News 
With the market showing signs of stability – or even a slight dip – knowing how much to bid is more important than ever. While we already offer a great tool to estimate the value of any classic or exotic car (have you checked it out yet?), we are always working on ways to improve our data and our tools.
With this in mind, we have been testing some AI improvements to CLASSIC.COM in the background all year, and we are excited to test those efforts in real-time!
Updated August 23, 2024 with results. Individual results included for each category.
Skip to the overall result here.
The Estimation Game: AI vs. Comparable Listings
We are challenging ourselves to predict a range of hammer prices for select Porsches at Monterey using two different data-driven methods, and one human method:
- Comparable Listings (“Comps”) Estimate: This is our standard method of comparing the listed car to similar cars that have recently sold. We use this today in our Garage by calculating a value range based on comparable sales and applying a statistical calculation, to account for market variations.
- AI Estimate: The new method we are testing uses a Machine-Learning AI model designed and fine-tuned with our team’s expertise produce a similar estimated range by leveraging historical sales data combined with key value drivers.
- Human Estimate: Many may not realize, but we employ a team of humans at CLASSIC.COM who are in charge of reviewing and curating all of the listings in our search engine. We asked our Porsche curator, Pablo, to give his estimate as well – just for fun.
We’re eager to see how our estimates compare to each other and to the official auction venue projections. This is an opportunity to put our methods to the test and see how they perform in a real-world auction setting.
Below, you’ll find select Porsche 911s set to be auctioned in Monterey, separated by generation along with their estimated value ranges using each of our approaches along with the venue estimate. Once the auctions conclude, we’ll analyze the results and post them here.
We are excited to see what happens, and we hope you are too!
Porsche F Body SWB
The value of SWB (Short Wheelbase) cars is influenced by a variety of factors, many of which are highly subjective. These include the car’s condition, whether it has matching numbers, the preservation of original details, the availability of documentation, and other unique characteristics that can significantly impact the car’s market value. For our analysis, we selected two 1965 Porsche 911s—one a ‘survivor’ and the other a restored example, both with matching numbers—as well as a restored 1967 911S with non-matching numbers.
At first glance, our AI and Comps estimates reveal a broader value range compared to the venue estimates, highlighting the complexity and the variety of subjective factors that can affect these cars’ value. For the 1965 models, the restored car is anticipated to fetch a higher price at auction, largely due to its superb restoration quality, excellent condition, and well-documented history. Interestingly, our human expert believes that while the car may not meet the venue estimate, it could potentially exceed both the AI and Comps ranges, with a human estimate set at $240K. Conversely, the survivor car, though well-preserved and highly regarded by collectors, is not in “time-capsule” condition but is instead a well-maintained original driver. This suggests it may not reach the venue’s expectations but should still perform well, likely landing in the upper half of both the AI and Comps estimates.
The 1967 911S, despite its non-matching numbers—a critical factor in the valuation of these models—is a high-quality restoration. However, it is expected to sell near the lower end of the Comps range, slightly below the venue’s estimate, with a projected value of $210K.
Only the 1965 survivor sold, and almost exactly at our human-prediction value, well within the three estimates and right near the middle of both the Comps and AI estimates, albeit closer to the lower end of the range.
For the restored 1965 car that didn’t sell, the high-bid was 4% from our human’s estimate, and close to the higher end of the AI and Comps estimate, but well below the lower end of the venue’s estimate. As we pointed out before the auctions, when the venue’s estimate is placed so above the AI and Comps ranges, it could be a sign that the car probably won’t meet its reserve, like this one.
Finally, the 1967 911S failed to sell (we could argue it was the lack of matching-numbers engine) and had a high-bid well below all of the estimates, although the AI range was the only one that contained it.
F Body LWB (excluding RS 2.7)
Long Wheelbase (LWB) F Body 911s follow similar pricing patterns to their Short Wheelbase (SWB) counterparts, with values heavily influenced by subjective factors such as condition, originality, and provenance. However, because LWB models are more common, the market tends to be more clustered, leading to narrower value ranges. This trend is evident in our analysis, where the estimates for LWB models—whether from the venue, AI, or Comps—were generally tighter compared to those for SWB cars. Notably, in two out of the three cars we analyzed, the venue estimates were significantly higher than both the Comps and AI estimates.
