Every brand has the data. Few act on it by 9am.
Terminal6 is the agentic operating system for consumer commerce. It unifies every marketplace, store and warehouse into one model, and runs function agents that tell each team what to do today, with numbers computed, not guessed.
Built by an operator, after 15 years running planning and finance for consumer businesses.
One model of the business. Agents on top. Memory underneath.
Five components, built for the way consumer brands actually run: many channels, a few warehouses, and a small team that needs answers before the day starts.
Every channel mapped to one product, one SKU and one day: sales, stock, purchase orders, returns, fees and ad spend.
Brand, category and team knowledge, plus a governed library of analyses that every agent and every question calls.
Every ask, owner and outcome recorded, so tomorrow's run knows what was done and what wasn't.
The same analyses, open to Claude and other assistants, scoped to each person's access.
One agent per function. Each one owns its actions.
An agent watches one part of the business and can only ask for what that function controls. Ownership is clear, and no two agents ask for the same thing.
Availability
Days of cover for every product, in every channel and region. Finds what runs out before stock can arrive, and where to pull it from.
Replenishment
Open purchase orders against forward demand. Catches the order that lands too late and the one that leaves too much behind.
Demand
This fortnight against the last, by product and channel. Splits traffic from conversion so the fix reaches the right team.
Growth
Ad spend read against stock and margin, so budget goes to products that can sell and stops on ones that can't.
Margin
Contribution margin by product and channel after fees, returns and ads. Flags where growth is costing money.
Planning
Plan against actuals, every day, for every channel. Turns the monthly review into a morning read.
The data says one thing. The business means another.
Most AI on commerce data fails in the gap between what a table says and what an operator knows. These are a few of the traps we hit in production, and the rule each one became.
Sales stopped. Is it ops or marketing?
An out-of-stock listing also shows falling traffic and falling ad spend. Report the loudest signal and you blame marketing for a supply problem every time.
RULETest causes in order: stock first, then price, spend and traffic. The first test that fires is the cause.
Quick commerce counts demand twice
One platform reports national sales on every warehouse row. Add them up and demand is overstated by exactly the number of warehouses.
RULERead stock by location and demand at national level. Never sum the two grains.
Ad spend that doubles on a join
One product can carry several marketplace listings. Join ad spend to listings directly and over a third of products show double the spend.
RULEResolve every ad to one product before it meets the catalogue.
A cancellation rate that reads zero
On the largest marketplace, 99% of cancelled orders carry zero units. A cancellation rate looks perfect and means nothing.
RULEMeasure returned units, and say which measure is used.
Today is always a collapse
At 6am the current day holds a few percent of a normal day's orders. Chart it and every morning opens on a crash.
RULEEvery window ends at each channel's own last complete day.
Nobody acts on "investigate"
The first agents ended half their findings with "investigate further". The teams read them and did nothing.
RULEEvery ask is a verb the owner can do today: send, expedite, pause, trim, restore.
Embedded with the team, live in weeks.
We work inside the brand, next to the people who will use it, and go live in about three weeks. Agents earn trust one function at a time.
Connect
Plug in every channel, warehouse and ad account. Map every product to one catalogue.
Model
Build the brand's context: categories, teams, owners and the rules they already use.
Deploy
First agents go live. Each team gets a morning page and checks it against their own sheets.
Compound
Every action is logged and every miss teaches the next run. New agents are added function by function.
Running in production, not in a deck.
A multi-channel consumer brand, India
Marketplaces, own store, quick commerce and three warehouses, all in one model. The supply, sales and marketing teams open their page every morning and act from it.
Reserved for the next brand.
We take a small number of design partners at a time.
Apply →Built so an operator can act on it without checking.
Numbers from code
Every figure is computed and checked before it's shown. The model picks which figure to cite and writes the sentence around it.
Missing is not zero
A late feed is shown as late. It never shows up as a collapse in sales.
Scoped access
Each brand's data is isolated, and every read is scoped to the person asking and logged.
Every ask audited
Each ask carries an owner, a verb and its evidence, and stays on record until it's closed.
Built where the domain and the engineering meet.
Most AI teams have engineers who learn the business, or operators who brief engineers. Terminal6 was designed and built by one person who has done both jobs.
"I spent 15 years running planning and finance for consumer businesses, and most of it in spreadsheets. So I built the system I wanted, embedded with a brand, next to the people who use it every morning."
- S&OP and demand planning across channels and sale events
- Inventory: cover, transfers, re-order lines, purchase orders
- Channel economics: fees, returns, ads, contribution margin
- Finance: P&L, plan vs actuals, monthly and daily
- How a planner, a buyer and a brand head actually decide
- Canonical data model across marketplaces, store, OMS and ads
- Governed analysis library, shared by agents and chat
- Agents with owned actions, evidence and a decision ledger
- Evaluation harness for answer quality, speed and cost
- MCP access, row-level security, nightly orchestration
Bring us your morning.
Become a design partner
Multi-channel brands with a team that will act on it daily.
- Marketplaces, own store, quick commerce
- Live in about three weeks
- One function first, then the rest
See it in production
Building agents for consumer and CPG companies? Let's compare notes.
- Live walkthrough of a production deployment
- What it took for operators to trust it
- Where it goes next
contact@terminal6.io