For example, you might say yes to quoting a price that's on your website, and no to agreeing a delivery date. That's a sensible mix. One setting can't hold it.
What should it never do?
- Let a buyer believe something nobody said. We work with several firms in your sector may be true. But the buyer hears you've done this exact job before. The problem is in the sentence itself, so no setting makes it OK.
- Give its word. You have my word means nothing coming from software. There is no person behind it.
- Promise a date when nothing is set up to meet it. This is the mistake that gets agents switched off.
What can it do once it has something to read?
| it may... | once this exists |
|---|---|
| quote a price or a detail about a plan | that fact, with an owner who has checked it recently |
| say someone at your company approved something | a written record of the approval |
| quote a number | where the number came from, named, with an owner and a date |
| make a promise that depends on something | that dependency written down, so the agent can say it out loud |
Most of these open up as soon as the fact exists. The agent doesn't have to wait through a trial period first. Good behaviour over time can't make it safe to quote a number that has no up-to-date source. Give it the source and it's safe straight away.
A fact only counts while someone stands behind it. Each fact has an owner and a date to check it again. If that date passes, the agent loses anything that relied on the fact, until the owner confirms it again.
What does it earn over time?
Some things do grow with a good track record:
- how much it may send
- how many people one mistake could reach
- when it has to ask a person
- how much of its work someone still reads
The last one never drops to zero.
How Reynforce helps
You set these up as Rules, for each agent and each kind of work it does. A rule says what the agent may promise, what it has to ask about first, and what it may never say. The facts it relies on live in Facts. Before the agent promises anything, it checks with Reynforce, and you can see the answer in Activity.
When Reynforce is the wrong choice
- Your agent only answers from a fixed script. If it can't say anything new, there's nothing to decide.
- You'd rather a person approved every message. That works when there aren't many messages. Do AI sales agents need human approval? explains when it stops working.
What it costs
£99 per agent a month for your first 10 agents. After that it's £79 for each agent, up to 49. If you pay for a year at once, you get two months free. If you'd like us to do the setup and sign-off with you, that starts at £2,500, paid once, and the price depends on the agent's job. Pricing has the rest.