Which metrics should a startup track before product-market fit?

A startup releases its first version, attracts five hundred visits, gains one hundred registrations and receives twenty positive comments on social media. The dashboard looks active, the numbers are rising and the team feels that the project is gaining momentum. Yet only five people complete the central action, two return the following week and nobody pays.

Is the startup making progress? That depends on what it was trying to prove. If the objective was to determine whether a campaign could attract curiosity, the visits provide information. If the objective was to learn whether the product solved an important problem, those numbers say very little.

Before product-market fit, the purpose of metrics is not to demonstrate growth. It is to reduce uncertainty. The startup is still trying to understand who needs the product, which value they recognise, what behaviour shows they have experienced it and why some people continue while others disappear.

This changes how the company should measure. It does not need a dashboard containing twenty indicators moving at once. It needs a small number of metrics connected to questions capable of changing the product, customer segment, price or next experiment.

A useful metric is not the one that creates the most impressive chart. It is the one that supports a decision.

Before product-market fit, the objective is learning rather than growth

A company that has discovered a repeatable pattern can concentrate on acquiring more customers, improving efficiency and expanding what already works. A startup searching for product-market fit is in a different position. It does not yet know with enough confidence whether the product solves an important problem, whether it has selected the right segment or whether the observed behaviour will repeat.

At this stage, measuring growth without measuring its quality can be misleading. Advertising may increase registrations without showing whether anybody receives value. Revenue may double because of two exceptional customers without revealing a proposition that can be sold repeatedly.

Separating learning from growth does not mean ignoring users, revenue or acquisition. It means interpreting every number according to the question it should answer. Early data helps uncover patterns; it should not be used to project a growth curve that does not yet exist.

The work before product-market fit consists of testing a chain of assumptions: that the customer recognises the problem, understands the proposition, reaches the value, decides to return and accepts some cost in order to continue receiving it. The metrics should observe that chain instead of measuring activity around the product without knowing whether it leads anywhere.

A metric is only useful when it is connected to a decision

Before choosing an indicator, ask what you will do if the result is high, low or ambiguous. If none of those outcomes changes a decision, you probably do not need to measure it yet.

Imagine a startup recording the number of pages viewed inside its application. The number may increase because users are exploring many useful features, but it may also increase because they cannot find what they need and move through several screens before leaving. Without an expected behaviour and a related decision, the same data supports opposite interpretations.

Measuring how many users complete the core task during their first session provides a more focused starting point. If few succeed, the problem may lie in the promise, onboarding, complexity of the workflow or audience selection. The metric does not provide the full explanation, but it identifies where further investigation is worthwhile.

The sequence should always begin with a hypothesis. The startup then defines the behaviour that would make the hypothesis observable, the data representing that behaviour and the decision that follows from the result.

The metric does not replace judgement. It makes judgement visible.

Activation shows whether the user reaches the first value

A registration does not mean that a user has been activated. Somebody may create an account out of curiosity, follow a recommendation or respond to an attractive landing page and leave without experiencing the product.

Activation happens when the user completes the action that allows them to perceive the main value for the first time. That action depends on the product. In invoicing software, it may be creating and sending the first invoice. In a learning application, it may be completing a lesson. In a marketplace, it may be submitting an initial request and receiving a valid response.

The activation rate can be calculated as follows:

Activation rate = users completing the activation action ÷ new users beginning the process × 100

The calculation is simple. Choosing an action that represents value rather than administration is more difficult. Confirming an email address, completing a profile or accepting terms may be necessary, but these steps do not necessarily show that the person understands why the product is useful.

The denominator also matters. Including every website visitor mixes acquisition with product performance. Including only people who have already completed several steps may hide part of the friction. The population should represent the journey the company genuinely wants to evaluate.

When activation is low, adding features is rarely the first answer. The startup should observe where users stop, what they expected to find and how much effort is required to reach the result. The problem may be a confusing experience, but it may also be a promise attracting people with a different need.

Time to value measures how long the promise takes to become real

Activation shows whether the user reaches the first value. Time to value shows how long it takes. The two metrics are related, but they answer different questions.

