Deal Approval Pack Agent · Use case
How can we use an AI agent to assemble the approval request for a non-standard deal on our own infrastructure?
A deal approval pack agent takes one non-standard deal and assembles the request its approvers need: the terms the customer has asked for, how each differs from your standard terms and your policy, who must approve each under the policy, and past approvals of the same kind with their conditions. The deal desk and the named approvers decide. The agent recommends neither approval nor refusal, starts no approval, agrees nothing with the customer and changes nothing on the deal or the quote.
The deal is not standard.
The approvers each ask for the same facts.
- ✗ 01A non-standard deal goes to finance, legal and sales leadership, and each asks the salesperson for the same facts in a different form.
- ✗ 02An exception agreed for one customer is cited by the next, so what was approved before, and on what conditions, has to be in front of the people who decide.
Where it stops for a person
Always. The deal desk and the named approvers decide.Always. It assembles, it does not weigh: it does not say whether the deal is worth the exception, and it gives no view of the customer or of the salesperson who asked. A past approval is shown as a fact with its conditions, not as a precedent to follow. A contract term outside your standard is marked for legal, who decide every term. The pack names the approvers the policy gives and does not choose between them.
It assembles the request.
The deal desk and the approvers decide.
Step · 01
The salesperson or the deal desk names the deal.
Step · 02
The agent reads the deal and the draft quote through your own APIs, read only, and lists the terms asked for: price, payment, length, liability, service levels.
Step · 03
It sets each term beside your standard and your policy, quotes the clause, and names the approver the policy gives for each difference.
Step · 04
It finds past approvals of the same kind in the records you allow it to read, and gives each with its date and its conditions.
Step · 05
The deal desk checks the pack and sends it to the approvers, who approve or refuse in your own system.
Reads from · connectors
Your own APIs
The one deal named and its draft quote, read through your own APIs with the operations that change something left out; the approval policy, the discount policy and your standard terms, indexed as a knowledge source; and the record of past approvals, where you hold one, read only. It runs with the permissions of the person using it.
Acts through · MCP servers
Nothing. It writes a pack. It has no tool that starts or records an approval, changes a deal or a quote, or sends anything to the customer.
What you set up
- Index the approval policy, the discount policy and your standard terms as a knowledge source.
- Import the description of the API that holds deals, quotes and past approvals, and keep only the operations that read.
- Build the agent in Katonic Studio and test it on past non-standard deals where you know who approved what.
- Publish it to the deal desk first.
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 terms, past approvals and the pack 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 approve the deal?
No. It assembles the request. The deal desk and the approvers named in your policy decide.
Does it recommend approval or refusal?
No. It sets each term beside your standard and names who must approve it. The people who decide read both.
Where do the past approvals come from?
From the record you hold of them, read only. Each is shown with its date and its conditions.