Take this sentence: We can have that live by the end of the month. It is polite and on topic, so every wording check will let it through. Whether it is true is something the sentence alone can't tell you.
This page lists the checks a sales agent needs. For what one guardrail is, read What is a GTM agent guardrail?
What are the two kinds of check?
| a wording guardrail | a promise check | |
|---|---|---|
| it asks | is this sentence OK to send? | is this true, and is this agent allowed to say it? |
| it reads | the words in the message | your company's facts, each with an owner and a date, and the rules you set |
| it catches | rude replies, leaked data, off-topic replies, the wrong tone | an old price, a fact nobody owns, a date nobody can keep, an approval that never happened |
| it misses | a polite promise nobody can keep | a rude sentence |
You want both kinds. Most teams that talk to us already have a wording guardrail. They have just found out they need a promise check too.
What should a promise check cover?
Make it a set of separate yes-or-no questions. One slider that goes from "careful" to "bold" can't hold a mix of answers. Here are some examples:
- May the agent quote a number when only one source backs it up?
- May it say someone at your company has approved something?
- May it make a promise without saying what the promise depends on?
Some answers should always be no. An agent should never promise a date when nothing is set up to meet it. It should never give its personal word, because there is no person behind it. And it should never let a buyer believe something that nobody actually said. These are mistakes in the message itself, so no level of risk makes them OK. Words like always and never are held back too, unless you choose to allow them.
Other answers can become yes once the agent has something reliable to read. For example, it can mention your security certificates once a named person owns that fact and has checked it recently. If the check date passes, the agent loses that permission until someone confirms the fact again.
How Reynforce helps
Reynforce is a promise check. Your agent asks Reynforce before it makes a promise. The answer comes from your Facts and your Rules, and it is one of allowed, ask first or not allowed. You can see every check in Activity. When a person needs to approve something, the request reaches them by email, Slack or text.
It is built so you can give agents more work. When there's a choice between blocking something by default and allowing it with a limit and a record, Reynforce allows it with the limit and the record.
When Reynforce is the wrong choice
- You're worried about tone, safety or prompt injection (someone hiding instructions in a message to trick the agent). Use a wording guardrail. Reynforce doesn't filter wording.
- Your agents never talk to buyers. An agent that only does research inside your company has nothing to promise.
- You run one agent. One agent rarely needs this. It starts to matter when several agents and people are talking to the same buyers.
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.
How Reynforce compares puts it next to tools you may already use.