A product may achieve a reasonable activation rate while requiring too much time, configuration or personal support. Highly motivated early users may accept that effort. The wider market may not.

Time to value begins when the relevant experience starts and ends when the user obtains the first outcome that justifies beginning. In a simple application, this may take minutes. In B2B software requiring data imports and process configuration, it may take several days.

There is no universally correct duration. It should be considered against the urgency of the problem, the effort the customer expects to make and the available alternatives. If a spreadsheet solves the need in ten minutes, a product requiring three hours of configuration must provide a sufficiently important improvement to justify that effort.

Reducing time to value does not mean eliminating every configuration step. It means distinguishing between the steps necessary to create value and those that exist because the product still transfers its complexity to the customer.

Early retention shows whether the value is worth repeating

Activation can happen once. Retention shows whether the user returns when the need appears again. For that reason, it is usually a stronger signal of fit than registration volume.

A startup may gain many sign-ups through a campaign, a viral publication or a waiting list collected over several months. If most users try the product once and never return, growth at the top of the funnel conceals a constant loss beneath it.

Retention should be measured through cohorts. A cohort groups users who began during a similar period and shows what proportion remains active after a selected interval.

Cohort retention = users from the cohort returning during the selected period ÷ users who began in that cohort × 100

The interval must reflect the natural product cycle. A messaging application may expect daily use. Invoicing software may need weekly or monthly analysis. A platform for preparing annual tax filings cannot be interpreted using the same frequency as a social network.

Measuring retention does not mean forcing every product to be used daily. It means learning whether people return when they would reasonably be expected to need the value again.

At a very early stage, a cohort may contain only a few people. The percentage will be unstable, but the individual cases remain useful. The startup should examine who returned, which problem they had, what they did and how they differed from those who left.

Usage frequency should match the frequency of the problem

Frequent use may indicate habit, dependency or recurring value. It may also be irrelevant when the problem does not occur frequently.

The mistake occurs when a startup adopts measurements such as daily or monthly active users because they are common elsewhere without determining whether they fit its own product. A tool used once per quarter may deliver exceptional value. Measuring it daily will only create a needlessly discouraging chart.

The correct frequency comes from the problem. How often does it appear? When does the customer need to act? Which behaviour would be reasonable if the product were working well?

The company should also distinguish between opening the application and completing a valuable action. A user may return repeatedly because the system sends notifications without actually accomplishing anything. Another may enter once a month, complete a critical task and obtain considerable economic value.

The question is not how often somebody opens the product. It is how regularly the product helps solve the problem for which it was chosen.

Depth of use distinguishes activity from value

Two active users may obtain entirely different outcomes. One logs in, reviews a screen and leaves. The other completes the main process, saves information, invites a colleague and uses the result in their work.

Depth of use examines how much of the product becomes part of the customer's behaviour. It does not mean counting as many clicks as possible. It means identifying actions that increase the likelihood that value has been created.

In a collaborative product, inviting a colleague may be significant because it suggests the tool is entering a shared process. In an analytics platform, saving, reusing or exporting a report may represent more value than visiting five unrelated sections.

These actions still require careful interpretation. A heavily used feature is not necessarily valuable if people use it because the rest of the workflow is confusing. Data shows what happens; interviews and observation help explain why.

Paid conversion is a signal, but it needs context

Payment introduces a commitment that a free registration does not contain. When somebody hands over money, accepts a price and gives up alternative uses for that money, they provide a stronger signal than somebody who merely says the product sounds interesting.

Paid conversion can be calculated against everybody who tried the product, users who reached activation or qualified commercial opportunities. Each denominator answers a different question.

Paid conversion = paying users ÷ users eligible to purchase × 100

Defining who was genuinely eligible prevents misleading conclusions. If a waiting list includes people in countries where the service is unavailable, including them distorts the result. The same applies when people attracted by a free offer are mixed with customers actively looking to solve the problem.

One early payment does not prove product-market fit. It may result from a personal relationship, an exceptional discount or additional work delivered directly by the founder. The useful learning comes from understanding why that person bought, what they expected, whether they received it and whether they would continue under normal conditions.

