Chasing Invoices With WhatsApp Agents

Almost every small business has the same quiet problem. Work gets done, an invoice goes out, and then the money does not arrive when it should. Chasing it is awkward, it is repetitive, and it competes with billable work for the owner's attention. So it slips. A reminder that should have gone after seven days goes after thirty, if it goes at all, and the business carries the cost in its cash flow.

This is exactly the kind of task an agent should own: real, repetitive, valuable, and unglamorous. We built JustPayMe to do it. It uses WhatsApp-integrated agents to chase late invoices, on a no-win-no-fee model aimed at SMEs. This piece is about how we thought through the design, because the interesting parts are not in the model. They are in the channel, the conversation, and the commercial model.

Start with the problem, not the technology

The temptation with any agent is to start from what the model can do. We started from why invoice-chasing fails in practice. It is not that owners do not know who owes them. It is that the chase is emotionally and practically costly. Nobody enjoys nudging a customer for money, the wording is fiddly to get right, and the follow-up has to keep happening on a schedule that a busy person will not maintain by hand.

That framing decides the whole design. The job is not to send one reminder. It is to run a consistent, polite, persistent chase over time, on the channel the customer actually reads, without the owner having to think about it. An agent is a good fit precisely because the work is repetitive and rule-shaped, the part of a process people skip not because it is hard but because it is tedious.

Why WhatsApp is the channel

Email is where invoice reminders go to be ignored. They sit unread, get filtered, or land under a week of other mail. WhatsApp is where a lot of small businesses and their customers already talk, and a message there gets seen. Meeting people on the channel they actually read is most of what makes a chase land.

Choosing that channel is also the source of the real engineering. WhatsApp is a managed messaging platform with its own rules about how businesses may message people, what templates are allowed, and how conversations are structured. The agent has to operate inside those rules, not around them. Building on it well means respecting the platform's constraints by design rather than treating it as a generic text pipe, which is a large part of why doing this properly is harder than it first looks.

Designing the agent

A reminder that says you owe us money is a notification, not an agent. The point of an agent here is that the chase is a conversation, and conversations do not go in a straight line. A customer replies that they have already paid, or asks for a copy of the invoice, or says they will pay on Friday, or disputes the amount. The agent has to keep the exchange on track across those turns and toward a single outcome: the invoice gets paid or the situation is escalated to a person with a clear record of what happened.

• It works from real invoice and payment data, so it chases what is genuinely owed rather than a generic template.
• It runs the follow-up end to end on a schedule, instead of firing one reminder and stopping.
• It holds a multi-step conversation, handling the common replies (already paid, send the invoice, pay later, query the amount) without losing the thread.
• It hands off to a person when a case needs judgement, with the conversation and context attached.

The data connection is what keeps the agent honest. Because it is tied to live invoice and payment records, it is acting on facts: who owes what, what has been paid, and what is still outstanding. That is what stops the embarrassing failure mode of chasing a customer who has already settled, and it is what lets the agent stop the moment a payment lands.

No-win-no-fee, and why it fits

We built JustPayMe on a no-win-no-fee model for a reason that is more than a pricing decision. The businesses that most need help chasing invoices are the ones with the least slack to gamble on software that might not work. Asking them to pay up front to find out is the wrong way round.

A no-win-no-fee model removes that. The SME carries no upfront cost and no risk to try it, and the incentives line up: the agent is worth paying for only when it actually recovers money that was stuck. It also keeps us honest as the builders. We are not paid for a clever conversation that goes nowhere. We are paid when the invoice gets paid, which is the only outcome the customer cared about in the first place.

The best tasks for an agent are the ones people skip because they are tedious, not because they are hard. Chasing an invoice is exactly that, and doing it over WhatsApp is where it finally gets done.

The general lesson

JustPayMe is a narrow product, but it is built on a pattern we use widely. Find a real, repetitive task that people avoid. Put the agent on the channel where the work actually happens. Tie it to live data so it acts on facts rather than assumptions. Design for the messy middle of a real conversation, and keep a person available for the cases that need judgement. Get those right and an agent stops being a demo and becomes something a small business can quietly rely on to recover money it was owed.