Product-led sales: Nobody asked you to call them

The usual framing is that product-qualified leads are marketing-qualified leads with better data attached. Same motion, higher hit rate.

That framing is how a PQL program disappoints, and the difference it misses is not a matter of degree. A marketing lead raised their hand. They downloaded something, filled in a form, asked to be contacted. A product-qualified lead did not. They were using your product, getting on with their work, and now somebody is going to interrupt them.

An MQL can be sold to, because they asked. A PQL has to be helped with the specific thing they were visibly doing, or the contact reads as surveillance.

What is a product-qualified lead?

A product-qualified lead is an account whose behavior in the product indicates it is ready to buy more, or to buy at all. Not activity: readiness. The distinction matters because activity is abundant and readiness is rare.

Practitioners consistently report PQLs converting several times better than marketing leads, and the direction is not really in dispute.¹ The multiple is, because it depends entirely on how loosely each company defines an MQL. A company whose MQL bar is “visited the pricing page” will report a spectacular ratio, and it means nothing.

Activity is not intent

The most common way to build a bad PQL model is to score activity. Logins, sessions, feature counts, seats: all easy to instrument, all mostly measuring how large and how busy an account already is.

The signals that predict a purchase are the ones that only occur when somebody is trying to do something they cannot currently do:

They hit a boundary. A limit reached, a feature attempted, a wall met. This is the strongest class, because the user has just discovered a need themselves, and the prompt they saw is a record of exactly what they wanted.

They brought other people. Invitations, especially to people senior to them, or a second team appearing in the account. Somebody spending their credibility on you internally is a stronger signal than any amount of their own usage.

They asked a question only a buyer asks. Anyone looking at your security page, SSO documentation, data-processing terms or invoicing options has stopped evaluating the product and started working out whether their company can approve it. That is the clearest buying signal in bottom-up adoption and the easiest to instrument.

They built something durable. Configuration, integrations, accumulated data they would not want to abandon. Investment predicts commitment.

One non-behavioral dimension belongs alongside those four: whether the account could ever pay. Company size, role, industry. A behavior-only model reliably surfaces enthusiastic individuals at organizations with no budget, and the strongest scoring combines what somebody did with whether their employer is a plausible customer.

It helps to see what companies have actually used, because the shape is consistent and unglamorous: a team crossing a message threshold, a user syncing files from a second device, an account creating a handful of documents and inviting collaborators, someone scheduling several meetings and connecting a calendar within two weeks. Volume of ordinary use, plus somebody else arriving.²

Deriving your own signals

The classes above are a starting point, not an answer. Your own conversion history is better, and the procedure is short:

Pull everyone who started paying in the last quarter. Look at what they did in the week or two before they paid, not their whole history. Find the actions that appear in most of those accounts. Then check each against accounts that churned or went quiet, because an action common to both groups is measuring engagement, not intent.

What comes out is a ranked list of candidate signals particular to your product, which is worth more than any general list including this one.

The threshold is a capacity decision

Everyone starts by looking for the PQL threshold: some natural break in the numbers where accounts become worth calling. Go and look for it, because the search is what convinces you. There is rarely a break. Readiness is a continuum, propensity rises gradually, and any line you draw is arbitrary in the sense that matters.

So the threshold is not a discovery, it is a decision about capacity. You have some number of people who can make contact this week; the threshold is set wherever that capacity meets the volume of accounts above it. It moves when you hire, and it should.

The calibration has two failure directions, and teams only watch one. A hit rate that is too low means the bar is loose and you are interrupting people who were not ready. A hit rate that is very high means the opposite: the bar is so tight you are only contacting accounts that would have bought anyway, and the outreach is claiming credit, not creating revenue.

That reframing has three practical consequences. Score accounts by propensity and work down the ranked list instead of defining a binary qualified-or-not. Expect the bar to move, so do not encode it as a hard rule in five systems. And when someone asks to lower the threshold to generate more pipeline, understand that you are agreeing to contact people with weaker signals, which will reduce the hit rate: that is a trade, not an improvement.

What product-led sales actually is

Product-led sales means the product does the qualifying and a human does the part a product cannot: answering questions from people who were not in the product, and negotiating with people who require negotiation.

It is not a retreat from product-led growth. Buyers want both. McKinsey’s work on the shift to product-led sales reports that 65% of software buyers prefer having self-serve access and a salesperson available, rather than choosing between them.³ The choice most companies agonize over was never really on offer.

What changes is the sequence. Traditional sales qualifies with a conversation and then demonstrates the product. Product-led sales inverts it: the product demonstrates itself, and the conversation happens once somebody has already decided they like it. That is a better conversation, and it needs a differently-equipped person on your side of it, which is the part companies underestimate.

The most common failure is organizational rather than technical. A sales team measured on new logos and a growth team measured on self-serve revenue will fight over the same accounts and route around each other. Putting both on the same targets is not an administrative nicety; it decides whether the motion works.

When to add the human

Not at a revenue milestone. The signals are qualitative and they are visible before the revenue is:

• Accounts are converting to your top tier without help, which means the tier above it is missing.
• Deals stall in security or procurement review, which is not something a product can answer.
• Users are asking for invoicing, contracts, or terms your checkout does not offer.
• People with authority are appearing in accounts your product never marketed to.

Any of those means the ceiling is administrative, not product-related, and no amount of onboarding work moves it. The reverse is also worth stating: adding sales before those signals appear buys you cost, and a team with nothing qualified to work will start qualifying things badly.

Which numbers tell you it is working

1. Conversion rate per signal, ranked. Not an average. The point of scoring is to learn which behaviors matter, and the distribution is always more skewed than expected.
2. Time from signal to contact. This decays fast. A limit hit on Tuesday and contacted the following week has lost most of its value, because the intent it recorded has been resolved some other way.
3. What share of self-serve revenue you interrupted. Some accounts you call would have bought anyway. If you are not measuring against a holdout, you cannot tell whether sales is adding revenue or claiming it.
4. Hit rate against the threshold. When you lower the bar, watch the conversion rate fall and decide whether the added volume was worth it.

Adoption figures for PQL scoring are published and not used here, along with the multiples usually attached to PQL-versus-MQL conversion.⁴ The comparison that means something is your own: PQL-sourced conversion against your other sources, with a holdout.

Getting this running

1. List the four signal classes above and check which you can actually instrument today. The security-page visit is usually the cheapest to instrument.
2. Score, do not gate. A workable first pass: give each signal a weight reflecting how strongly it predicted conversion in your own history, sum them per account, and sort. Hitting a paid limit might be worth several times a login streak. Work down the ranked list instead of defining a line and calling everything above it qualified.
3. Set the threshold from your capacity, write down that this is what you did, and revisit it when the team changes size.
4. Measure time from signal to contact and fix the worst delay before adding signals.
5. Hold out a random slice of qualified accounts from outreach for a quarter. It is the only way to know what sales contributed.
6. Put growth and sales on shared targets before the first argument rather than after it.

Everything in this chapter depends on numbers you can trust: which action predicts retention, what an account is worth, whether expansion is real. Those numbers are also the easiest place in product-led growth to fool yourself, because most of the metrics in circulation are either uncomparable between companies or quietly measuring something other than what they claim. The metrics chapter is about which ones to keep.

Updated August 2026