AIagentsystemformanagingafranchisenetwork
For a 180-location network we deployed a system of 12 AI agents covering sales, support, procurement and marketing — to cut back-office load and bind the processes into a single, manageable model.
A 180-location franchise network. Back-office headcount grew faster than revenue and standards drifted between regions. They didn't need "task automation" — they needed a system that holds whole processes end-to-end and doesn't depend on the person on a given site. We shipped an agentic OS: 12 agents, one orchestrator, one panel.
The problem
40 people in the back-office. 70% of their time goes on routine: reconciliations, supplier email threads, report assembly.
Constraints
No breaking existing 1C and SAP integrations. Security — inside the corporate perimeter.
Goal
Cut manual operations 60%+ with no layoffs: free people up for negotiations and growth.
Manual operations in the back-office
- 40 people in the back-office
- 70% of their time on routine and reconciliations
- Reports stitched by hand over 3 days
- Every uptick in volume = a new hire
- 40 people working on growth, not routine
- 12 agents handle the routine
- A report assembles in 10 minutes
- Growth without new hires
Shipped in 6 weeks
From first call to first production launch. No phase two, no endless iterations.
Discovery
Mapped 18 processes for agents. Locked metrics and baseline.
Design
Architecture: 1 orchestrator + 12 agents. Roles and SLAs defined.
Build
Built agents and connectors to 1C, SAP, email. Tested on live data.
Launch
Shipped to production. Handed the KPI panel to the client team.
How we built it
Discovery
In 2 weeks we identified 18 processes an agent closes faster than a human.
Orchestrator
One LangGraph "brain" routes tasks across 12 agents. Every action logged.
Integrations
Direct connectors to 1C, SAP, email, Telegram. No middleware, no vendor lock-in.
KPI panel
One metric — share of manual operations. The drop is visible in real time.
Orchestrator and agents
The orchestrator reads context, picks an agent, verifies the result. Humans handle exceptions.
Solution architecture
One LangGraph orchestrator, 12 narrowly specialised agents, direct connectors to existing systems. No middleware, no vendor lock-in.
What came out
“We didn't fire anyone. We just stopped hiring. The back-office didn't grow for two quarters — and revenue went up 40%.”
Maria Goncharova
Igor Golikov
Vitaly Kust
James Park
Mariia Golikova
Marketing platform with AI agents
We built and shipped an AI system that unifies market and competitor research, audience segmentation, offer crafting, landing-page and creative generation, and performance launches across any acquisition channel. One person runs ×10 the campaigns without losing the brand voice.