Venues may be overestimating the value of the 1972 911T Coupe and the 1973 911E Targa, likely due to consignors setting reserve prices too high. While the true outcome will be revealed once the auctions conclude, these discrepancies could serve as strong indicators for evaluating reserve-setting practices. However, if these cars sell at or above the venue’s projections, it might suggest that our models underestimated certain subjective value drivers—like exceptional condition or unique history. Such insights would be valuable for refining our algorithms in future assessments.
In this particular case, the blue 1972 911T comes from a notable collection and is in superb restored condition, while the orange Targa is a two-owner original car featuring a recently rebuilt engine. Despite these positive attributes, we believe they may not be sufficient to justify the significantly higher venue estimates. We predict the blue 1972 911T will sell for around $120K, and the orange Targa for approximately $150K.
For the 1971 911S, the venue’s estimate is broader and more closely aligned with the Comps and AI estimates, suggesting that this car’s specifications and condition are more typical of what the market values—practical, enjoyable vehicles rather than the ultra-high-quality show cars often seen at Monterey. Given this alignment, we anticipate the 911S will perform within its expected range, with our in-house prediction set at $180K.
On one of the three cars, the 1971 911S, the AI estimate had a 0.5% error, Comps had a 5% error and our human estimate saw a 7% error, so this was possibly the best result against our three predictions.
On the other hand, we had one of our worst predictions for the ‘72 911T: with everyone missing the price point entirely, but as we saw during the Broad Arrow live auction, it was the result of a bidding-competition, and therefore we consider it to be an outlier data point without much more to explain for the 100%+ error on all estimates.
For the 1973 Targa, our human estimate was relatively close the the highest bid, yet this one fell above the Comps and AI ranges, yet fell short on the venue’s estimate. This car showed that a super high-quality car could be the most challenging listings for the data-driven estimates, and that even when our human estimate accounted for this, it still fell short of the seller’s expectations.
Carrera RS 2.7 Touring & Lightweight
The 1973 Carrera RS 2.7 represents the pinnacle of early F Body Porsche 911s and is widely regarded as one of the most prestigious models in the marque’s history. This distinguished car is offered in two distinct variants: the Touring, with 1,308 units produced, and the Lightweight, of which only 200 units were made, rendering it nearly twice as valuable. The valuation of these vehicles is contingent upon several critical factors, including condition, provenance, historical significance, and accompanying documentation. At the upcoming Monterey auctions, five Carrera RS 2.7s will be presented—two Tourings and three Lightweights—each providing an exceptional and rare opportunity for discerning collectors during Car Week.
Carrera RS 2.7 Touring
For the white Touring car, the venue’s estimate falls right in the middle of the Comps estimate and in the lower range of the AI estimate, between $600k and $750k. Although both the Comps and AI ranges are broader, they still encompass the venue’s estimate, suggesting that it may be well-aligned with market trends—or perhaps, it’s just a coincidence. This car is being sold without reserve, which could explain why the venue’s estimate appears more closely tied to the market. Given its interesting provenance, though not extraordinary, our human prediction was set at the intersection of the three estimates: $620k.
The orange car, on the other hand, also has notable prior ownership but has been slightly modified with Lightweight-style bumpers and tweaked suspension, designed for enhanced driving enjoyment. While these modifications are minor, they can still influence its value. In this case, the venue’s estimate is at the higher end of our Comps range, whereas the AI model estimates the value on the opposite end, well below the venue’s range. Given that it isn’t a top-market example, we believe it should still sell at the higher end of the Comps range: $560k.
Only the white RS Touring sold, and at a price just below the AI and venue’s estimate but very close to the Comp’s mid-point estimate, and only 15% below from our human prediction.With the venue’s estimate so close to the data-driven ranges, it was a sign that the expectations for the car were very well aligned with the market, and therefore sold just below the Comps estimates, but not that well when compared to the AI estimate.
