The Signal Engine
How the Signal Engine works.
The Signal Engine is a patent-pending methodology that converts nontransactional engagement activity into explainable, product-specific readiness scores and recommended actions for institutional staff.
It uses deterministic, explainable behavioral scoring and does not depend on generative AI to calculate product readiness.
Patent Pending | U.S. Provisional Application No. 64/087,878
Why nontransactional behavior matters
Transaction data describes decisions a household has already made. By the time a new loan or account appears in core records, the choice of provider is usually settled.
Engagement activity happens earlier. When a student completes a lesson on credit, a parent works through a module on college savings, or a supporter makes a contribution, that activity can indicate an emerging financial need before any account is opened anywhere.
The Signal Engine treats these supported engagement events as signals. It does not use core banking transactions to score readiness.
Why household participation is useful
Financial decisions are often made by households, not individuals. College Cash links students with the parents, grandparents, and other supporters they invite. Many of those supporters may not currently be members of the institution.
Combining signals across linked participants produces a household view that no single person's activity would show on its own.
The Methodology
Eight stages from engagement event to institutional action.
-
STAGE 01
Behavioral signal collection
The platform records supported engagement events, such as financial education module completions, savings milestones, scholarship activity, StackU decisions, and supporter contributions. Events come from students and supporters, each in their own role.
-
STAGE 02
Product-intent mapping
A configurable mapping connects each signal type to the financial products it relates to. One signal can relate to several products, and each product can draw on several signal types.
-
STAGE 03
Weighted signal assignment
Each signal type carries a numerical weight that reflects how strongly it relates to a product. Not every behavior carries equal significance. Weights can differ by participant role, and an institution can have its own reviewed weights where they differ from the standard configuration.
-
STAGE 04
Temporal relevance processing
Recent activity counts more than older activity. Each product has its own decay rate, so a signal's contribution declines over time and older signals eventually stop counting toward the score.
-
STAGE 05
Multi-stakeholder household aggregation
Signals are combined across the linked participants in a household: students and the supporters connected to them, whether or not those supporters are current members. The result is a household-level view of emerging needs.
-
STAGE 06
Product-specific readiness scoring
The engine calculates a readiness score for each product. Households are ranked so staff can see where to focus first.
-
STAGE 07
Institutional action recommendations
Staff see the recommended product, the reasons behind it, and a suggested outreach script. Each readiness card shows no more than two products, to keep outreach focused.
-
STAGE 08
Closed-loop outcome tracking
Staff record what happened after outreach, such as an account opened, an offer declined, or not yet. Those outcomes feed the review process described below.
How recommendations are explained
Every readiness score can be opened to a "Why this score" view. It lists the engagement events behind that product's score, so staff can see what the household actually did before they reach out.
How audit information is preserved
Scores are calculated by fixed rules from recorded events, so the same inputs produce the same result. Every change to a signal weight is written to an append-only change log that records what changed and when.
How reviewed feedback works
Recorded outcomes are used to generate suggested weight adjustments, for example when households with a particular signal open a product more or less often than expected.
- Suggestions never change a weight on their own.
- Each suggestion must be reviewed and approved by an authorized administrator before it takes effect.
- Approved changes apply only to the institution they were generated for, and each one is logged.
Outcome-based suggestions depend on the institution recording outcomes, and on enough recorded history to compare. Review arrangements are agreed with each institution as part of deployment.
Why core scoring does not depend on generative AI
Readiness scores are calculated with deterministic rules: mapped signals, weights, and time decay. No large language model is involved in calculating a score, which keeps results consistent and explainable.
Other parts of the College Cash application, such as optional coaching features and content moderation, use generative AI where it is appropriate. Those features are separate from readiness scoring.
Where the signals come from
The Signal Engine scores engagement captured through the College Cash platform. Institutions that want to use it alongside existing engagement programs can discuss it where the required event data and technical access are available. We scope that work with each institution rather than assuming a ready-made connection.
See the Signal Engine on a working platform, with household readiness cards, score explanations, and recommended actions in a simulated institution.