What is deterministic commerce intelligence?
Deterministic commerce intelligence is commerce intelligence whose findings are computed from a fixed set of rules and the store's own data rather than generated by a language model. The engine ranks and decides; a language model may only phrase the result.
The same store data always produces the same finding, and every finding traces back to the rule that fired and the signal beneath it. That is the whole difference: a copilot tells you what a model thinks, and this tells you what your data says.
Grounded, not generative
Most tools sold as commerce intelligence ask a language model what it thinks about your numbers and hand you the reply. Ask twice and you can get two answers. When the answer is wrong there is no rule to inspect, because there was never a rule, only a probability.
oddly inverts that. A fixed rule set computes each finding against your own store and channel data. Nothing generative sits in the decision path. A language model is used only to phrase a result the engine has already decided, never to decide it. So a recommendation can always be read backwards: this rule fired, on this signal, worth this many dollars.
The practical test is repeatability. Feed the same store data through the engine twice and you get the same finding twice. If a recommendation changes, your store changed, not the model's mood.
Why determinism matters when the system touches money
Determinism is not an aesthetic preference. It is the property that makes every other guarantee enforceable. A finding that cannot be traced to a rule and a number cannot be checked before it is applied, or audited after.
Three constraints follow from it, and each is enforced in the product rather than promised in the copy:
Decrease-only. oddly may reduce, pause, or block. Increasing, granting, or escalating always requires a human. It cannot raise your ad spend, by construction.
Held-mutate. Every change is a dry run plus explicit human approval before it is applied, and it stays reversible within a known window. The system proposes and verifies; a human disposes.
Audit as a contract. Every action is recorded with the rule that fired and the signal beneath it, so the log is a product surface you can read, not plumbing you have to trust.
A generative decision path cannot honour any of the three, because there is no rule to record, no verdict to reproduce, and nothing to inspect when it goes wrong.
How the term relates to the rest of oddly's vocabulary
Deterministic commerce intelligence is what a merchant experiences: findings about their store, computed and ranked by dollar impact. Underneath sits the deterministic governance substrate, the layer that decides what an automated action is allowed to do before it touches a real account. The two are non-competing. The first names what you get; the second names what makes it safe.
For the fuller picture, see what oddly is, read a worked example of how oddly decides what to recommend, or compare oddly against the tools it is usually weighed against on the comparison pages.
Common questions
What is deterministic commerce intelligence?
Commerce intelligence whose findings are computed from a fixed set of rules and the store's own data rather than generated by a language model. The engine ranks and decides; a language model may only phrase the result. The same store data always produces the same finding, and every finding traces back to the rule that fired and the signal beneath it.
Is oddly's intelligence generated or grounded?
Grounded. No generative model sits in oddly's decision path. Every finding is computed by a deterministic rule against your own store and channel data, so it can be traced back to the rule that fired and the numbers underneath it. A language model is used only to phrase a result the engine has already decided, never to decide it.
How is this different from a marketing copilot?
A copilot asks a language model what it thinks and returns the answer. The answer can change between two identical questions, and there is no rule to inspect when it is wrong. Deterministic commerce intelligence inverts that: a fixed rule set computes the finding, the same inputs always produce the same verdict, and the record shows which rule fired on which signal.
Why does determinism matter for a store owner?
Because the system acts on real money. If a recommendation cannot be traced to a rule and a number, it cannot be checked before it is applied or audited after. Determinism is what makes oddly's other guarantees enforceable: it can only reduce or pause ad spend and never raise it, every change waits for human approval, and every change is reversible and logged.
Does oddly use language models at all?
Yes, but never where a decision is made. Language models phrase findings the engine has already computed, and lightweight models group related topics on the higher tiers. Nothing that changes your store or your ad spend runs on one.
Does deterministic mean the recommendations never change?
No. Findings change as your data changes, which is the point. What determinism guarantees is that the same data and the same rules always produce the same finding, so a change in the recommendation always reflects a change in your store rather than a change in a model's mood.
See it against your own store
Connect your store and your channels for free on Watch and see what oddly catches before you pay. Every connector is free to connect. Paid tiers act on what it finds.