“Where and when will artificial intelligence make a difference, good or bad, to the companies I own?” In the end, that is the only question about this nascent technology that matters to most portfolio managers, investment advisers, and stock analysts. Will AI be “Déjà vu all over again?” Over the last decade we’ve seen Amazon remake retailing, Apple redefine the mobile phone, and Facebook/Google dominate online advertising.
Those case studies prime the investment community to expect that AI will generate the same sorts of disruption, but on a much grander scale. It was hard enough to see the Amazon juggernaut coming a decade ago (when the stock was $71, not $1,500 as it is now). How can investors assess the disruptive potential of AI when it touches basically everything?
McKinsey is out with an in-depth study today that offers a roadmap to consider this question. Researchers/consultants there looked at 400 use cases across 19 industries and 9 business functions. They define AI as a neural network (organized in a similar fashion to the human brain) with many layers of interconnections.
We’ve included a link to the report at the end of this note, but here are our three key takeaways for public equity investors:
#1. Adopting robust AI solutions is hugely complex and comes with significant front-end costs. Yes, consultants always make things seem more difficult than they are. If every business challenge had an easy answer, they would be unemployed. Still, in this case McKinsey makes a strong case and outlines the following:
- AI can’t do its thing (i.e. “learn) without organized and labeled data.
- The data requirements to bring AI up to human level performance can be large. McKinsey cites the number of 10 million examples in its report.
- Unless a company already uses very advanced analytical tools, they are unlikely to have this large an already-labeled sample size available to train an AI system.
- The technology to correctly label historical data is still evolving, so for now the creation of fully labeled datasets falls on human programmers and business people.
- The journey doesn’t end with a good historical dataset. AI systems need fresh data on a daily (preferable) or monthly basis to adapt to new trends. That means every front-end system a business uses needs to label incoming data as well.
#2. The benefits are incremental, rather than dramatic. McKinsey is careful to under promise the benefits of AI, and for good reason. As companies adopt this technology, the benefits will tend to flow to consumers rather than stay within the four walls of the business due to competitive pressures.
Here are the anticipated financial benefits cited by the consultants, and a few hypothetical industry examples:
- “Per industry, we estimate that AI’s potential value amounts to between one and nine percent of 2016 revenues”. Hardly life changing, in other words. But enough to get the attention of most Fortune 500 CEOs, who would give their eye teeth for a 100 basis point improvement in operating margin.
- Potential impact on Retail: 1-2% incremental sales growth by adopting AI-powered personalized customer promotions.
- Potential impact on Consumer Goods: 5% reduction in inventory levels and 2-3% increases in sales (presumably from better in-stock performance).
- Potential impact on retail banking: no numbers given, but McKinsey cites fraud protection and risk assessment as two areas where AI could help.
#3. There is a wide spread in terms of which industries stand to benefit the most from developing AI solutions.
In terms of where you should spend your time quizzing managements and Wall Street analysts about the impact of AI on particular industries, here is McKinsey’s top 5. The percentage changes cited are the notional improvement AI might offer over existing business processes:
- Travel: 128% potential improvement using AI
- Transport and logistics: 89%
- Retail: 87%
- Auto and assembly: 85%
- High Tech: 85%
Two final thoughts:
1. McKinsey is a global business thought leader, so for them to publish a report that emphasizes the difficulties and expense of adopting AI is very notable. We still can’t ignore progress on this front; AI will get better. But don’t expect it to be the magic bullet for any non-tech public company any time soon.
2. The only businesses with large labeled datasets are tech companies. Some non-trivial part of their lofty valuations comes from this competitive advantage. And, of course, they are all at the forefront of developing AI applications to harness that information. As a result they will certainly beat non-Tech companies to the finish line. Perhaps those valuation premiums are actually too small.