How to convert more free users to paid in PLG, and when to say no

Free users convert more when they aren't left alone, and when you know who isn't ready yet. The PLG (product-led growth) companies that convert best onboard each user around the problem they came to solve, follow up with those who stall, keep support open to free accounts and hold the sales pitch until the moment is right.

A free user who never finds value doesn't stay free for long. They go looking somewhere else.

What follows comes from Zentesimal's work on onboarding, follow-ups and support across thousands of self-serve accounts: the tactics that move conversion, and the numbers behind each one.

Why do most free users never pay?

In most PLG products, nobody talks to them. Free users are treated as a cost to keep low, so they get a product tour, an email sequence and a help centre.

The result is well documented. Vineet Kumar, a Harvard Business School professor who studied freemium models (a free plan with paid upgrades), found that most companies convert between 2% and 5% of free users, as he wrote in Harvard Business Review.

The other 95% leave without saying why.

Freemium is designed to let the product sell itself. That works for users who find the value on their own. Everyone else disappears quietly, and the company only sees a number going down in a dashboard.

AI changes the cost of that conversation. Talking to every free user used to need a team nobody would fund. Today an AI agent can do it for every account, every day, and every tactic below builds on that.

Why does onboarding built around the user convert better?

Because the user only cares about the problem they came with. In internal Zentesimal data across more than 4,000 accounts, onboarding focused on the user's problem got up to 8 times more users to finish onboarding than onboarding focused on showing what the product offers.

In self-serve products, finishing onboarding is the step that makes conversion possible. A user who never gets there rarely pays.

B2B teams have worked this way for years. On a kickoff call, a CSM (customer success manager) asks what the customer wants to achieve and shows the three features that get them there. Self-serve onboarding skipped that step because asking every user and adapting the path didn't scale, so most products settled for the same tour for everyone.

As an example, picture a user who signs up for a project management tool and, asked what brought them there, writes: "I spend every Monday chasing my team for status updates."

A product tour walks them through boards, timelines, automations and integrations. A user-centred onboarding shows them recurring check-ins, helps them set up the first one for next Monday, and stops there.

Boards and timelines can wait until the check-ins are working.

Usability research points the same way. In an analysis of app onboarding, the Nielsen Norman Group, a user experience research firm, found that generic tutorials didn't help people complete tasks, and kept onboarding for the cases where it tailors the experience to the user.

Zentesimal does this at scale with AI agents that match what each user says they need with the product features that solve it, and walk them through using those features step by step. Expanding to the rest of the product comes later, once the first problem is solved. The conversion use case shows how it works.

Nobody signs up to learn a product. They sign up to fix something.

What happens to users who don't finish onboarding?

Many of them were never going to be customers: students, people comparing tools for a blog post, accounts created by mistake. Around 44% of the stalled onboardings Zentesimal has examined belong to a different group, users who didn't find the value and went to try a competitor.

That 44% is Zentesimal's estimate, built from reviewing stalled accounts and the replies users gave to follow-up messages.

You're losing a customer and handing one to a competitor at the same time.

Public data points the same way. In a 2020 survey of 216 people by Wyzowl, the video marketing company, 80% said they had deleted an app because they didn't know how to use it.

The good news is that a lot of these users can be brought back. With the proactive follow-ups Zentesimal runs, up to 30% of incomplete onboardings end up completed (internal Zentesimal data across self-serve accounts in production, 2026).

What makes those follow-ups work is precision. The message picks up from the exact step where the user stopped, with the problem they described at sign-up in mind.

As an example, a user who connected their data but never built a report gets a message about that report and the one step they missed. A generic "we miss you" email gets ignored. A message about their own unfinished report gets opened.

Should free users get support?

Yes. A surprising number of companies cut support for free accounts to save resources, and that saving has a price: you know nothing about those users, and they get no help when they get stuck.

Metricool, the social media management platform, kept support open to non-paying users. Close to half of its chat conversations came from free accounts (Metricool support data).

When Metricool brought in an AI assistant, converting those users became much easier, because they were already used to asking questions in the chat.

