Moving Away from Gut Feelings: AI-Driven Vehicle Metrics in B2B Trade
Every trade-in involves a pricing decision worth several thousand euros, made in just a few minutes at the sales counter. AI-powered vehicle valuation in B2B trade refers to the automated determination of purchase and market prices based on completed transaction data, vehicle master data, and statistical models that continuously factor in new transactions and replace manual market research for each vehicle.
Dealerships remain the key sales channel. In 2025, 74% of all used cars in Germany were sold through independent or brand-affiliated dealerships, with independent dealerships accounting for 38% of title transfers and brand-affiliated dealerships for 36% (Deutsche Automobil Treuhand, DAT). Whoever sets the purchase price determines the margin and the vehicle’s time on the lot before it even arrives there.
This guide compares book value, portal research, and transaction-based AI valuation. It explains the calculation logic behind the pricing algorithms, outlines selection criteria, and shows where your own judgment remains indispensable.
Summary
Anyone who prices trade-ins based on gut feeling or outdated book value is giving away profit margins or losing customers to the competition. An AI-powered valuation of completed B2B transactions at Basis provides a purchase value, a channel recommendation, and comparable vehicles in seconds. Book values and portal research still have their place, but they provide average and asking prices rather than actual purchase prices. This guide categorizes the methods, outlines selection criteria, and shows where the decision ultimately rests with the dealer.
Why Gut Feelings and Book Value Cost Margins in B2B Trade
An error of 500 euros on the high side costs margin. An error of 500 euros on the low side costs the sale. Both errors don't become apparent until weeks later: either the vehicle sits on the lot too long, or the customer has signed with a competitor.
The problem is rarely a lack of experience. The problem is the data set. Anyone who bases their calculations on averages and asking prices is working with a systematically skewed picture of the market.
Book Value vs. Actual Transaction Price
A book value is a model value: an average across many vehicles, updated periodically. A price paid, on the other hand, is a specific transaction between two dealers, at a specific point in time, in a specific segment. There is a spread between the two, which, in the case of a trade-in, is directly reflected in your profit or loss.
The trend in residual values for electric vehicles illustrates just how widely segments can diverge. In March 2025, three-year-old used electric cars were worth only 51.5% of their original list price, whereas three-year-old Petrol were worth 64.0% (Deutsche Automobil Treuhand, DAT). A metric that reflects such movements with a lag leads to misvaluations in both directions when it is incorporated into the analysis.
This does not mean that book values are worthless. They are well-established and recognized for appraisals, insurance claims, and lease calculations. However, they are too imprecise to serve as the sole basis for determining a purchase price. Anyone familiar with the fundamentals of vehicle valuation in the B2B sector uses them as a reference, not as the basis for a decision.
Empirical values and individual variation
Without a standardized process, every sales consultant evaluates vehicles differently. Three consultants, the same vehicle, three different valuations: one bases their assessment on the most recent comparable case, the second on yesterday’s search results on the portal, and the third on the sales target for new-car sales.
Experience is most valuable where it is needed: in assessing a vehicle's condition, evaluating the backlog of repairs, and determining the cost of reconditioning. When it comes to determining a model's price level in a volatile market, however, it is a less reliable source because it is based on only a few observations.
The result is a problem of dispersion. Individual purchases may seem reasonable, but when you add up over a hundred vehicles, the discrepancy adds up to an amount that no one can account for in the monthly financial statements. If you want to protect your margin throughout the process rather than in individual deals, this is exactly where you start.
Time Required for Manual Portal Research
In many dealerships, it’s standard practice to spend about 20 minutes researching the market for each vehicle: mobile.de, AutoScout24, auction platforms—each one checked anew for every listing. With 60 trade-ins per month, this ties up about 20 hours of sales time that could otherwise be spent talking to customers.
These challenges come at a time when the market has already become tighter. 34% of car dealers surveyed by BCG described the supply of used cars as a major or very major challenge, while another 42% described it as a moderate challenge; and according to BCG and Wards Intelligence, the average time it takes to sell a vehicle rose from about 30 days in early 2023 to nearly 60 days in 2024 (Boston Consulting Group, BCG).
