive one extra sale. Until then, keep these facts visible:

• cash spent;
• your hours used; and
• new paying customers attributable with reasonable confidence.

Revenue should mean money collected when cash timing affects the decision. Keep invoices sent and payments received as separate figures, particularly for services or longer B2B sales.

The smallest stack you can maintain

Your measurement plan can begin in a spreadsheet or Notion database. Add software only at the point you outgrow it.

Plausible supports page and custom-event goals, including sign-ups and purchases. Google Analytics records site and app interactions as events. PostHog is designed for product events and funnels. Those capabilities only help once you decide which event counts.

A prettier dashboard doesn't justify rebuilding your stack: a plain table you update beats an immaculate control room whose “activation” figure stopped working during the last product release.

RelayReport's first scorecard

RelayReport is a fictional reporting product for small marketing agencies. Its current objective is to learn whether trial users create and share a client report, which is the first result the product was built to provide.

One paid account can't prove a repeatable channel. It can provide a customer to learn from and a route to reconstruct. The next question is why two trial accounts shared a report and two stopped earlier.

Your measurement plan

Use the measurement plan with your activity log, customer records, analytics and payment information.

I am preparing my go-to-market measurement plan. Use only the information I
supply.

Don't connect accounts, change a live system or invent missing figures. Don't
infer why a result changed.

1. State my current commercial objective in ordinary language.
2. Identify the customer action that would show progress towards that objective.
3. Separate my information into activity, customer movement and commercial
   results.
4. Recommend one main measure and no more than five supporting measures.
5. For each measure, write its definition, source, period, update frequency,
   baseline and the decision it can inform.
6. Keep counts and denominators visible beside every rate. Mark small-sample
   rates as provisional.
7. Separate reported results from targets and planning assumptions.
8. Record first known source, latest prompt and unknown attribution separately.
9. Show cash spend and founder hours in separate fields.
10. Add every missing definition, broken source or other data gap as a caveat.
11. Prepare a review with three sections: What happened, What remains unknown,
    and What needs investigation. Don't assign a cause without evidence.

Return standard Markdown in my first person. Stop with the plan and blank
scorecard ready for my review.

You're ready to move on when the scorecard has one current objective, one main measure and enough supporting figures to explain what happened. The next chapter covers how to collect customer feedback and work out why.