Value Investing in the AI Age: How Traditional Methods Still Apply

Ask around investment circles right now and you'll hear a familiar line: stock picking the old way is finished, and AI-driven momentum is the only game left. Growth has beaten value for more than a decade. A handful of mega-cap tech names carry most of the index returns. Machines trade in microseconds. So why would anyone still read Benjamin Graham in 2026?

I get the skepticism, but the answer is pretty simple. Value investing was never a strategy that pays off every calendar year. It pays off across full market cycles, and the reason it keeps working is arithmetic rather than nostalgia.

Put the performance gap in context

Value has underperformed growth for most of the period since 2010, and it hasn't been close in some stretches. If you parked money in a value fund fifteen years ago, you spent a long time watching growth investors post far better numbers while you waited. That gap is real, and pretending otherwise doesn't help anyone.

Zoom out, though, and the shape changes. Over long horizons, value has tended to come back and outperform, and the reason is cyclical. Growth stocks earn premium valuations during expansions, and those premiums compress hard when the cycle turns. Value stocks are already priced for disappointment, so they have less room to fall and more room to recover when sentiment shifts.

The 2020 to 2025 run was unusual less because growth won and more because of how long and how big the win was. Years of near-zero interest rates made distant cash flows extraordinarily valuable in today's terms. When money is almost free, a dollar of earnings a decade out is worth nearly a dollar now, which flatters any company selling a story about future growth. That math works in reverse once rates normalize, and a lot of investors are only now sitting with what that means.

What value investing actually means

People hear "value" and think "cheap stocks." That's a mistake. Value investing means buying a business for less than it's worth, and that distinction does real work. A stock at 5x earnings can be expensive if the business is quietly dying, and a stock at 25x earnings can be cheap if the company compounds at a high rate behind a real competitive moat.

Graham and Dodd laid the framework out in 1934. Estimate what a business is worth based on its assets, its earning power, and its prospects, then buy it at a discount to that number. The margin of safety isn't really a hunt for low P/E ratios. It works as the buffer that protects you when your analysis is wrong or something you didn't see coming shows up, which it always eventually does.

That principle doesn't care what century you're in. It held when companies made steel, and it holds when they train large language models. The question underneath never moves: is this business worth more than what the market is charging for it today?

How AI actually helps a value investor

Here's the part the doom narrative gets backwards. AI tools make value work easier, not harder. The historical bottleneck was always information processing. Reading 10-Ks, pulling apart financial statements, comparing companies across a sector, and spotting the mispricing took real hours, and there were only so many hours.

That timeline compresses now. Language models can read through thousands of SEC filings and surface the ones where the numbers hint at undervaluation. Screening tools can apply Piotroski F-Scores, Altman Z-Scores, and Graham-style filters across the whole market in the time it used to take to open one filing. The grunt work that used to eat a week of an analyst's life is mostly automatable now.

There's a nice paradox here too. Machine-driven markets tend to create more mispricings, not fewer. When algorithms crowd into the same momentum trades, the names they ignore drift further from fair value. The more capital piles into whatever is working this quarter, the wider the opportunity set gets for someone patient enough to look at what everyone else is skipping.

Where the old methods need updating

None of this means value investing is frozen in 1934. A few things genuinely need adjusting for the modern market.

Book value has lost a lot of its power. When Graham was working, most companies held tangible assets that gave you a real floor. Today a large share of corporate value sits in intangibles: software, patents, brands, customer relationships. Accounting rules mostly don't carry those at fair value on the balance sheet, so price-to-book overstates how cheap asset-light companies look and understates the value hiding in intangible-heavy ones. Lean on it blindly and you'll get fooled in both directions.

Earnings quality has gotten murkier. Most large companies now report adjusted, non-GAAP numbers alongside the official GAAP figures, and the gap between the two can be meaningful. You have to normalize for that yourself, which means actually understanding stock-based compensation, restructuring charges, and the many polite ways a company presents its results in the best possible light.

