All news

Kensho's sale makes data incumbents the exit for AI-first start-ups

A small cobalt cube sliding into the one empty slot of a wall of white and graphite blocks

S&P Global agreed to pay about $550 million for Kensho, a year after its $50 million round. Data owners now buy machine learning teams rather than build them.

About $550 million in cash and stock is what S&P Global has agreed to pay for Kensho, a five-year-old machine learning company from Cambridge, Massachusetts, that sells analytics to large banks and to the national security community. The deal, announced on 6 March, values Kensho above the price of its last funding round, a $50 million raise in February 2017.

The price shows where exits in applied AI are forming. The buyers are not only technology giants hiring researchers but owners of proprietary data and customer relationships, who would rather buy a working machine learning team than spend years building one. For an AI-first company selling into a single industry, the incumbent that holds that industry's data is now the most likely acquirer, and the one with the most reason to pay up.

S&P Global is acting on that logic at speed. Two weeks before Kensho, it agreed to buy Panjiva, a company that applies machine learning to a billion records of global trade transactions, on undisclosed terms. Two such deals in a month from one data company make a strategy rather than an opportunistic purchase. The scale is meaningful for the buyer as well: S&P Global's revenue was $6.06 billion last year, so Kensho alone costs about 9% of a year's sales.

The accounting shows what is being bought. S&P Global cut its 2018 earnings guidance under standard accounting rules by about 20 cents a share while leaving its adjusted figure unchanged, because the adjusted number excludes amortisation, replacement equity awards and retention plans. Much of the cost goes to keeping people in place. Kensho will run independently in Cambridge under its own brand, with its founder reporting to the buyer's chief financial officer. The team, not the software alone, carries the price. Founders building for one industry should treat the data owner as their first customer and their likeliest buyer, and keep the team that makes the product work intact, since that is what a buyer pays for.

For Fanaura, a buyer list like this one matters at the moment of investment, not at the exit. It tells us which AI-first companies have a strategic acquirer waiting for them, a test we apply when we assess founders and their markets.

Scarcity is the weak point in this reading. A premium paid for a team reflects a shortage of skills, and shortages end. As machine learning skills spread and tools become cheaper, incumbents will find it easier to build in-house, and a company sold mainly for its people was not yet worth more on its own. The signal arrives over the next year: whether Kensho's tools show up in the buyer's products and lift its market intelligence revenue, and whether other exchange and financial information groups follow with deals of their own. That is a question for capital that can wait years for an answer.

In applied AI, whoever owns the data sets the exit price, and this month the price was $550 million.

For founders

Building an AI-first company on the data of one industry, with its incumbents as customers? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.