AI workflowIn production

A salesperson describes the deal in a chat thread. The contract request builds itself.

A chat bot pulls contract details out of a conversation, asks for what is missing, takes corrections and, on a “yes”, sets up the task, announcement, contract and dashboard entry.

5set-up steps from one “yes”
9simulated scenarios passed, plus a live run
A chat thread where a salesperson describes a deal and a bot drafts, corrects and submits the contract request
Rebuilt interface · illustrative data

The business problem

New contracts started with a form. The salesperson filled in fixed fields, and a workflow turned them into tasks and documents. Forms cannot ask a follow-up question, accept a correction or answer “which template will this use?”. Missing or wrong details were found later.

What changed

Yash built a Slack bot on n8n. A salesperson mentions it and describes the deal in plain text. An AI model extracts the contract fields and the bot asks in the thread for anything missing. It takes corrections, answers questions about the deal and posts a summary. A “yes” in the thread then creates the onboarding task, announces the request, builds the contract document (for automation deals in USD or EUR), adds an account dashboard task with the value in euros and replies with what happened.

Before & after

AreaBeforeAfter
Capturing the dealA fixed formA conversation that asks for what is missing
CorrectionsStart again or fix downstreamSay what changed in the thread
What gets approvedWhatever the form heldThe exact saved draft shown in the summary
QuestionsAsk someone in opsAsked and answered in the thread

The business value

The bot recognises five kinds of message: update, approve, cancel, question and chit-chat. One approval sets up five things that were handled separately. No time or error reduction has been measured, so none is claimed.

How to read the return

The bot is live, but no before-and-after timing has been measured. The value claimed is the five set-up steps one approval now covers, not hours saved.

Why the system is dependable

Every summary saves its exact deal to a data table, and approval submits that saved copy. Rewording by the AI model cannot change what was approved. The draft is marked submitted before anything is created. Nine scenarios passed in simulation with the real AI models running, including corrections, cancellation and a change request after submission. A live run created a real task correctly, then was cleaned up.

A closer look at the build

Deal-to-contract conversationA deal-to-contract conversation ↗ (full size, opens in a new tab)

Rebuilt chat thread. Company, people and figures are fictional.

Could this work for your team?

If requests arrive as free-text conversations and are re-typed into a form, a bot that reads the conversation and confirms before it acts can save the re-typing.

Talk through a similar project
Keep exploring

More case studies

All case studies