Competitive dynamics move faster than they used to. Graham could buy a cigar-butt stock and wait years for the market to catch on. Now a business model can unravel in a couple of quarters. A screen might flag a legacy media company as cheap on trailing earnings right as a streaming competitor pulls the floor out from under it. Reading the competitive picture alongside the financials isn't optional anymore.

Running Graham and Dodd with modern tools

In practice, the workflow looks like this. Start with a quantitative screen to find companies trading below a rough estimate of intrinsic value, and use more than one valuation method. Combining a discounted cash flow model, relative multiples, and an asset-based or liquidation view gives you triangulation, so no single wrong model sends you off a cliff.

Then layer in quality filters. The Piotroski F-Score is genuinely useful here because it screens for improving financial health rather than raw cheapness. It scores a company from 0 to 9 across profitability, leverage, and operating efficiency, and Joseph Piotroski's 2000 study found that within the pool of cheap, high book-to-market stocks, the strong scorers went on to beat the weak ones. It's a decent way to steer around value traps, meaning stocks that are cheap for a good reason.

Next, run some forensic checks. The Beneish M-Score flags companies that show the statistical fingerprints of earnings manipulation. The Altman Z-Score estimates bankruptcy risk, and a reading in the distress zone, generally below 1.8 for the classic manufacturing model, is a signal to slow down. Running these next to your valuation work keeps you from buying something whose apparent cheapness is really a warning light.

Finally, look at competitive position, and this is where AI earns its keep. Automated reads on patent filings, hiring patterns, customer reviews, and web traffic can tell you whether a company is holding its ground or quietly losing it. A name that screens cheap on the financials but shows a deteriorating competitive picture is usually a trap, not a bargain.

The behavioral edge hasn't gone anywhere

The most durable advantage in value investing is behavioral rather than anything to do with spreadsheets or models. Value investing asks you to buy when others are selling, hold when everyone around you is panicking, and keep conviction through long stretches of looking wrong. That's hard, and most people can't do it.

AI and algorithmic trading haven't fixed that. If anything they've made it worse. Momentum algorithms speed up sell-offs. Social media manufactures echo chambers of panic. Quarterly performance pressure pushes institutions to chase whatever is working this minute. All of that produces bigger, more frequent mispricings for anyone willing to sit in an unpopular position and wait.

The underlying pattern is old and stubborn because it runs on human psychology, not market plumbing. People extrapolate the recent past, over-react to bad news, and consistently underprice the odds of a recovery. Machines that trade faster don't remove that tendency. In a lot of cases they amplify it.

Building a modern value framework

If you're assembling a value process for today's market, here's what I'd want in it.

• A multi-model valuation approach using at least three independent methods. Discounted cash flow for the cash economics, relative multiples for market context, and asset or liquidation value for downside protection. When all three point the same way, the odds you've found a genuine bargain go up.
• Quality filters that screen for improving financial health rather than static cheapness. The Piotroski F-Score, return-on-invested-capital trends, and consistency of free cash flow all belong here.
• Forensic checks that catch accounting problems before they turn into write-downs. The Beneish M-Score and Altman Z-Score are well-established and easy to automate.
• Competitive intelligence on whether the company is defending its position, drawing on hiring trends, patent activity, customer sentiment, and market-share data.
• Behavioral discipline, the commitment to buy into pessimism and trim into euphoria. This is the hardest piece and the one no algorithm can hand you.

Where this leaves you

Value investing isn't dead. It's out of fashion, which historically is close to when it sets up for its better runs. The core idea Graham and Dodd wrote down ninety years ago still holds: buying a dollar of value for fifty cents works over time, whether the company makes cars or artificial intelligence.

What's changed is the toolkit. AI-powered analysis makes it faster and more accurate to find undervalued companies, gauge their financial health, and watch their competitive footing. The investors who pair the old principles with the new tools should do better than either the pure quant crowd or the spreadsheet-only holdouts. If you want a place to start, open a filing on EDGAR, read the footnotes and the proxy statement, and run one company all the way through the process above. Do that a dozen times and the framework stops being theory.