On the other hand, the orange car failed to sell, yet the highest bid was right in the middle of the Comps range, above the AI range and well below the venue’s and our human estimate. But, it showed that the AI was able to predict that this car was going to be valued below the venue’s expectations, possibly due to the modifications.
Carrera RS 2.7 Lightweight
The Lightweight 2.7 RS models are considered the holy grail of F Body (long-hood) Porsche 911s, embodying the pinnacle of what collectors seek: performance and rarity. With only 200 examples ever produced, it’s remarkable that three of them—representing 1.5% of the entire production run—will be offered at Monterey this year.
The venue’s estimates for these cars are notably the broadest we’ve seen, reflecting the unique subjective value drivers that often influence RS valuations—factors that may not be fully captured by AI and Comps estimates. The two cars with the widest estimate ranges, and the highest upper ends, are meticulously restored examples with excellent documentation, yet they lack special or notable provenance, which can significantly affect their value. For these two cars, our human estimate aligns more conservatively at $1 million each, reflecting a more tempered view of their market potential.
The third car, an orange model with RSR components and a richer history, has a higher lower-end estimate, suggesting that its provenance is driving a higher valuation from the venue. Our human estimate for this car is $1.2 million, recognizing its enhanced appeal due to its history and modifications.
However, with cars of this caliber, outcomes can be unpredictable. The final sale price will depend heavily on pre-auction inspections, thorough examination of the documentation, and the bidders in the room at the time of the auction.
None of the Lightweight RS sold, and overall the AI estimates did better than the venue, the Comps estimate and our human for all three cars. We won’t take into consideration for our analysis the white car, as there was an addendum to the sale (the catalog failed to note the engine was not original to the car), therefore even though the high bid was well within venue’s the estimate range, the car ended up not selling.
For the other two cars, our human estimate for the RSR-parts car was right at the highest bid, above the AI estimate and below the venue, but close to the Comps estimate mid points. Again, we saw that when the venue estimate is so above the AI estimate, it is likely that the car won’t sell.
On the other orange car, the AI range predicted the high-bid very well, yet comps and venue estimates were a bit higher. Our human estimate recognized that the comps and venue ranges were a bit too high, yet also over-estimated the car by $150k (15%).
G Body – Turbo (Early 930)
We selected three relatively standard early 930s for analysis. The first is a 1977 model, which shows the largest discrepancy between the AI, Comps, and venue estimates. This car is particularly notable for its exceptionally low mileage and well-documented history, yet the venue’s estimate is significantly higher than both the AI and Comps predictions. Despite the car’s unique qualities, our human estimate takes a more cautious approach, predicting it will sell for $240k.
For the 1979 and 1980 models, the venue estimates are also higher than the AI estimates, though the gap is much smaller compared to the 1977 car. The AI estimates for these models exceed the Comps estimates, possibly indicating that these cars are positioned at the upper end of their market spectrum due to subjective factors not fully captured by the AI or Comps models. Alternatively, it’s possible that both the venue and AI estimates are slightly overvalued given the auction context. Our human estimate for the 1979 930 is set at $200k, while for the 1980 930, it’s estimated at $185k, reflecting a more measured and realistic view of these vehicles’ market potential.Notably, the 1980 930, like the 1967 911S mentioned earlier, is being offered without reserve, which probably explains why the venue estimate aligns more closely with the AI and Comps ranges.
The 1977 car proved to be another example (perhaps the strongest one), of the venue over-estimating the value of the car: 42% error. Our human estimate had the smallest error with this car, yet the Comps also had the highest bid within its range.
For the 1979 car, our human estimate was also close, but the AI estimate performed the best. Even though the venue slightly overestimated the value of the car, it was interesting to see that the comps had the highest error.
Finally, for the 1980 car, the AI and comps estimates had the smallest errors, while our own human estimate and the venue slightly overestimated the car, with 27% error each.
For the two cars that sold, it was the AI that did the best.
