AltVector — CEO Q&A Brief

We sell a decision layer: a small set of portfolio rules that turn your existing data into actionable lending policies — approval, pricing, and guardrails — with measurable outcomes. In a pilot we deliver it as a contained package (policy logic + cohort definitions + measurement). After proof, it becomes an API/service that plugs into your stack.

Scorecards and ML models predict risk. We focus on what happens when you change a decision: which actions reliably improve outcomes, for which cohorts, and under what limits. The output is not just a score — it is a policy you can execute, plus a measured trade‑off curve and a safe operating point to deploy.

The fastest proof is usually False‑Negative Revival: identify good borrowers hiding inside rejects, approve them under tight caps, and track performance versus a matched baseline. It’s operationally simple, shows a clean ‘lift story’, and it translates directly into growth without loosening standards.

We commit to jointly defined primary metrics — for example: approval lift at a fixed bad‑rate ceiling, collection rate improvement, or loss reduction — plus secondary metrics like ROA/ROE proxies, early delinquency, and vintage stability. We report the full trade‑off curve and recommend the safest operating point, rather than a single headline number.

We run with guardrails by design: cohort caps, channel/branch whitelists, exposure limits, early‑bucket monitors, and stop‑loss triggers. If any metric crosses an agreed threshold, the pilot pauses automatically or reverts to baseline. This makes experimentation controlled, reversible, and audit‑friendly.

For a pilot, a secure extract is usually enough: application and rejection history, key borrower and product fields, pricing, collections outcomes, and timestamps. We start with batch files so your core systems stay untouched. API integration comes after value is proven and governance is agreed.

Everything is versioned: policy logic, cohort definitions, thresholds, and roll‑out configuration. Each decision can carry an audit_id and a human‑readable rationale, so reviews are straightforward. We keep a decision log, a measurement log, and a rollback plan — the same ingredients regulators and internal audit expect.

We watch drift the way a credit team does: performance by cohort, channel, and time, plus leading indicators like early delinquency and acceptance mix. When drift appears, we tighten caps, freeze learning, or revert to a safe policy until stability returns. The system is designed to degrade safely, not silently.

A pragmatic timeline is 6–8 weeks for a controlled pilot: define outcomes and baselines, prepare data extract, run analysis, configure guardrails, and start a capped roll‑out. Early indicators show within weeks; vintage confirmation follows your normal reporting rhythm.

We can do fixed pilot fee, implementation fee, success share — or a hybrid. In practice, leadership prefers something that aligns incentives and is easy to audit: a modest fixed component plus a clearly defined upside share tied to verified portfolio deltas.

Then we stop and we do not scale. The pilot is intentionally small and reversible, so the cost of being wrong is bounded. A no‑lift outcome is still useful because it prevents an uncontrolled roll‑out and clarifies where the constraint really is (data, policy, channel mix, or operations).

You retain your data and outcomes. We deliver your policy artifacts — cohort logic, thresholds, guardrails, and measurement spec — and we retain only reusable tooling and non‑portfolio IP. If you want, the policy artifacts for your book can be contractually treated as your deliverable.