RepSensAI — AI Disclosure & Automated Decision-Making Statement

Effective: November 4, 2026 · Version: 1.7 Published at https://repsens.ai/ai-disclosure

This document describes how RepSensAI uses artificial intelligence and statistical scoring, what the outputs mean, and what they cannot be used for. It is written for three audiences: the dealership deploying the platform, its HR and legal advisers, and the salespeople being measured.

It is provided so a deploying employer can meet its own obligations under laws governing automated decision-making, profiling, and automated employment decision tools. Provider is the developer of this system; the dealership is the deployer. Where such a law requires a developer to give a deployer documentation of a system's purpose, inputs, limitations, and known risks, this document is that documentation.

1. What the system does

RepSensAI measures how a salesperson performs and trains them to improve. It produces:

  1. Performance metrics — arithmetic on activity and sales data (e.g. closing rate = sales ÷ ups). Not AI.
  2. An effectiveness index and a 0–99 rating — the index is a weighted composite of the metrics that can be measured; the rating converts each rated metric to a 20–99 scale against its reference and averages them (§4.5). Statistical, not AI. Neither uses practice results.
  3. Training marks and belts — a large language model reads each practice conversation for its step's written checkpoints, quoting the salesperson's own words as its evidence. A salesperson earns a belt in each part of the sale — white through black — only on a belt test, never a score and never a peer ranking (§4.3, §4.3a).
  4. Earlier drill grades and practice ratings (records only) — until September 28, 2026 a large language model graded full practice conversations and rolled the grades into 0–99 practice ratings. Nothing produces these any more; the ones already stored are kept as dated records (§4.4).
  5. Coaching text and daily briefs — generated by a large language model.
  6. Manager and owner reports — narrative summaries generated by a large language model, including the preparation a manager reads before a weekly 1:1 with a salesperson.

2. What the system does NOT do

3. What goes in

Performance inputs — units sold, ups, leads assigned, appointments set and shown, demos, write-ups, and activity counts (calls, texts, emails). Sourced from the dealership's CRM or DMS exports, or self-reported by the salesperson where no feed covers that field.

Training inputs — the salesperson's typed or spoken responses during practice exercises and practice conversations.

Coaching inputs — what a manager records about a weekly 1:1: the focus, the salesperson's commitment and whether it was kept, and the manager's own observations of the salesperson at work.

Configuration inputs — the store's own goals and standards, set by the dealership's management.

Not used as inputs — protected characteristics; commission, earnings, or personal budgets; demographic or inferred data of any kind.

4. How the rating is calculated

4.1 Metrics. Each funnel metric is computed arithmetically and compared against a target that the dealership sets. Example defaults: 40 contacts per day; a 20% up-close rate (sales ÷ ups).

4.2 The availability rule — the most important design decision. A metric is shown, scored, and coached on only if the underlying data exists. If a store's systems cannot supply a denominator, the metric reads "not tracked" and is excluded from the index, which is then re-normalized across whatever remains.

There is a second gate: a metric that exists but has too small a sample is displayed honestly (for example "2 of 8 ups so far") but does not become a verdict and is not scored.

Consequence, stated plainly: two salespeople at different stores may be scored on different metrics, because their stores measure different things. Scores are meaningful within a store, and comparisons across stores should be treated with caution.

4.3 Belts and belt tests. The sale is divided into twelve parts — the eight on the lot, from Meet and Greet to Delivery and Follow-up, and four phone calls: the Phone Up, the Outbound Call, the Confirmation and the Unsold Follow-Up — and a salesperson holds one belt in each — white, yellow, green, brown, black. A belt is not a number, is not averaged, and is never ranked against other salespeople.

White is earned by working through that part's step on the Road to the Sale. Every belt above it is earned only on a belt test: a full conversation with a practice customer at that belt's level, with the coach silent from the first line — no notes, hints or examples, and no retrying from a line. The practice customer decides the test by the rule in §4.3a. The salesperson chooses which belt to test for: any belt above the one they hold in that part. A pass also counts toward the belts below it. One test per part per day. The two highest belts take two passed tests on separate days. A belt, once earned, is never taken away. Belts earned before this version under the earlier rules are kept.