G Body – Turbo Slantnose
Three prominent slant-nose examples of the 930 were selected for our continued estimate analysis. These include two sub-10k mile cars (one coupe and one cabriolet) and a 70k mile coupe, all finished in red over beige. As observed with other vehicles, the venue’s estimates for these three cars are positioned at the higher end of the ranges provided by the Comps and AI models, with the Cabriolet and the higher-mileage Coupe showing the most significant discrepancies between the venue’s estimates and those generated by the Comps and AI models.
For the 8k-mile Coupe, the estimates across all models are relatively close, although the AI estimate is noticeably wider than the Comps and venue estimates. This greater variability in the AI estimate invites further analysis to understand the underlying factors contributing to this spread. Considering the relatively tight range and recent sales data, our human estimate for the 8k-mile Coupe is set at $240k.
The Cabriolet, being a 1989 model, benefits from the desirable G50 transmission, which adds to its value, as reflected in the ’89 Comps. However, the Comps algorithm indicates considerable variability in the price, more so than with the Coupes, while the AI estimate significantly underestimates the car’s value compared to the venue’s estimate. For this Cabriolet, our human estimate is positioned at $280k.
For the 70k-mile Coupe, which falls into one of the higher mileage tiers in this market, both the Comps and AI suggest it will sell for less than the 8k-mile Coupe. This suggests that the venue estimate may be overly optimistic, and the Comps are likely more accurate in this case, while the AI estimate, being quite broad, appears to cover a range that bridges the gap between the venue and Comps estimates.
We were surprised to see that the 70k-mile 1987 coupe didn’t sell for a price well above our Comps range and our human estimate. However, the high-bid landed short on the venue’s estimate yet close to the higher end of the AI estimate. AI performed the best in this case, but once again proved that when one of the data-driven estimates has such a gap vs the venue estimate, the car probably won’t sell, especially when the cars lack a super-strong value driver.
The other two cars, the sub-8k mile cars, sold on the higher end of the venue estimate range, signaling perhaps that the market is stronger for the low-mileage, more recent 911’s.
964 – Turbo 3.6 and Carrera RS
The 964 Carrera RS is one of the cars with the closest estimates among the three we analyzed, suggesting that the car is likely to sell within its expected range. Our human estimate for this car is set at $290k, reflecting a balanced view of its market potential.
For the 3.6 Turbos, the situation is different. The 1993 model is estimated by the venue at a value significantly higher than both the Comps and AI estimates. This discrepancy is largely due to factors that the Comps and AI models may not fully account for, such as the ultra-rare Japanese specification and unique color combination, coupled with its low mileage. These elements drive the car’s value beyond what the AI and Comps predict. Our human estimate, considering these unique aspects, predicts the 1993 car will sell for $520k.
The 1994 Turbo is also valued higher than the Comps range, though it intersects with the upper part of the AI range. This indicates that the AI model is recognizing factors similar to those considered by the venue’s estimators. Our human estimate for the 1994 model is set at $410k, aligning with this more nuanced understanding of the car’s value.
All three cars sold, and all three were below the venue’s estimates and very close to or above our human’s estimates.
For the Carrera RS, all four estimates were close to the final price, signaling that there is a very strong, clustered market around these cars. The venue estimates had the most difference, but error-wise this sale was one were the estimates did the best.
On the 1993 car, and as we well noted, the ultra-rare Japanese spec and unique color combo + low mileage drove the price significantly above the factors that the data-driven estimates can “see”. The result was the final price well above the Comps, AI and our human estimate, and a bit below the venue’s estimate, which was closer than all of our three estimates.
The 1994 car, was not specced as “special” as the ‘93 car, and thus had a result much closer to the data-driven estimates and our own human prediction, signaling that the greatest weakness is indeed in valuation the “super” special subjective value drivers for these cars.
993 Carrera
The 993’s analyzed are all no-reserve, ex-White Collection cars with intersecting ranges between the Comps, AI, and venue estimates. The Targa has the closest estimates among the three, and given it last sold for $168,000 in December last year, it makes sense. However, our human expert predicts that it will sell for a bit less now given recent market activity and climate, estimating its value at $140k.
