Platform
The difference is architectural, not a marketing line. Raw data never leaves the company that produced it. What reaches the centre is derived, aggregated and irreversible.
The problem
Every market intelligence product faces the same wall. The data is only valuable in aggregate, and no serious company will hand over its cost base, its suppliers and its margins so that a competitor can benefit from them. Promising confidentiality does not solve this, because a promise is a policy, and policies can change with an owner.
We resolved it by removing the possibility rather than prohibiting the behaviour. The centre never holds anything that could be traced back to a contributor — not because we choose not to look, but because the record was transformed before it ever arrived.
Architecture
Data flows upward and narrows at every step. Decisions flow back down and widen again — carrying context that only the individual company has.
Isolated per tenant
Inventory, purchase costs, supplier terms, customer records, margins. This layer belongs entirely to the company that produced it. It is not pooled, not sold, and not visible to us in aggregate. Every query against it is scoped to a single tenant at the database level, not by a filter someone could forget to apply.
Shared and canonical
One identity per physical product, shared across every tenant. When four companies spell the same cartridge four different ways, the graph decides they are one thing. This is the layer nobody sees and everything depends on: get identity wrong and every price, index and recommendation downstream is quietly wrong too.
Derived and irreversible
Private events become market signals only after they cross a contributor threshold and lose their origin. A single company's sale never becomes a visible data point. Below the threshold nothing is published — there is no override, because the aggregation happens before anything is stored.
Private again
Shared signals meet private context: your cost base, your margin floor, your cash position, your capacity. A competitor can copy a market signal. They cannot copy a decision that depends on what you paid and what you need to earn. This is why the last layer is private again.
Decision quality
Asking a language model to produce a price is how you get a confident, fluent, invented number. Ours never does. The figure comes from optimisation over observed transactions, bounded by hard constraints — cost floor, margin target, cash available. The model's only job is to put that result into a sentence a person can act on.
When the evidence is thin, the system says so and returns a range. A wrong number stated precisely is worse than an honest interval, because someone will trade on it.
Example output
List at 31,900 — expected sale in 6 days
Confidence: high · 34 comparable observations · expected margin +2,400
Feedback
A specific action with a target, a timeframe and an expected value.
Taken or dismissed — both are logged. Ignoring advice is information too.
When the item sells, realised margin is compared to the forecast. The gap tunes the engine.
This closes the loop that most analytics products leave open. It also produces the only number that matters commercially: what the platform earned its customer this month.
Tell us what you are trying to decide. If we are not the right people for it, we will say so.