AI Can Make the Relative-Valuation Process Less Subjective

Relative valuation—using simple metrics to compare a firm’s value to its peers—is a cornerstone of financial decision-making. If a company earns $2 billion in profit, and if similar firms trade at 15 times earnings, then that company might be valued at $30 billion. This approach offers an intuitive starting point for understanding valuation ranges, standing alongside the more detailed discounted cash-flow analysis methodology.

Yet current approaches to relative valuation are far from perfect. The process of selecting comparable firms often involves subjective judgment—and even determining what industry a company belongs to can be tricky. The Global Industry Classification Standard (GICS), for example, groups Alphabet (Google) under code 5020 or “Media and Entertainment,” whereas the North American Industry Classification System (NAICS) allocates it to code 518210, which is “Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services.” So Google could equally be described as a media company or a computing-infrastructure provider, and in fact operates in both industries. Such differences in classification can lead to inconsistencies in peer selection and, ultimately, valuation estimates. Moreover, it’s simplistic to assume that all firms within an industry share similar margins, growth rates, or risk profiles.

A New, AI-Driven Approach

Two of us (Paul and Helen) have recently published research forthcoming in the Journal of Accounting Research, in which we devise a methodology that incorporates artificial intelligence in the relative-valuation process. Instead of relying solely on subjective judgments, our methodology uses algorithms to analyze a wide range of financial inputs and select comparable firms. In plain terms, it uses AI to review historical data—such as revenues, earnings, and debt levels—to detect patterns and relationships related to historical valuations that traditional methods might miss.

At one level, this is not a new approach: As Thomas H. Davenport and Peter High have recently pointed out on HBR.org, analytical AI has become an essential tool for many organizations. Our work relies in particular on Gradient Boosting Machines (GBMs), which, because they can uncover complex patterns and nonlinear relationships in large datasets, are one of the key AI techniques used in financial prediction. What’s new about our methodology is that it uses GBMs to break down individual valuation estimates into a weighted average of peer-firm multiples from the training data. We refer to the weights used to construct this weighted average as “peer weights” since they provide a measure of the contribution of each peer firm in the training data to the GBM valuation estimate. This added layer of transparency allows finance professionals to see exactly how much each comparable firm contributes to the final valuation, making the results both more interpretable and actionable.

Peer Pressure

Suppose we were interested in valuing Mastercard. A traditional relative valuation might compare the company with other payment processors, including both emerging fintech companies and well-established competitors. However, these firms have vastly different risk profiles and growth dynamics. A broader industry view would group Mastercard with financial-services firms. Yet Mastercard, unlike banks, does not bear credit risk, making its business fundamentally different from consumer lenders such as Citigroup or Wells Fargo.

To apply our AI-driven peer-selection method, we first train a machine learning model to predict valuations using historical data for large U.S. firms. We then use our novel algorithm to extract peer-weights from the model for each model prediction. Since these weights come from a model designed specifically for valuation prediction, they capture the most relevant firm similarities for valuation purposes.

When we apply our method to Mastercard, we find that the top-five most comparable firms are Apple, S&P Global, Eli Lilly and Company, Lockheed Martin, and United Parcel Service. This is followed by a longer list mostly consisting of technology firms. Interestingly, none of those firms is normally considered to be a financial-services firm. It seems that the data is trying to tell us something: When it comes to relative valuation, Mastercard may have more in common with technology and specialized service firms than with traditional financial institutions.

Our method also identifies peer firms that align with the most informative predictive variables for valuation. For Mastercard, the most significant variable, return on equity, shows a tight clustering among peer firms, as does quarterly return on assets. In contrast, gross profit and cash-flow variance are more dispersed. Overall, our approach appears to select peer firms that are quite similar across at least some key predictive variables.

The power of this methodology becomes apparent if we realize that every firm has a peer-weight that relates it to every other firm in the data. These inter-firm relationships can be represented visually as a network, as in the diagram below. The thickness of the connecting lines indicates the relative strength of the peer-weight association among firms. The layout of the graph attempts to bring firms with stronger peer-weight associations closer together. Mastercard is in the top-left corner; its connections are rendered in shades of red for ease of identification. It seems to be positioned in an area of the network with a strong technology focus, with Nvidia, Apple, Microsoft, Texas Instruments, and Tesla all close at hand.

See more HBR charts in Data & Visuals

In comparison, the upper-right quadrant contains several financial-services firms, including Citigroup, Wells Fargo, Goldman Sachs, Berkshire Hathaway, Bank of America, and Morgan Stanley.

Healthcare firms appear to be located toward the bottom of the network, with Merck & Co., Johnson & Johnson, Abbott Laboratories, Procter & Gamble, and Stryker Corporation all located close together.

The diagram above paints a very different picture from the traditional static hierarchies that define standard industry classifications. For one, the results are specifically geared toward valuation, as the results are extracted from an AI model trained to predict valuations. Had we trained the AI model to predict future sales growth or earnings quality instead, we would have ended up with different networks altogether. In short, the network in the diagram is one that is specifically geared toward the objective of valuing firms and is therefore particularly relevant for peer selection.

Algorithms may be free from emotions, but they are not infallible—nor are the humans who apply them. This means that analyses like the one presented here should be viewed as a starting point for discussion rather than a definitive conclusion. As the valuation expert Aswath Damodaran has pointed out, “soft data”—qualitative insights into management vision and market dynamics—often provide crucial context that algorithms alone cannot fully capture. To truly harness the power of new technologies, AI’s computational strength must be balanced with the creativity, flexibility, and, indeed, the skepticism of experienced human analysts.

Next Steps

This AI-driven method challenges traditional benchmarking and valuation strategies. If Mastercard is more comparable to Apple than Citigroup, should finance executives adjust how they benchmark their performance? Should investors reconsider how they assess cross-industry valuations?

For professionals looking to integrate AI into valuation, we suggest the following steps:

Select key inputs.

Use both quantitative financial metrics and qualitative factors (such as brand strength or management quality).

Train and test models.

Employ historical data to train machine learning models and predict valuation ratios accurately.

Generate transparent outputs.

Using the trained model, extract peer-weights that explain how the predicted valuation of each firm relates to those of other firms.

Review and refine.

Cross-check the outputs with traditional financial models and expert intuition. Adjust the inputs and peer selection as needed to improve accuracy and relevance for the final application.

. . .

By integrating AI into the relative-valuation process, organizations can transform a traditionally subjective art into a more rigorous, transparent, and data-driven science. This new approach not only enhances valuation accuracy but also builds confidence among stakeholders by clearly showing the rationale behind each valuation decision. As companies navigate uncertain economic landscapes, combining AI’s computational power with human insight offers a path to smarter, more-informed decision-making.