Longer holding periods mean tied-up capital and greater pressure to adjust prices. Anyone who wants to improve inventory turnover must quantify—rather than estimate—the relationship between the purchase price and the expected holding period.
How an AI pricing algorithm evaluates vehicles for B2B wholesale
A pricing algorithm decodes the FIN, assigns the vehicle to a model and trim level, searches for comparable sales, and uses this information to calculate a price range and an expected time on the market. The result is not a single number, but a range with an explanation.
Data Source: Asking Prices vs. Completed Transactions
The key difference between two valuation tools lies not in the model, but in the data source. Listing prices are asking prices, not final prices. They include room for negotiation, outdated listings, and vehicles that will never be sold.
An analysis from the electric vehicle segment illustrates just how large this gap can be: For used electric cars, the difference between the initial market price and the actual sales price in May 2025 averaged 3,000 euros (Deutsche Automobil Treuhand, DAT). Anyone basing their calculations on classified ad prices ends up paying this difference when making a purchase.
Transaction-based models, on the other hand, are based on actual transaction prices. Their reliability depends on the volume and liquidity of the underlying data source. Approximately 1.3 million vehicles pass through the CarOnSale system annually, and the overall transaction rate in 2025 was 76%. The higher the proportion of actual completed sales, the narrower the price range a model can derive from them. The white paper on European market dynamics describes the regional differences underlying this.
Model Architecture: Which Vehicle Characteristics Are Included
The calculation chain is structured similarly in most systems. It consists of five steps:
- FIN Decoding: Master data, engine type, transmission, and factory-installed equipment are automatically retrieved.
- Normalization: Mileage and date of first registration are adjusted to a comparable Basis .
- Comparison Set: The model creates a set of similar vehicles based on sales records and current listings.
- Price Distribution: From this data set, the median, range, and trend direction are calculated for this specific configuration.
- Liquidity and Holding Period: Based on the density of listings and the pace of sales, a forecast of the sales duration for each channel is derived.
The key factor is the level at which calculations are performed. Pricing algorithms operate at the market and model levels, not at the vehicle level. A stone chip, a poorly repaired scratch, or worn-out brakes do not appear in any data field.
Current Events and Market Dynamics
Models that continuously incorporate new transactions reflect market movements more quickly than lists that are updated periodically. This is particularly important when the overall market appears stable but individual segments are shifting.
This is exactly what the data for 2025 shows: The average used car price on mobile.de fell by only 0.7% in the first half of 2025 compared to the same period the previous year (mobile.de). While the average has barely changed, there are significant variations among different powertrain types and model years.
The industry is aligning its investments accordingly: 51% of automotive executives surveyed worldwide plan to invest in AI to optimize prices, discounts, and purchase incentives (KPMG). Pricing is the lever with the shortest chain of effects, as demonstrated by margin calculations in the B2B automotive trade.
What valuation methods are available to traders today?
Three approaches dominate everyday practice: traditional valuation services, portal-based market price analyses, and AI-driven valuation based on transaction data. They are not mutually exclusive; rather, they address different questions. This also applies to the procurement side, where the fragmented nature of procurement in Europe further complicates price determination.
Traditional valuation services: DAT SilverDAT and Schwacke
DAT and Schwacke are deeply rooted in Germany. Their ratings serve as a benchmark for appraisers, insurers, and leasing companies, and that is exactly what they were designed for: to provide a transparent, standardized classification of a vehicle.
The limitation lies in the logic. A reference value indicates the average value of a vehicle of this type. It does not indicate what a commercial buyer would pay today for this vehicle in this configuration.
Portal-Based Market Price Analyses
Listing data from the major portals is widely available and can be viewed for free. They are the most direct source for determining how many comparable vehicles are currently on the market.
Two distortions remain. First, these are requests, not completed transactions. Second, the wholesale market is completely missing: vehicles traded between dealers do not appear in any portal search results.
AI-Powered Transaction Price Valuation
The third approach compares B2B and B2C prices side by side and uses this information to derive a channel recommendation. Caddie AI, as an example of this approach, uses FIN input, automatically retrieves master data, and provides the purchase value, market prices for both channels, expected sales duration, and a reasoned recommendation.
