Cookbook
Tell an agent's claim from a verified fact
Your agent will be wrong sometimes. The job is not preventing that, it is making sure you can tell which facts are worth trusting before you repeat one to a customer.
The situation
You are about to tell a customer something your CRM says. In an ordinary CRM every field looks identical: the number somebody typed after a phone call and the number an agent guessed from a website are the same grey text in the same box.
That is fine until it is not. The failure is specific and it is embarrassing rather than catastrophic: you repeat a figure in a meeting, and it turns out nobody ever checked it.
What must be true when you are done
- Every value carries how it was come by, not just who wrote it.
- You can ask what nobody has stood behind, before you rely on it.
- A person's correction outranks an agent's next import, and stays.
Tell your agent how it knows
Every write can say how the value was come by. It is one field and it is the whole mechanism.
| Kind | What it means |
|---|---|
| observed | The agent saw it. A price on the page it was reading. |
| inferred | The agent worked it out. A company size from a headcount page. |
| imported | A file said so. The weakest thing on this list. |
| user_supplied | A person typed it. |
Ask for it once and the habit sticks, because it costs the agent nothing to say.
You
When you write to the CRM, always say how you came by the value. Observed if you read it, inferred if you worked it out, imported if a file said so.
Ask what nobody has checked
This is the question an ordinary CRM cannot answer. Not "who wrote this", which is forensics after you have already been embarrassed, but "which of these is nobody standing behind", which you can ask beforehand.
You
Before I take this to the call: which values on this account has nobody checked, and how did you come by each one?
Agent
Now the two facts are different facts. The website came out of a spreadsheet and the industry is the agent's own reading, and you can decide how much weight each one carries in a sentence you are about to say out loud.
Narrow it when the list is long. Asking for just the imported values surfaces the weakest claims first, which is usually where the wrong ones are.
Correct one, and it stays corrected
The important half. When you fix a value, it is a person's assertion, and a person's assertion outranks anything an agent can write. The next import does not quietly undo you.
You
Their industry is freight forwarding, not logistics software. I checked with Priya.
Re-run the import that produced the wrong value and it is held rather than applied, and the result says so by name. That is what makes the correction worth making: in a CRM where the next sync overwrites you, people stop correcting things.
What your agent cannot do
An agent can see what nobody has checked. It cannot check it. There is no tool for that and the absence is deliberate: a confirmation means a person looked, so an agent confirming its own assertion would be a signature on its own work rather than a check on it.
Correcting a value IS the check, and it is the one people actually perform. You are not asked to work through a queue agreeing with things.
Verify it
- Open a record. Values an agent asserted and nobody has checked are marked as such, in the place you are already looking.
- Re-run an import over a value you corrected. It should report the field as held, with what it wanted to write and what it kept.
Run it again
Before anything that leaves the building: a call, a proposal, a board number. It takes one question, and it is the difference between a CRM you quote from and one you check first.