An ML platform that predicts each customer's upcoming purchases from behavioral history, runs a vendor matching optimization to find relevant discounts, filters the match against preference rules (including distance), and triggers delivery via internet or SMS fallback.
Businesses send the same promotional offers to every customer, regardless of purchase timing or expressed preferences. Meanwhile, financial institutions have rich transactional histories that signal exactly what a customer is likely to buy next — and when. That signal has never been systematically used to route relevant vendor offers.
A computing platform learns each user's purchase activity pattern from historical data, identifies upcoming anticipated purchases, then determines — via an ML optimization algorithm — a vendor whose sales offering includes a discount matching that anticipated purchase.
Before triggering the action, the platform retrieves the user's pre-stored preference rules (including distance willing to travel) and confirms they apply to the purchase attributes. Only then does it route the offer — using internet or SMS fallback depending on connectivity.
The platform operates in two simultaneous tracks: a user behavioral track (learning purchase patterns from historical activity) and a vendor track (receiving and cataloguing current sales offerings). The ML optimization algorithm runs at the intersection — producing a match that satisfies both sides.
Preference rules act as a gate: a valid match that violates a user constraint (e.g., the vendor is too far) is not triggered. A time-based expiration rule can also block offers that are no longer within the user's defined window.
An intelligent chatbot initiates an interactive session with the user to collect preference rules via NLP. These rules are stored in a repository and retrieved at matching time — ensuring the system never surfaces an offer the user wouldn't accept.
The patent explicitly names distance as a preference rule attribute — a geographic constraint that eliminates irrelevant local offers before they reach the customer.
The matching pipeline is explicit in the patent claims — it's not just a recommendation engine. The ML optimization algorithm identifies optimized resources; the preference rule gate confirms the match is valid for this specific user; and the time expiration check confirms it's still timely.
The platform doesn't treat all anticipated purchases the same way. Preference rules, vendor proximity, and the user's interaction history modulate what gets matched and how it's delivered. Select a scenario below to see the full outcome.
Like the companion patent (P15 — ML Account Management), the connectivity check is not a fallback workaround — it's a structural element of the core claim. The platform must determine connectivity before triggering, and must use one of the two communication modes to deliver.
This dual-mode design ensures that customers without internet access at the moment of purchase-pattern relevance still receive the matched offer — via SMS — without any manual intervention.
The combination of behavioral prediction, vendor matching, and preference-rule gating enables a new category of commerce infrastructure — one where offers arrive at the right moment, for the right product, from the right vendor, filtered by what the customer already said they want.
This patent has been cited by Intuit and a Chinese industrial supply platform — indicating that the customer-vendor purchase-matching framework is being applied both to individual consumer applications (Intuit's recipient attribute prediction) and supply-chain procurement optimization.