The 1996 and 1997 Carrera 4S cars are nearly identical, with the primary difference being that the 1997 car features a factory aero-kit. And, unlike the Targa, these two cars were not part of the auction from the White Collection in December last year. Both are being offered without reserve and are expected to sell close to the venue estimates, which are well above the Comps range. These two cars are among the few where the venue estimates sit between the AI and Comps estimates, with the AI estimates being the highest.
Given the exceptional quality of these cars, and the rising interest for 993 4S’s, our human estimates for these models are $235k for the 1996 car and $245k for the 1997 car, reflecting a reasonable position amongst the three different estimates.
All three cars sold as well, all of them out-performing our human and comps estimates, and in the case of both Carrera 4 very close to the AI mid-points, but well above the venue’s estimate. This proves special 993’s are hot in the market and that there is strong demand for them, but best of all is that our AI model was able to see that.
For the Targa, the result was best aligned with the venue’s estimate, selling for only $5k of what it sold for 8 months ago. It is a very special car amongst its market, so we understand the result, but underestimated it slightly.
993 Turbo
The value of 993 Turbos has been steadily increasing over time, which likely explains why the 1996 model has a higher Comps range compared to the Venue or AI estimates. This trend also justifies our human estimate being placed close to the middle of the venue estimate, at $215k.
On the other hand, the XLC models show consistency between their Comps and AI estimates, but the lower mileage XLC is valued almost three times higher than those estimates. XLC Turbos, with their GT2 performance in a 4WD package similar to the Turbo S, are highly sought after, especially since only 200 were made. The exceptionally low mileage of one example makes it particularly valuable. However, given the significant disparity in the valuations of the two XLC cars, our human estimates reflect a more nuanced approach. The low-mileage example is valued below a Turbo S, at $540k, while the higher-mileage XLC is estimated to go for $250k.
Like we saw with the 964’s and 993 Carrera cars, these more modern gen cars had a better result in the Monterey auctions than the older cars, and we continued to see the same trend for newer cars as well: all 997 and 991 cars considered in this analysis also sold.
The ‘96 Turbo (non XLC package) sold on the intersection of the three estimates, on the higher end of the venue and AI estimates, and almost right in the middle of the comps estimate, which aligns with our initial theory that the market for these is still very strong (even stronger than we predicted).
We did much better with the high-mileage example than with the low-mileage example, which once again confirms that the toughest challenge for these estimates is gauging how much strong value drivers like “low-mileage” affect a vehicle’s final price. In this case, the result was right in the middle of the gap of the Comps/AI estimate and the venue’s estimate. For the higher mileage one, the result was almost exact to our human estimate and just slightly above the AI and Comps estimate, but also very slightly below the venue’s prediction.
997
The 997 cars in Monterey exhibit varied relationships between the different estimates. The 2008 GT2, for instance, shows a much wider Comps range compared to the AI and venue estimates, although it intersects with both. The AI estimate sits at the lower end, while the venue estimate is at the higher end of the Comps range, which is typical for these cars. Given this, our human estimate aligns more with the higher end of the Comps range, setting it at $330k.
For the more common 997 Turbo, the venue and AI estimates are nearly identical, while the Comps estimate is significantly lower. This discrepancy is largely due to the car’s extremely low mileage—under 200 miles. The Comps algorithm, which relies on comparable vehicles, groups this car with others that have ultra-low mileage but doesn’t fully account for the fact that this particular car has only 100 miles. The AI and venue estimates, however, do consider this detail. As a result, our human estimate for this 997 Turbo is set at $130k, which could potentially set a record for 997.1 Turbos.
For the GT2 RS, the venue’s estimate is not far from the high-end of the comps and AI range, yet its ultra low mileage (under 400), should place it at least at the top of the comps range, at $740k.
On the 997’s, the venue estimate out performed both the Comps, AI and human estimate on the GT2, again a show of the strength of the market on these “newer” cars.
The 997 Turbo saw a very interesting result, with the AI estimate being the best predictor of value for this ultra-low mileage example of a 997, which unfortunately the comps couldn’t account for. This car shows-off one of the biggest strengths in the AI predictions: it can accurately measure how much the ultra-low mileage value driver can affect price above the market.
