Яндекс.Метрика

How do you set up purchase approvals?

Five monitors for the new designers take nine days to arrive, and nobody’s dragging their feet: the request is scattered across three DMs, and no one sees the whole picture. In Pachca, the request arrives as one message in a shared chat, an AI agent checks it and pulls up the budget, and a person makes the call with a reaction. Walk through it yourself below.

Purchasing

12 members · AI agent

Today
Member avatar
Paul Evans09:34

Purchase request

ItemDesign tool license renewal, 12 seats
Amount68,400 ₽
Budget lineSoftware, Q3
Needed byOct 1
VendorSoftpoint
Invoice #1180, Sep 19, 2026.pdf96 KB
✅2
Member avatar
Natalie Owens09:58

Purchase request

ItemPrinter cartridges, 4 pcs
Amount12,800 ₽
Budget lineOffice supplies, Q3
Needed bySep 25
VendorPaper Plus
Invoice #2043, Sep 22, 2026.pdf78 KB
🕐1
Member avatar
Laura Kramer10:12

Purchase request

ItemFive 27″ monitors for new designers
Amount142,500 ₽
Budget lineEquipment, Q3
Needed bySep 29, designers' first day
VendorTechnomir
Invoice #4471, Sep 22, 2026.pdf214 KB

Requests land in one chat using a template: item, amount, budget line, due date and invoice. Open the request's thread.

The real problem is the route

For large tenders, companies usually set up a dedicated request system, and there it makes sense. But printer cartridges, licenses and monitors get bought every week. Here, what stands between "we need it" and "we bought it" is usually a question asked in the wrong place, not the lack of a system.

You can build that route right in your team chat app, where the team already spends the whole day. It has one more advantage: there's room in it for more than just people.

People and an agent in one thread

In this setup, the AI agent is a real member of the team, not a separate service you have to copy the request into. You bring it into a thread with an @mention, just like a teammate. It reads the whole thread and replies right there. The difference is that the agent handles the routine work people never get around to, and does it right away.

What you can hand off to it:

  • check a request before a person sees it: is the budget stated, is the invoice attached, is it clear why we're buying this;
  • pull in background info: what's left in the budget line, past purchases from this vendor;
  • bring approvers into the thread and nudge them if a request stalls;
  • write a short summary of the thread for the CFO, who didn't read the back-and-forth;
  • after approval, pass the data to the accounting system so nobody has to retype it.

The agent can do all this only if it has the right tools. For example, the Pachca team chat app has a whole environment built for this:

  • an open API and webhooks: the agent uses them to read requests and learns about new ones the moment they're posted;
  • MCP to connect to the accounting system and the budget spreadsheet;
  • a CLI for the developer who builds the workflow;
  • Agent Skills and docs in llms.txt: the model uses them to figure out on its own how to work with Pachca.

Don't trust the agent with the money decision itself. Its job is to prepare the request so that all a person has to say is "yes" or "no." The final sign-off belongs to whoever owns the budget.

How to set up purchase approvals in Pachca

One chat as the single entry point. Create a dedicated chat for requests and add the agent to it. Anyone who wants to buy something posts there, not to "a friend in procurement."

One request, one message. Agree on a template: what you're buying, how many, why, which budget line and by when. Attach the invoice or link right away. Follow-up questions go in the thread under the message. The agent checks requests against the same template.

The route is known in advance. Set a rule: up to a certain amount, the department head decides; above it, the CFO. The right person gets pulled into the thread with an @mention, either by hand or by the agent. Pachca has cross-chat threads for this: you can bring a teammate into a thread even if they're not in the chat. Your finance lead won't have to join yet another chat just to sign off on one thing.

Status is visible without asking. Agree on reactions: a clock means "looking into it," a checkmark means "approved," a cross means "declined, reason in the thread." The "so what's happening with my request?" question disappears on its own.

The decision stays on record. Six months later, accounting asks who approved this purchase. Search finds the answer in a minute, because the request, the invoice and the "yes" all live in one thread.

Why the chain gets shorter

In DMs, approval works like a relay race. Everyone waits for the person before them, retells the request in their own words, and halfway through it gets sent back for a missing invoice.

In a team chat app, what speeds up the chain is that agents and people work in the same chat or thread. The agent checks first that everything's in place and pulls up the budget info. People see a ready-to-go request and respond in parallel, without waiting on each other. Nothing gets sent back, because the agent already cleared up the questions. Nobody retells anything, because everyone has the same context.

Relay through DMs

≈ 1.5 days

Request
Finance
Manager

One thread (with an agent)

≈ 2–3 hours

Request
Finance
Manager

Where to start

You don't need to rebuild all your purchasing at once. Pick one category, say office supplies or software subscriptions, and run it through the new chat for two or three weeks. See where requests still get stuck and adjust the route. You can add the agent right away, starting with one task: checking requests against the template. Add the rest once it's clear where it saves the most time.

If you want to set up a process like this at your company, get in touch and we'll do our best to help.

How to build an AI agent

API, webhooks, MCP, CLI and Agent Skills: where developers should start