Net future value per customer
One score per customer, net of returns, forward and reverse logistics, support, acquisition cost and discounting. The factors come with the score.
PRODUCT
The allocation engine
Six functions that answer a single question: how much can we spend on this customer without eroding margin?
01 —
What the engine does
Calibrated credible interval
Lower, median and upper bounds over the requested horizon, computed in offline batches, with the calibration status of the band.
Treat or suppress recommendation
For a proposed action cost: customer archetype, cost ceiling, expected net gain and kill switches. A simulation is flagged as such, distinct from a committed action.
Gateway product catalogue
For each first-purchase SKU: archetype, repeat trajectory and habituation effect. You see which first purchases open a relationship and which close it.
Configurable margin policy
Margin floor, hurdle rate, maximum budget per customer, maximum discount, and an optional tier for key accounts. Each parameter is explicit and editable.
Traceability
Audit of the policy applied to each customer, cohort analysis, export and erasure of a customer's data on request.
02 —
How you query it
Two ways in. Through the API, your systems send structured requests to the engine. Through MCP (Model Context Protocol), your AI agents call the engine’s functions as tools: an agent can answer an allocation question in natural language, for example “what cost ceiling for this customer, and should we treat them?”, based on the engine’s results. The engine computes; it is the agent that receives the natural-language question and calls it.
03 —
Who it is for
Brands and retailers in any industry that hold a per-customer transaction history and decide on discounts, offers or follow-ups.
04 —
What the engine does not do
- It does not replace the commercial decision: it gives its ceiling and its risk.
- It promises no result: every value is an estimate, given with its band.
- It is not an ad-targeting tool.