The adoption of AI in the retail sector is no longer a topic for the future. In Germany, 26.0% of companies were using AI technologies in 2025, up from 19.5% in 2024 (Eurostat). AI is already widespread among car dealers: 82% of the car dealers surveyed were already using AI in some form in the second quarter of 2026, and among the dealers using AI, those that made decisions primarily based on data reported revenue growth driven by AI significantly more often than other dealers—34% versus 18% (Cox Automotive).
What to Look for When Choosing an AI Assessment Tool
The marketing claims made by providers are very similar. These tools can be distinguished based on three factors: data source, traceability, and the effort required in day-to-day operations.
Data source: prices paid or classified ads
Ask this question first, and make sure you get a specific answer. Is the model based on completed transactions, listing prices, or a combination of both, the weighting of which no one discloses?
A second criterion is the source of the transaction data. B2B wholesale prices reflect what merchants pay each other. B2C data reflects the retail market. Both are necessary if you want to make a channel decision.
Transparency: Comparison Vehicles and Rationale
A figure without an explanation is worthless in a conversation with a customer. If your sales consultant can’t explain the trade-in value, they’ll lose the negotiation to a customer who has researched their vehicle on the portal.
Therefore, check whether the comparable vehicles are visible alongside a recommendation: price, mileage, date of first registration, and number of days on the market. Visible comparison data turns a review into a selling point.
Integration and Workload for the Team
An evaluation tool must work during a sales pitch, not in a project plan. Minimal input, results in seconds, and no training required—that’s the standard.
The second criterion is the ability to integrate with your existing processes. When appraisal, intake, and marketing are handled separately, data transmission errors occur—as is often evident in typical errors in B2B transactions. Practical tips on digitization at car dealerships show how businesses can bridge these gaps.
What Caddie AI Actually Does—and What It Doesn't
An evaluation tool is judged by what it reveals and what it leaves unsaid. Both aspects are part of the assessment.
Procedure: Enter FIN; results in about 30 seconds
The master data is retrieved via the FIN; you enter the mileage and date of first registration. After that, five values are displayed side by side: purchase price, B2B market price based on completed transactions, B2C market price with median and range, expected time to sell, and demand.
In addition, there is a recommendation for how to use the channel, along with an explanation. About 30 seconds per vehicle instead of about 20 minutes of market research: that’s the calculation that makes all the difference in day-to-day business.
Model Risk Radar and Visible Comparison Vehicles
Known engine weaknesses at the model level are automatically flagged. This does not replace a thorough inspection; rather, it draws the consultant's attention to the specific areas where, based on experience, this particular model tends to incur costs.
Each recommendation is accompanied by comparable vehicles, including their price, mileage, date of first registration, and time on the road. More than 85% of the inventory on the CarOnSale platform is listed exclusively there—that is, vehicles that do not appear in any other portal searches. This is relevant for B2B pricing because these transactions would otherwise remain invisible.
Clear Boundaries: Condition, Prior Damage, Decision
The specific condition of the vehicle is not taken into account. Previous damage, pending repairs, and wear and tear are excluded, even for the comparison vehicles. The calculation is based on the market and model.
The division of roles is just as clear: Caddie does not place bids, set selling prices, conclude contracts, or communicate with your customers. The decision rests with the retailer, and that is the correct division of responsibilities. An algorithm knows neither your customer relationships nor your inventory strategy.
How to Incorporate AI Valuation into Your Purchasing Process
A tool doesn't make a difference as long as it remains separate from the process. The benefit arises when evaluation becomes an integral part of the intake process.
Standardization Across the Entire Team
Establish a policy that every trade-in is appraised before the price negotiation, without exception. This will eliminate variations among sales consultants and make results comparable across locations.
The second benefit is the ability to track performance after the fact. If the purchase price is documented, you can check after the sale to see how much the purchase price differed from the proceeds received. This figure is the benchmark against which every tool must measure up.
Compare the price estimate with your own calculation
The price estimate is the starting point, not the final result. You then factor in your target margin, the estimated refurbishment costs, warranty and liability risks, and your current inventory levels.
