Your funnel has opinions. This layer has evidence.
Every lender believes things about what converts: about transparency, about pricing presentation, about which leads are worth a call-back. The intelligence layer (we call it Datum) turns those beliefs into measured signals tied to real outcomes, so you make the next experience decision on evidence.
When we say “intelligence”, we mean measured. Nothing here is a prediction. Every signal on this page is a cohort we compute from outcomes your own files produced. No artificial intelligence (AI) or machine learning (ML) model produces, influences, or ranks any output that determines a regulated outcome: not credit, not pricing, not valuation, not adverse action, not a disclosure, and a continuous integration (CI) gate fails our build if a model endpoint ever becomes reachable, by import, from the code that decides one. Our AI governance posture →
Where it stands today. We’ve built the framework, the event taxonomy and the cohort math. No lender is live yet, so the signals in a walkthrough run on illustrative data, and we label them that way. The funded-loan half waits on your loan origination system (LOS) milestone feed.
Evidence signals
Does a borrower who opens the “show me the math” panel submit more often? Does adopting the automated value estimate change completion? Does a loan officer (LO) pricing off par close better files? Each signal is a with/without cohort against submitted and funded outcomes. We label each one honestly and flag the thin cohorts.
An experiment framework
Experience variants run as assigned arms with deterministic bucketing, and every outcome event carries its arm. Correlation isn’t causation. We built the framework so a promising signal can graduate into a controlled test before it becomes your policy.
Funnel & source economics
Attribution rides the file from first click. We built it to carry through the LOS milestones once you connect that feed, so you can compare lead sources on what they actually produce: submittable files and funded loans.
Re-engagement
A stalled journey enters a timed follow-up cadence that carries the context the borrower left behind, so they hear about the file they started. They don’t get a cold blast. Our suppression rules structurally keep completed and opted-out borrowers out of the queue.
Measured against the loan file, not the pageview
The layer’s spine is our own event taxonomy: a small, disciplined set of journey events that carry the outcome, the experiment arm and the evidence signals on every record. The same platform runs the borrower journey and pushes the LOS file, so once you connect the milestone feed, the signals join to loan milestones without a data-engineering project.
A leadership view, and it isn’t surveillance. The console reports cohorts and talk tracks, meaning what your best-performing experiences and officers do differently, and it preserves borrower privacy. It shows references and statuses, never raw borrower payloads. More on the data posture →