VARTEQ

Automotive case study

Predictive Scoring and a Shared Workflow for Vehicle Acquisitions

A US vehicle auction operator needed a more consistent way to evaluate inventory on Manheim, Mbondemand, and other sources.

At a glance

Project type
Vehicle acquisition platform
Industry
Automotive
Location
United States
Services
AI Development, Custom Software Development
Acquisition costs
40%lower
Buying decisions
60%faster
Delivery
3months, plan was 8

Overview

The Summary.

A US vehicle auction operator needed a more consistent way to evaluate inventory on Manheim, Mbondemand, and other sources. Although it had extensive auction history, buyers still checked prices manually and used different criteria when reviewing deals.

VARTEQ built an internal acquisition platform that brings auction records into one review process. Predictive scores help buyers evaluate deals, while purchase outcomes feed back into subsequent assessments. Data, review steps, and AI assistance operate within the same application.

About the project. A vehicle auction operator whose buyers assess inventory across multiple auction sources.

Engagement length. 3 months to delivery against an 8-month plan.

01Where they started

The challenge.

The operator’s buyers worked across auction sites, spreadsheets, and email. Historical data was available, but pricing checks remained manual and evaluation criteria varied between teams and regions.

The acquisition team needed to:

  • Collect listings from multiple sources in one queue, reducing the need to switch auction platforms.
  • Apply common scoring criteria using historical results across buyers and regions.
  • Move deals through review, price checks, and approval without manual handoffs.
  • Record purchase outcomes and use them to refine future scores.

Project Highlights

  • Shared scoring and review rules made results comparable across regions.
  • Expert assessments were recorded during review and used to refine scoring.
  • Scheduled processing updated listings without manual refreshes.
  • API validation and consistent error responses supported reliable data exchange.
  • Task-specific agents automated repeatable steps within the acquisition workflow.
02What we built

The solution.

The internal platform combines auction data, predictive models, buyer review, and procurement actions. Reusable workflow and automation components support the acquisition process.

  1. 01

    Deal Scoring

    • Machine learning models use historical auction records to identify potentially valuable purchases before manual review.
    • A reinforcement learning process updates predictions using actual purchase outcomes as well as historical training data.
    • Experienced buyers annotate edge cases during review. Their recorded feedback is used to tune scores in later deal cycles.
  2. 02

    Review and Acquisition Workflow

    • Listings from Manheim, Mbondemand, and internal sources appear in one prioritized queue.
    • Each deal has a visible status as it moves through review, price checks, approval, and action.
    • Purchases are linked to later performance, allowing managers to compare results across buyers, regions, and vehicle categories.
    • Common scoring rules make regional results comparable without relying on each buyer’s individual method.
  3. 03

    AI Assistance

    AI handles supporting tasks that save time while preserving the established decision rules:

    • Generating structured deal summaries from auction records.
    • Tagging and grouping listings by vehicle type, condition, and risk.
    • Adding internal notes to the existing deal records.
    • Optional agent support covers summaries, tags, and structured notes within the workflow.
  4. 04

    Data Processing and Storage

    • Background jobs collect listings and price changes on a schedule.
    • API validation and standardized errors keep auction data exchange consistent.
    • Auction records, model inputs, decisions, and results are held in a shared data store.
  5. 05

    Delivery

    The team first documented how buyers decide: which sources they trust, which signals they check, when they reject a deal, and when approval is required. Defining successful purchase outcomes established a common basis for evaluation across regions.

    Auction data, score inputs, decisions, and outcomes were then brought into one application. Buyers could record the reasons for a purchase and track its performance. Managers could compare activity across teams without manually combining reports.

    An initial scoring model was refined using purchase results and buyer feedback. Reinforcement learning incorporated outcomes, and human reviewers assessed cases the model could not handle reliably.

    Agent assistance was added for deal summaries, listing tags, and internal notes linked to existing records. These tasks reduced administrative work without changing the acquisition decision rules.

Services on this project

Technologies used

  • Machine learning
  • Reinforcement learning
  • Web APIs
  • Manheim and Mbondemand data sources
03What changed

The outcome.

  • 40%

    Lower Acquisition Costs

    Predictive scores filtered low-value opportunities earlier in the process, giving buyers more time for deals with stronger return prospects. Automated checks and approvals also reduced acquisition administration.

  • 60%

    Faster Buying Decisions

    One review queue replaced repeated switching between auction sites, spreadsheets, and email. Deals that previously took days to assess moved through review in hours.

  • 70%

    Improvement in Prediction Accuracy

    Historical data, actual purchase outcomes, and expert feedback improved prediction accuracy compared with manual assessment. Each deal cycle supplied further evidence for model refinement.

  • Delivered 5 Months Ahead of Plan

    Reusable workflow, dashboard, and API components reduced the infrastructure work required. The project was delivered in 3 months against an 8-month plan.

More Client Stories

All case studies

Start the conversation

Facing a Similar Challenge?

Tell us what you're building or fixing. You'll meet the engineers who would work on it before you sign anything.