Record Check Agent · Use case
How can we use an AI agent to list the deal records that are incomplete or out of date on our own infrastructure?
A record check agent reads open deals, when a person starts it or on a schedule you set, and lists those that break a rule you have written down: a required field left empty, a close date in the past, no next step, an amount that disagrees with the quote. For each deal it gives the rule and the field. The deal owner corrects each record. The agent changes no record, moves no stage, and its list is about records, not people: it counts nothing by salesperson and gives no score.
The pipeline review starts.
Half the hour goes on the records.
- ✗ 01A forecast and a pipeline review are only as good as the deal records behind them, and nobody has the time to read every record against the rules.
- ✗ 02A list of records to correct turns easily into a table of who keeps poor records, and that is a different tool from a check of the records.
Where it stops for a person
Always. The deal owner corrects each record.Always. It lists, it does not correct and it does not judge: it fills in no field, even where the right value looks plain from the quote. The list is by deal. It gives no count, score or rank by salesperson, and it is not built to be added up by person. It says nothing on whether a deal will close. A deal that is not on the list has not been cleared: the list shows what broke a rule and nothing more.
It lists what is missing or out of date.
The deal owner corrects each record.
Step · 01
The record rules are indexed as a knowledge source: which fields each stage needs, and what counts as out of date.
Step · 02
A sales operations analyst starts a run, or it starts on the schedule you set.
Step · 03
The agent reads the open deals and their quotes, read only, and checks each against each rule.
Step · 04
It lists each deal that breaks a rule, with the rule and the field, set out so that each owner sees their own deals.
Step · 05
The deal owner corrects each record in your own system. The agent changes nothing.
Reads from · connectors
Open deals and their quotes, from a reporting database, read only, through a view your team has prepared, or through your own APIs with the operations that change something left out; and your record rules, as a knowledge source. Where your API checks the caller, a salesperson who runs it sees their own deals and a manager sees their team's.
Acts through · MCP servers
Nothing. It writes a list. It has no tool that edits a deal, moves a stage, changes an amount or a date, or writes a note to a record.
What you set up
- Write the record rules as facts that can be checked in the record, and index them as a knowledge source.
- Prepare a read-only view of open deals and quotes in the reporting database, or import the description of the API that holds them and keep only the operations that read.
- Build the agent in Katonic Studio and test it on invented deals, and on a past month you have already checked by hand.
- Tell salespeople and, where you have them, staff representatives what the list is and is not used for, and start with one team.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. Deal records and the list stay inside the company.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does it correct the records?
No. It lists what is missing or out of date, with the rule and the field. The deal owner corrects each record.
Is the list used to rate salespeople?
It is not built for that. It is a list of deals and fields, with no count, score or rank by person.
How does it decide a record is out of date?
By the rules you have written down, such as a close date in the past or no next step. Each entry names the rule that put it on the list.