This often leads to the actual decision: retail or wholesale. If the retail markup barely covers the capital tied up for several weeks, selling trade-ins online is the safer option. The overview of price drivers at car auctions shows which factors drive the proceeds in this scenario.
Evaluation as a basis for discussion with the customer
Visible market data changes your negotiating position. Instead of discussing a random number, show the customer comparable vehicles along with their price, mileage, and number of days on the market. The customer sees the same information you do.
This also taps into untapped potential. Only 14% of customers received a trade-in offer during a visit to a repair shop, even though 33% would be very interested in one (Cox Automotive). Since an appraisal takes just 30 seconds, this conversation can take place at the service reception desk, and the vehicle can then be marketed to commercial buyers in Europe.
Key Points at a Glance
- Before running the test, ask each provider whether the value provided is based on paid transaction prices or on quoted prices, and make your decision accordingly.
- Make the evaluation a mandatory step before every price discussion so that the outcome does not depend on the experience and daily performance of individual consultants.
- Decide on a channel-by-channel basis for each vehicle, rather than automatically routing returns and trade-ins to retail.
- Measure success based on the difference between the purchase price and the proceeds received from the last thirty vehicles sold.
- Keep the condition assessment in-house: You’ll be the one to assess previous damage, the backlog of repairs, and wear and tear.
Frequently Asked Questions About AI-Powered Vehicle Appraisals
Does an AI valuation replace the DAT or Schwacke valuation?
No. The two sources follow different logic. Book-value services provide recognized reference values for appraisals, insurance, and leasing. Transaction-based models provide the current market price for a specific model. In practice, many businesses use both: a reference value for context and a transaction price to guide purchasing decisions.
What data do I need for an AI-powered assessment?
Generally, the VIN, mileage, and date of first registration are sufficient. The remaining master data is retrieved automatically through the VIN lookup; manual entry is optional. A complete condition report is not required for the price estimate.
Is the specific condition of the vehicle taken into account in the appraisal?
No. Pre-existing damage, outstanding repairs, and normal wear and tear are not taken into account, even for the comparable vehicles used in the analysis. The calculation is based on market and model levels. You determine the discount for the specific condition yourself at Basis based on your inspection and your preparation costs.
How up-to-date is the price data used in an AI valuation model?
That depends on the provider and should be discussed during every selection meeting. Transaction-based models rely on a constant stream of new sales rather than periodic lists. Caddie AI’s calculations are updated daily, with new completed sales and new listings added on an ongoing basis. As a result, the same vehicle may be valued differently after just a few days.
How do B2B and B2C market prices differ in the output?
The B2B price reflects what commercial buyers have actually paid in the wholesale market. The B2C price reflects the retail market, typically as a median with a range. The difference between the two is your retail markup, against which you offset capital tied up, reconditioning costs, service life, and warranty risk.
Why does a price quote differ from the book value?
A book value is a model value; a transaction-based indication is a market value. Upside and downside deviations are normal and do not constitute an error. They show how far the current price level of a model has deviated from the projected average.
Is an AI assessment worth it even if there are only a few trade-ins per month?
The time savings are realized starting with the very first vehicle, because they apply per transaction and do not depend on volume. The smallest monthly package per account includes 10 appraisals starting at 29.00 EUR net per month.
In which markets is Caddie AI available?
This offer is available to customers based in Germany and Austria. Reviews that have already been created can be viewed again in the query history without using additional credits.
Conclusion
The purchase price determines the bottom line long before the vehicle is displayed in the showroom. As long as this price is based on book values and experience, the discrepancy remains hidden until it manifests itself as downtime or a lost customer.
Transaction-based valuation brings this process to the forefront. It turns an estimate into a figure with a rationale, a channel decision into a comparison, and 20 minutes of research into half a minute. Assessing the current situation, managing customer relationships, and developing a portfolio strategy remain your responsibility.
If you want to calculate the initial costs: Depending on the package size, the valuation costs between 2.90 EUR and 1.75 EUR (net) per transaction. Test Caddie AI on the last thirty vehicles you’ve taken in, and compare the estimated value with the actual proceeds.