When there are few customers, the story behind every conversion matters as much as the percentage.

Signal quality matters more than signal volume

Not every action represents the same level of commitment. A positive interview response, an email submitted through a landing page, a trial, repeated use and a payment contain very different amounts of evidence.

A startup can order signals according to the cost accepted by the user. Saying that an idea sounds good requires almost nothing. Sharing real information, spending time configuring the product, inviting a colleague, paying or changing an existing process requires progressively greater commitment.

This helps interpret small datasets. Ten generic registrations may provide less evidence than three users from the intended segment who complete the workflow, return and ask to continue using the product.

Quality also depends on alignment with the ideal customer. A startup may attract substantial activity from students interested in a free version while intending to sell to companies. That activity may teach the team about usability, but it may not validate the chosen market's problem, price or purchasing process.

Before product-market fit, a signal is valuable when it comes from the right person, occurs in the right context and requires an action consistent with the hypothesis being tested.

Acquisition metrics still have a purpose

Visits, impressions, reach, clicks and registrations are often described as vanity metrics. They are not inherently useless. They become vanity metrics when they are presented as evidence of progress without being connected to subsequent behaviour.

Acquisition can reveal whether the segment is reachable, whether the message attracts attention and which channels bring people with a greater probability of activating. What it cannot establish on its own is whether the product solves the problem.

A campaign producing many clicks but little activation may indicate an attractive promise the product does not fulfil, incorrect targeting or an overly complicated initial experience. A campaign with few clicks and strong retention may mean that the product creates value for those who arrive, but the message or channel cannot yet find them efficiently.

Acquisition data becomes useful when connected to activation, retention and payment. Without that connection, increasing traffic may simply fill a product that users leave more quickly.

What turns a metric into a vanity metric

A vanity metric increases the appearance of progress without improving the company's understanding. It is often cumulative, lacks context or combines users with very different behaviours.

Total registrations are a common example. The number almost always increases because it includes everybody who has ever created an account, even when most have never returned. The same problem affects downloads, followers, impressions and historical transaction volume.

These numbers can be useful for particular questions, but they should not obscure current behaviour. One thousand registered accounts do not mean one thousand users. One hundred interested businesses do not mean one hundred customers. A waiting list is not a sales pipeline.

A simple way to identify a vanity metric is to ask which decision it supports and which behaviour it represents. If the only interpretation is “more is better”, important context is probably missing.

Why a dashboard with twenty metrics can reduce understanding

Measuring more does not guarantee learning more. At an early stage, a large dashboard can create an illusion of control while dividing attention among indicators whose relationship with the business remains unclear.

When one metric falls, another rises and a third remains stable, it is always possible to construct a favourable explanation. The team ends up commenting on charts instead of deciding which hypothesis has gained or lost credibility.

Every indicator also needs a definition. The company must decide what counts as an active user, when a conversion starts, how internal testing is excluded and what happens when an action repeats. If twenty metrics are implemented without agreeing on these rules, the dashboard may display data precisely while nobody interprets it consistently.

Before product-market fit, it is usually more useful to work with one primary metric related to the current uncertainty and a small number of supporting measures. The primary metric can change when the question changes.

If nobody completes onboarding, activation becomes the priority. Once most people complete it, attention may move to retention. If users return but nobody buys, the next learning concerns price, perceived value or the conversion process.

Cohorts reveal what averages conceal

Averages combine users acquired at different times, through different channels and across different versions of the product. That mixture can hide both improvement and deterioration.

If a startup changes onboarding and average retention remains stable, it may conclude that the change had no effect. The new cohort may actually be retaining better while older users continue to pull down the average.

Cohorts make it possible to compare groups sharing a relevant condition: registration date, segment, channel, use case, plan or product version. There should not be so many segments that every group contains one person, but there should be enough to separate different hypotheses.

This distinction is particularly important before product-market fit. The product may retain poorly overall but perform well within a particular customer type. That concentration may reveal the segment deserving closer attention.

