AI Agents for a Coffee Chain: Ops and Purchasing

This case shows how a coffee chain used three AI agents to run daily finance reports, staff onboarding and outlet purchasing. The onboarding agent answers 95% of new-staff questions, and questions to the manager fell by 65%. Purchasing for an outlet went from 40 minutes to 5, photos and SKU lookup included, and all of it now runs through one system.

Context

The chain runs multiple outlets. Each one has its own numbers, its own new hires and its own stock running low. Managers held it together by hand: they handled reports, answered staff questions and sorted out purchasing.

That works at one outlet. It breaks as the chain grows, because every outlet adds more reports, more questions and more orders to the same few people. PHII Labs built three agents, one for each pain. This is the pattern behind our AI automation work for multi-outlet businesses.

Problem

Reports depended on people. A finance report for each outlet had to be built every day, then rolled up into one summary. A wrong number could travel upward before anyone caught it
Managers were the help desk. New staff brought their questions to the manager, which took time away from running the outlet
Purchasing was slow. Turning "what has run out" into a purchase request took about 40 minutes

What was built

Daily finance reports with guards. At the end of every day, an agent builds a finance report for each outlet and a consolidated summary for the chain. Validation checks run before anything is sent, so wrong numbers do not go out
Onboarding agent. New staff get step-by-step guidance. It answers from a large Q&A knowledge base, so people get an answer right away instead of waiting for a manager
Purchasing through Telegram. Staff write in a chat what has run out and what is needed. The bot finds each SKU in the database and builds the purchase request
One system for buying. Purchasing no longer sits outside the process. All of it runs through the agent

The agents follow the same rules we describe in how to build agent systems that ship: clear inputs, checks before action, and a human where it matters.

Results

Purchasing: 40 minutes → 5 minutes, including photos and the SKU lookup
95% of onboarding questions closed by the AI
65% fewer questions to the manager
All purchasing now runs through the system
Daily reports for every outlet and the chain, with checks before they go out

Takeaway

The three agents work because each one has a narrow job and a clear source of truth. Reports pull from the numbers and are checked first. Onboarding answers only from the knowledge base. Purchasing matches words to SKUs in the database. The biggest win was not speed alone. Managers got their time back, and the chain got one record of what it buys.