AI Transparency — How Louie Works

Built to assist.
Never to decide.

Every dealership deserves to know exactly what its AI is doing, why, and who remains in control. This page answers those questions without marketing language.

Section 1

What data Louie uses

Louie draws exclusively from data that Louie Auto Group already owns and operates. Nothing is scraped or purchased from outside sources without an explicit integration that the dealership configures.

  • Live inventory database — all vehicles on the lot with cost, age, reconditioning status, and current asking price
  • Deal history — closed transactions from SalesLogs and SalesHistories: front/back gross, lender, rate, term, F&I products sold
  • Lead and CRM activity — inbound leads, follow-up timestamps, stage progression, BDC call logs
  • Floor and traffic signals — active walk-ins, current desk activity, appointment queue
  • Market data — NHTSA recall feeds, Federal Reserve economic indicators (FRED API), MarketCheck comparable listings for live price-to-market calculation
  • Weather and calendar context — local conditions and day-of-week patterns that affect floor traffic, factored into staffing recommendations

Louie never accesses individual consumer credit files, social media accounts, or any data source not explicitly listed here. There is no "shadow" data collection.

Section 2

How recommendations are generated

Louie's brain uses advanced AI reasoning as its foundation. At each decision point, the system assembles a structured context packet — current inventory state, relevant deal history, applicable business rules, and market signals — and submits it to its AI engine with explicit instructions about the dealership's operational goals.

Every recommendation carries a confidence tier so staff always knows how much weight to give it:

High confidence
Data-backed

Multiple corroborating data points. Example: a vehicle with 90+ days age, falling market comps, and zero recent activity.

Moderate confidence
Pattern-based

Fewer data points or mixed signals. Louie flags uncertainty and explains its reasoning inline so the manager can weigh it.

Louie does not generate recommendations without a stated reason. Every action card includes a plain-English explanation of what triggered it. There are no black-box outputs.

Section 3

Human-in-the-loop controls

Louie is an advisor, not an actor. The three highest-stakes workflows are explicitly gated behind human approval and cannot be triggered automatically:

Price changes require manager sign-off. Louie surfaces a recommended price adjustment with full supporting rationale (age, market position, gross impact). A manager must tap Approve — the price does not change until they do.
Mass texts require explicit blast approval. Louie can draft a customer segment and message, but bulk SMS is only sent after a staff member reviews the recipient list, reads the message text, and clicks Send. There is no scheduled auto-blast.
AI responses can be rejected and corrected. When Louie drafts a customer-facing reply, the salesperson sees it before it goes out and can edit, reject, or send as-is. A rejection teaches Louie the correct approach for future similar prompts.

Louie has no ability to independently wire funds, submit lender applications, post online listings, or contact customers without a staff member initiating the action.

Section 4

Privacy and security architecture

Louie Auto Group's data stays within Louie Auto Group's infrastructure. The system is designed so that the data never needs to leave.

  • AES-256-GCM encryption at rest — every uploaded document (credit apps, stips, contracts) is encrypted before it touches disk. Plaintext never exists on the server.
  • Private VPS, no shared cloud tenancy — the application runs on a dedicated server (143.198.6.154). There is no multi-tenant SaaS layer where another dealership could access your data.
  • No third-party data sharing — dealership data, customer records, and deal financials are not sold, licensed, or transferred to any party outside Louie Auto Group.
  • AI inference is stateless — each API call to Louie's brain contains only what is needed for that specific task. Dealer data is never used to train any third-party models.
  • JWT-gated access — every internal route requires a valid JSON Web Token. Sessions expire and are not persistent across browser restarts without explicit re-authentication.
  • Role-based data visibility — sales reps see only their own deals and leads. Managers see their team. Owners see the full store. No role can escalate its own access without an admin change.
Section 5

Accountability and audit trail

Every action Louie takes — and every action a human takes in response — is recorded. This is not a log that exists for compliance optics. It is an operational record that the dealership can use to audit, retrain, and improve.

  • Each Louie recommendation is stamped with a timestamp, the data signals that triggered it, and the confidence tier assigned at generation time
  • Human approval or rejection is recorded alongside the staff member's name and timestamp — there is no anonymous override
  • Outcome tracking ties each recommendation to its downstream result: did the price drop close the deal? Did the follow-up text get a response?
  • The full audit trail is accessible in the Tech Support console to the owner and admin roles at any time — no special request or vendor involvement required

If Louie makes a bad call, the dealership can see exactly what data it had, what it was told to optimize for, and where the reasoning went wrong. Accountability is built in, not bolted on.

Section 6

Data retention policy

Retention periods are set to match the operational value of the data — long enough to improve Louie's accuracy over time, bounded enough that stale data does not skew recommendations.

Conversation history
12 months

Staff–Louie chat retained for one year. Enables pattern learning and supports dispute resolution if a recommendation is later questioned.

Encrypted file uploads
Until deleted

Credit applications, contracts, and stips are retained encrypted until an admin manually deletes them. The dealership controls the lifecycle.

AI action log
Indefinite

The audit trail of what Louie recommended and what humans decided is retained permanently as an accountability record.

Session tokens
24 hours

JWT access tokens expire after 24 hours. Users must re-authenticate. No persistent "remember me" tokens exist in the system.

Our commitment

The standard we hold ourselves to

AI in the dealership is still new enough that most vendors avoid explaining how it works. We believe that opacity is a feature dealerships should not accept, especially when the AI touches customer communications, deal structure, and pricing.

This page is not marketing. It is a technical and operational description of exactly what Louie does. If anything on this page is inaccurate or incomplete, we want to know. The people who work this floor every day deserve a tool they can fully understand and fully trust.

Last reviewed: June 2026  ·  Contact: admin@louieauto.com

Report Issue