Certification honesty. No belt, at any level, says a salesperson is ready for live customers. That requires a manager's validation — a human observation, recorded with the method used (observed roleplay, reviewed call, live shadow, outcome review, or an explicit override). The software cannot award this to itself.

4.3a Practice conversations. Two tracks put the salesperson in a full conversation with an AI customer.

The scrimmage is practice. Every part of the sale, on the lot and on the phone, plays the way a belt test does (below), with a language model coaching in the margin: the practice customer decides whether the conversation moved on, and a language model reads it for the step's written checkpoints, quoting the salesperson's own words. Phone conversations practiced before the Phone Up replaced the First Ring and Lock the Visit topics were judged once, at the end, against five written dimensions, and are kept as they were judged. There is no score, and a scrimmage never moves a belt. Management sees how many conversations a salesperson has held — never the words or the marks.

The Road to the Sale is a teaching track: each step's reading, then a practice conversation with a language model coaching in the margin. It earns no belt; working through a step earns that part's white belt.

How the practice customer decides. In a scrimmage or a belt test, the practice customer, itself a language model, is ready to move on no sooner than a set number of the salesperson's lines (ten to fifteen, rising with the belt), once the customer has opened up fully, and on a line that lands with the customer and that the reading model credits. On a call the practice customer is on the other end of the line. On the Phone Up, the Outbound Call and the Unsold Follow-Up, ready to move on means they agree to come in and say when — or, for a customer with a reason they cannot come in yet, that they agree to the next step: the numbers in writing and a time to talk again. On the Confirmation it means the visit they booked is kept, or moved to another real day and time. A line on a call that names a dollar figure the salesperson's numbers don't support, and the customer never said, is never credited. A customer on a call who needed only part of the deal, and is handed the rest of it unasked before being given a reason the visit is worth it, hangs up. A conversation that ends in a walkout or a hang-up, or reaches its line limit, does not move on.

The numbers on a call. On a call the salesperson can ask for written numbers on the caller's deal, take it to a practice desk and put a question to a practice manager. The numbers are worked out in code at practice figures — a 7% tax, $699 in fees and 7.9% financing on approved credit — and the manager's answers are a practice store's. Neither is the dealership's own, and neither is a quote for a real customer.

What the practice customer shows. In a Road conversation, a scrimmage or a belt test, the practice customer reports after every line how that line landed, and the screen shows five readings — Trust, Rapport, Value built, Patience and Commitment — as segments and words, never numbers, with the question the customer is still waiting to have answered, in their own words. They are the practice customer's reactions, read by a language model; they are not a score and not a judgment of the salesperson. Commitment is worked out in code from the other readings, not read from the model, and it fills when the customer decides: when they move on, or on a call, when they agree to come in or to the next step. Trust and Commitment decide nothing about how a conversation ends or what it earns, and Value built counts in only one: whether, on a call, a caller who needed only part of the deal hangs up when handed all of it (above). When the customer moves on is the rule above. During a belt test the readings are hidden, and they are shown when it ends. The readings are the salesperson's alone: no manager's screen shows them, and they are not part of the store's data export.

4.4 Earlier drill grades and practice ratings — records only. Until September 28, 2026, full practice conversations were graded on five rubric dimensions, each on a 1–5 scale, summed and multiplied by four into a drill score of 20–100, and drill scores rolled into per-subject practice ratings on a 0–99 scale. That grading has stopped. Stored grades and practice ratings are kept, unchanged and dated, as records the salesperson can see. They no longer move, do not feed the rating in §4.5, and are not shown to management.

4.5 The 0–99 rating. Up to three metrics are rated, each over up to the last three months: units sold, against the store's own average over the same months; and contacts per day and closing rate, against the targets the dealership sets. Each is converted to a 20–99 scale — meeting the reference reads 80, half again better reads 99 (the cap), 80% of the reference reads 60, and 40% or less reads 20 (the floor) — and the rating is the average of whichever of the three the store's data can measure (§4.2). It uses no practice results. Because units are rated against the store's own average, a rating describes a salesperson relative to their store, and ratings are not comparable between stores.