Small datasets are not useless, but they require a different interpretation

A startup with eight users cannot interpret a difference of one person as a stable statistical trend. Moving from two retained users to four doubles the percentage, but the sample remains too small for reliable projections.

That does not mean the company must wait for hundreds of users before learning. It means combining the numbers with individual cases. Who are the four people who returned? Which problem do they share? What did they do differently? Why did the other four leave?

At this stage, metrics help locate important conversations. Analytics reveals where a difference exists, and qualitative research helps explain it.

Interviews should begin with observed behaviour rather than abstract questions about whether somebody likes the product. How to run customer interviews that yield real insights explains how to reconstruct past decisions and behaviour without leading the participant.

Different business models require different interpretations

Basic metrics may share a name without representing the same behaviour. In SaaS, activation may mean completing a configuration and receiving the first result. In a marketplace, value does not appear when one side registers but when supply and demand complete a useful interaction.

In a B2B product, the activated user may not be the buyer. The company will need to measure how people use the tool and, separately, how the commercial process, approval and account conversion progress.

For a service addressing an occasional need, retention may appear as repeat purchasing when the problem returns. In a subscription product, it also matters whether the customer continues paying even when use is not daily.

Copying another startup's dashboard can introduce familiar metrics that do not represent the value of your own business. The definition should begin with how the customer receives value and how the company captures it.

A simple measurement system before product-market fit

Measurement can begin in a straightforward spreadsheet. A complex infrastructure is unnecessary when there are few users and the process can still be reviewed manually. What matters is maintaining consistent definitions and enough context to interpret each case.

For every priority hypothesis, record:

• The assumption being tested.
• The behaviour that would suggest it may be true.
• The metric representing that behaviour.
• The people or accounts included.
• The period over which it will be observed.
• The result that would lead to keeping, changing or rejecting the hypothesis.

For example, a startup may propose that small professional firms receive value when they generate their first report within three days of registration. It measures the percentage that succeeds, the time required and how many return to create another report during the month.

If many begin but few finish, the team should review the workflow or its complexity. If users finish but do not return, the startup needs to investigate whether the result was useful or the need occurs less frequently. If they return but do not pay, the question moves towards price, the buyer or the distinction between the free and paid proposition.

The primary metric changes as the learning progresses. That is more useful than preserving a dashboard designed when the product was answering different questions.

Which signals begin to suggest product-market fit?

Product-market fit is not proved by one number and does not appear on the day a chart crosses a universal threshold. It becomes visible when several signals begin to reinforce each other.

The right users reach the value without requiring constant explanation. A meaningful proportion returns at a frequency consistent with the problem. Some pay under normal conditions, continue using the product and resist the idea of losing it. Recommendations, expansion requests and adoption inside real processes begin to appear without every action needing to be pushed manually.

The effort required to keep each account active also starts to decrease. The founder continues speaking with customers but no longer rebuilds the product around every individual case. A pattern begins to appear across who buys, why they buy, what they use and which result they obtain.

None of these signals is sufficient alone. Strong retention among five friends does not demonstrate a market. Numerous payments followed by immediate cancellation do not either. The evidence becomes stronger when activation, repeat usage, payment and satisfaction appear within the same segment and across several cohorts.

How Foundeia approaches measurement before product-market fit

Foundeia does not treat metrics as a checklist that every startup must complete. It connects them with the stage, hypotheses and unresolved decisions of each project.

Before launch, the priority may be to define which behaviour will demonstrate that the MVP solves the problem. After launch, the data should be compared with that expectation and with what appears through interviews, sales and support.

Early users are particularly valuable because they connect behaviour with context. How to find your first customers as a startup explains why this stage is not simply about creating volume but about finding people from whom the company can learn.

Before product-market fit, measuring well does not mean possessing more data. It means knowing which uncertainty you are attempting to reduce and recognising which signals deserve greater confidence.

Visits may indicate that a message attracts attention. Activation shows that somebody reached the first value. Retention indicates that the value deserves to be repeated. Payment demonstrates an additional level of commitment. When those signals come from the same customer type and begin to repeat without exceptional explanations, the startup stops accumulating activity and begins building evidence.

Frequently asked questions