The channel already existed. The AI added the ability to spot who was ready to buy and help them get there. The full story is in the Metricool success story.

The easiest free user to convert is the one already talking to you.

This changes when the free base is huge and the product is simple, like consumer apps with millions of users. There, an AI assistant is the only kind of support for free accounts that makes financial sense, since a human team would never pay for itself.

Can the support chatbot spot who is ready to buy?

Yes, and it's often the best placed to do it, because it talks to more users than anyone else in the company, free users included.

A user asking about pricing, visiting the pricing page, asking about features from a higher plan, asking for a discount or sitting at the limit of their plan is showing buying intent. A chatbot that reads those signals can start the conversation instead of waiting for a ticket.

It also needs to know when to stay quiet, which the next section covers in detail.

In the accounts where Zentesimal runs the support assistant, selling this way brings in an average of $45k a month in new and expansion ARR (annual recurring revenue), against less than $2k in the control group (internal Zentesimal figures across accounts in production, 2026).

The signals, the timing and the cases where the chatbot holds back are covered in how to use AI chatbots to sell more.

When should you say no to a free user?

When their behaviour doesn't look like a future customer's, or when the moment is wrong. Saying no keeps follow-ups and offers where there's business, and spares the people who were never going to pay from messages they'll never read.

Users who don't look like customers

Some signals that usually point to a real prospect:

• They invite a teammate.
• They connect an integration or import their own data.
• They come back several weeks in a row.
• They sign up with a company email from the segment the product sells to.

As an example, two users sign up on the same day. One uses a personal email, opens the product once and never sets anything up. The other connects their company's data on day one and invites two colleagues by day three.

Both are free users, but only one of them is worth a conversation about the paid plan.

Users at the wrong moment

The moment matters as much as the user. A frustrated user, one who stopped logging in or someone in the middle of a complaint gets help and no offer, because a badly timed pitch can turn a free user into a lost one.

Knowing who not to sell to is half of selling well.

When is the right moment to offer the upgrade?

Right after the user gets value, or when they hit a limit. An email on day 14 of a trial arrives when the calendar says so, regardless of where the user is.

As an example, two users get the same "your trial ends soon" email on the same morning. One just published their tenth report and hit the free limit yesterday. The other hasn't logged in for nine days.

For the first, the email arrives a day late. For the second, it arrives with nothing to upgrade to.

An upgrade offer works when it describes something the user already wants more of.

How do you know it's working?

By measuring against a control group. Part of any free base converts on its own, so a good month of upgrades says little without a group that didn't get the new onboarding, follow-ups or chatbot to compare against.

A control group (also called a holdout) is a random slice of eligible users who get none of the new tactics while everyone else does. The gap between the two groups is what the tactic actually earned.

Without a baseline, every user who would have paid anyway looks like a win.

The two numbers worth tracking are onboarding completion and free-to-paid conversion. Onboarding completion moves first, often within weeks. Conversion follows, at the pace of the product's own buying cycle.

Reading the result as a funnel shows where a tactic breaks. For stalled onboardings that means follow-ups sent, users who came back, onboardings completed and users who paid. Many returns with few completions point at the onboarding itself. Many completions with few payments point at the offer or the timing.

The same loop we use for churn applies here: a baseline, a hypothesis with a reason behind it, a control group and decision rules agreed before the test starts. It's laid out step by step in how to reduce and prevent churn with Claude.

Key takeaways

How Zentesimal fits in

Zentesimal builds AI Managers for customer experience and customer success teams. Workers (AI agents specialised in each stage of the customer lifecycle) run inside the team's existing stack, with access to each account's plan, usage, conversations and tickets.

For free-to-paid conversion, that means an onboarding Worker that matches each user's problem with the features that solve it, follow-ups for users who stall, and a support assistant that helps free users and spots the ones ready to buy. Playbooks (the rules behind each action) decide what happens in each account: help, follow up, sell or wait. The conversion and support use cases show how it works.

Free users were always worth talking to. Now it's affordable to talk to all of them.