How to use AI chatbots to sell more: turn support into a sales channel

To sell more with an AI chatbot, the support chatbot you already have is the best place to start. It talks to more users than anyone in the company. Once it can spot users ready to buy or expand, and knows when to hold back, conversations that used to end in an answer start ending in upgrades.

Most companies already have their best salesperson. It's sitting in the support inbox.

What follows comes from giving that job to the support assistants Zentesimal runs in production: which signals to watch, when to start the conversation, when to stay quiet, and how to tell whether it's working.

Why do most chatbots never sell?

Because they were built to close tickets. Almost every support team is building or running a chatbot, and the goals behind it are cost and self service, rarely revenue.

In a survey by Gartner, the research and advisory firm, of 321 customer service leaders, run in October 2025, 91% said leadership was pushing them to implement AI. The goals behind that push were customer satisfaction, operational efficiency and self service. Revenue wasn't on the list.

Who talks to more potential customers than anyone in your company?

The support chatbot does, and by a wide margin. A sales rep talks to the leads that reach them. A CSM (customer success manager) looks after a set list of accounts. The chatbot talks to everyone who has a question, free users included.

And someone asking a question about a product is already using it.

At 6,000 conversations a month, that's 6,000 chances to convert a free user, expand an account or catch one that's drifting away. The number that matters here is the one already sitting in your own helpdesk.

Every conversation that ends with just an answer is a sale that closed without anyone noticing.

The free base is where a lot of that potential sits. In freemium models (a free plan with paid upgrades), most users never pay.

Vineet Kumar, a Harvard Business School professor who studied these models, found that most companies convert between 2% and 5% of free users, as he wrote in Harvard Business Review. The other 95% keep using the product, and plenty of them keep asking support questions.

Paying customers are the other half. Every account close to outgrowing its plan passes through support sooner or later.

Which signals show that a user is ready to buy or expand?

A user is warm when their behaviour starts to look like a customer's, or like a bigger customer's. An AI assistant can pick up on that behaviour through signals in the conversation and in the account.

Every product has its own. These are some of the ones used in the accounts Zentesimal runs, both for free users about to pay and for paying accounts ready to grow.

They ask about pricing

A question about cost usually means the user has already decided the product is useful. What's left to settle is whether it's worth the price.

They visit the pricing page

A visit in the middle of a support conversation says they're comparing options, often right after finding something their current plan doesn't cover.

They ask about features from higher plans

Someone asking whether they can do something on a higher tier is already picturing their work on that tier.

They ask for a discount

A discount request is a negotiation, and people only negotiate for things they want to keep.

Nobody negotiates the price of a product they plan to leave.

They sit at the limit of their plan, or use the product heavily

The product is already doing the job and the plan is the only thing in the way. For a free user, the next step is the paid plan. For a paying account, it's the next tier.

These are also the users least likely to open a ticket, which is why a chatbot that only waits for questions never meets them.

Should the chatbot wait for the user to ask?

No. The users closest to buying are often the ones who never open a ticket, so a chatbot that only reacts misses the best part of the opportunity.

This was the biggest change in how the assistants Zentesimal runs are set up. Signals in the account start the conversation too, alongside tickets.

As an example, a free user at 95% of their monthly limit for the third month running, who has never opened a ticket, is invisible to a reactive chatbot.

A proactive one reaches out before they hit the wall. It shows them what they've been doing with the product and what the next plan would let them do.

When should a chatbot not sell?

When the user is frustrated, barely uses the product or is complaining about something. Of everything learned while teaching support assistants to sell, this rule matters most, because an offer at the wrong moment trades a sale for a churn.

A user who has reported the same bug twice this week reads an upgrade message as proof that nobody is listening. The same user, a week after the fix and with usage back up, is a completely different conversation.

Customers don't give chatbots many second chances either. In a Gartner survey of 3,566 customers run in early 2026, only 27% said they would try a chatbot again after a negative experience.

That's why the chatbot reads the account before it offers anything: open tickets, recent complaints, how usage is trending. When any of those points the wrong way, the conversation stays on helping.

Why does account context decide what the chatbot offers?

A generic chatbot offers things blind. One that sees usage, plan and history offers what fits the account in front of it.

As an example, two users ask the same question on the same day: how much is the Pro plan? One is at 98% of the limit on a feature Pro doubles. The other hasn't logged in for six weeks.

The first gets a walkthrough of what Pro unlocks. The second gets help getting back into the product, and no price.

The question is identical. The account is what changes the answer.

That judgement is only as good as the data the chatbot can reach. We broke down which sources matter most, and how much each one moves the answer rate, in which sources to connect to an AI assistant.

How do you know the chatbot is actually selling?

By measuring two things against a control group: how many warm users the chatbot detects, and how many of those convert or expand.

That baseline matters because some users upgrade on their own. With 2% to 5% of a free base converting anyway, a good month of upgrades says very little without something set before launch to compare it with.

Without a baseline, every upgrade looks like a win.

What happened when Zentesimal's assistants started selling

Support turned into a revenue line.

In the accounts where Zentesimal runs the support assistant, giving it the ability to sell, together with the signals that show a warm user, now brings in an average of $45k a month per account in new and expansion ARR (annual recurring revenue).

That figure combines free users converting to paid and existing accounts moving to bigger plans. The control group, defined before the assistant started selling, closed less than $2k a month. The assistant closes more than 22 times that.

The number didn't arrive on day one. The average climbed from $30k to $45k a month over recent months, and two changes explain most of the jump.

The first was sharper signals. The assistant got better at telling a user who is ready from one who is only curious, so offers reached the right people more often.

The second was the way it sells. The assistant learned to help first and then explain what the next plan would do for that specific user, the same way a good salesperson would.

A price on its own rarely convinces anyone. A clear picture of what changes with the upgrade does.

Selling is a skill the assistant gets better at, the same way a new hire does.

These are internal Zentesimal figures across accounts in production during 2026. You can read more about them in our success stories.

What this looks like in practice

• The support chatbot already holds the most sales opportunities in the company. At 6,000 conversations a month, that's 6,000 chances to convert or expand.
• Buying intent leaves signals the assistant can read, such as pricing questions, pricing page visits, questions about higher plan features, discount requests, and usage at the plan limit.
• Waiting for tickets leaves the best opportunities on the table, because heavy users close to upgrading rarely open one.
• Frustrated, inactive or complaining users don't get an offer. That rule is what keeps a sale from becoming a churn.
• Results only count against a control group, since part of any user base upgrades on its own.
• Selling improves with iteration. The same assistants went from $30k to $45k a month once their signals got sharper and they learned to explain the value of the next plan like a salesperson would.

How Zentesimal fits in

Zentesimal builds AI Managers for customer experience and customer success teams. The assistant runs inside the team's existing stack, with access to the plan, usage, conversation history and tickets of every account.

Workers (AI agents specialised in each stage of the customer lifecycle) watch the signals in each account, and Playbooks (the rules behind each action) decide what happens next: help, sell or wait. The conversion use case shows how that works for free to paid and expansion.

The support conversation was always the one closest to the sale.

Book a demo