4.6 Narrative text. Coaching text, daily briefs, and manager reports are generated by a large language model from the metrics above. The model is instructed to work only from supplied numbers, but generated text may still contain errors and should be verified against the underlying figures.

5. Known limitations and risks

Stated directly, because a deployer needs them for its own assessment.

LimitationWhat it means in practice
Data quality drives everythingA wrong, late, or misconfigured CRM/DMS export produces wrong scores. Nothing downstream can detect this.
Unequal measurement between storesDifferent stores support different metrics, so scores are not strictly comparable across rooftops.
Practice ≠ the floorPractice measures a simulated conversation. It is a proxy for skill, not a measurement of real selling.
LLM marking variesThe same transcript may be marked slightly differently on different runs. Marking reduces this by asking for a yes/no judgment on one written criterion at a time — the shape a language model judges most reliably — and by requiring quoted evidence; it does not eliminate it. Treat single marks as evidence, not proof.
Belts are not a person's abilityA belt describes what one or two test conversations showed about one part of the sale, practiced in writing. It is not a rating of the salesperson, and no belt is a statement that they are ready for live customers — only a recorded manager validation is.
A rating is relative to its storeUnits are rated against the store's own average, so the same month's work can rate lower on a strong team than on a weaker one, and a newer salesperson's rating rests on a shorter record.
Assignment effectsMetrics based on assigned leads reflect how leads were distributed as much as how the salesperson worked them. Lead distribution is controlled by management.
Proxy-discrimination riskAlthough protected characteristics are not inputs, activity-based metrics can correlate with schedule, shift, tenure, language, or lead assignment — which can correlate with protected characteristics. A deployer that uses these outputs in consequential decisions should monitor for disparate impact. We will supply the data needed to do so.
Generated text can be wrongNarrative summaries may misstate or over-interpret figures. The figures govern.

6. Required human oversight

Every dealership using RepSensAI agrees in writing that it will:

Where the law of a deployer's jurisdiction requires an independent bias audit of an automated employment decision tool, that obligation falls on the employer using the tool. We will provide the methodology documentation, input definitions, and output data an auditor needs. Contact legal@repsens.ai.

7. Voice features

Where voice practice is enabled, speech is converted to text by the user's own web browser. We receive and store the text transcript and timing measurements only. We do not receive or store audio, and we do not generate voiceprints or any other biometric identifier.

This is a deliberate architectural choice: by never holding audio, the system does not process biometric identifiers of the kind regulated by the Illinois Biometric Information Privacy Act, the Texas Capture or Use of Biometric Identifier Act, or comparable statutes.

If this ever changes — if audio is stored, or a voice is used to identify a person — this disclosure and the underlying analysis must be revisited before launch.

Timing-derived measurements (for example, pause length after a price is presented, or the ratio of speaking to listening) are measurements of conversational behavior, not biometric identification. They cannot be used to identify anyone.

8. How a salesperson can question a result

  1. Raise it with your manager. They can see the underlying numbers and correct store data.
  2. Ask for the inputs. You are entitled to know which metrics were scored and which read "not tracked."
  3. Read the evidence. A training mark of met quotes the words it relied on. Marking varies slightly between runs, and a mark that looks wrong may be. A test that didn't pass can be taken again another day, and a belt already earned is never taken away.
  4. Contact us at privacy@repsens.ai for questions about the data itself. Questions about how your employer uses the results belong with your employer.

No adverse action should follow from a score alone. If you believe one has, raise it with your employer's HR — and you may tell us at legal@repsens.ai, because a deployer using our outputs that way is in breach of our agreement with them.

9. AI providers

Large language model processing is performed by Anthropic, PBC in the United States. We have contracted for terms under which data submitted for processing is not used to train Anthropic's models. We do not train any model on customer data.

10. Changes

Material changes to scoring methodology will be reflected here with a new version and effective date, and dealership customers will be notified.

Questions: legal@repsens.ai

Book a Personalized Demo