For the GT2 RS, our human estimate had the best result, although the price also fell within the comps range, but above the AI estimate and well above the venue’s estimate. In this case, the AI didn’t measure the low-mileage of this car as well, which points to one of its weaknesses: the ultra-rare low-volume markets.
991
The AI estimate was the tightest for both the Targa 4 GTS and the Turbo S models, and it is also set below the Venue and Comps estimates. In the case of the Targa, the three estimate ranges are very similar, and our human estimate aligns with the intersection of these three, setting it at $145k.
For the Turbo S, the venue and Comps ranges are noticeably higher than the AI estimate. Given the similarities between these estimates, our human estimate is positioned right in the middle of the Comps range, at $280k.
The GT3 RS, another example from the White Collection with delivery mileage, also shows AI and venue estimates that are significantly higher than the Comps estimate. This discrepancy is likely due to the car’s ultra-low mileage. With a recent sale of a 500-mile GT3 RS for $256k, our human estimate takes a more cautious approach, predicting the sale of this car will close at $330k, acknowledging the high value but factoring in recent market behavior.
For the 991 cars, these were all very different cars, and the results vs our estimates show that.
The Targa 4GTS was a special example, and sold very close to the human estimate and the comps mid point. The AI predicted a lower range, and was more closely aligned with the venue’s estimate. We think that perhaps the special color drove its value up.
Something similar happened to the Turbo S, with the comps, human and venue estimates being much closer to the result, and the AI grossly underestimating the final price, which we will look into.
And finally, the GT3 RS from the white collection also shows that the AI had the ability to accurately measure the ultra-low mileage on this example, which again the comps failed to do.
Why we picked these Porsche 911s in Monterey
This year a total of 86 Porsche 911s will be crossing the block across five major auction events. This lineup is diverse not only in terms of models but also in terms of condition and modifications. We chose to focus on three cars for each generation that is represented on the Peninsula that have comparable features.
Both our Comps and AI methods are 100% data driven based on comparable sales – producing accurate estimates for all 87 cars is difficult due to the number of outliers in terms of quality and rarity of some listings. There are also several cars that are unique in nature, such as one-off custom 911s, or race cars whose value depends on unique factors that just don’t have enough (or any) comparable sales. So we picked cars that have comparable sales to provide the best test cases. Our little baby AI tool is just getting started. We need to give it some time to grow and learn – this is its first big test!
The final verdict: Can AI predict sale prices?
AI performed the best out of our tested methods of estimating the final sale price in Monterey:
- 50% of prices landed within the estimated AI range
- 38% of prices landed within the estimated Comps range
- 28% of prices landed within the Venue’s estimated range
However, we aren't ready to call this a success quite yet. These results underscore the difficulty of predicting prices at live auction events like Monterey.
For these 32 listings, only 50% of the final prices fell within any of the three estimated ranges.
- For the 20 listings where Comps missed the range, 17 were under-estimated, and only 3 were over.
- For the 16 listings where AI missed the range, 15 were under-estimated, and 1 was over.
- For the 22 listings where the venue missed the range, 19 were over-estimated.
This indicates a tendency for data-driven algorithms to underestimate prices, possibly because the data points often reflect cars that aren’t as exceptional as those presented at Monterey... or because the algorithm is yet to take into account the excitement of live auction events. On the other hand, Venue estimates are likely biased in favor of the seller's interests, leading to overestimations. Our Human estimates, which integrated both Comps and market data with a qualitative evaluation of each car, displayed a more balanced outcome, with 20 prices exceeding the estimate and 11 falling below.
Comps and AI estimates also provided a wider value range for each of the listings, making their average error larger than the Human and Venue estimates. As we train our systems with more data and qualify even more value drivers, we will be able to narrow the value ranges.
We are looking forward to continually improving our algorithms and building tools to help both buyers and sellers to know what their cars are worth, and when/where the best time is to sell or buy.
Stay tuned for more explorations like this from your friendly data geeks at CLASSIC.COM!
