Welcome to the AI acquisitions club: a framework on how AI acquisitions are priced

AI startups have grabbed the recent headlines on exit activity. These headlines are a confluence of multiple trends, including the lack of 2016 exits, the increasing acceptance of AI as a core technology, and the largest acqui-hire in history through DeepMind. But as common as AI acquisitions have been, the understanding of the process have been and still pretty opaque.

Part of the opacity is the differences per acquisition deal. Some deals will be executed quickly due to familiarity between the acquirer and acquiree (see Gnips and Twitter), and some a much longer ordeal. Some will be executed very smoothly with both sides happy, and some disappear into the sunset because of bad negotiation, timing, or players. Some force relocation of a startup into the parent company’s headquarters, and some like Pie in Singapore is the bedrock for the parent company in a new market.

However, like in most things, there are patterns to be discerned. These patterns serve to provide a framework for future acquisitions to happen and to smooth the way for future processes. This framework serves to guide and should improve over time.

There are four distinct elements of how an acquisition is priced. These elements are interrelated and the ultimate price may be a factor of one, or all four, of these elements. For clarity, acquisitions of public companies are not considered, since they are already priced by market forces. Furthermore, these four elements are in the order of importance, from most to least.

The first element is the acquirer’s strategic intent. It is important to assess why a possible acquirer is interested, based on the factors of product, team, or the opportunity cost of not acquiring. In some cases, the motive behind an acquisition is for an acquirer to accelerate development in a product, with human resources that are not currently available within the acquirer’s company. Thus, it is cheaper to acquire a startup that has been developing in the space than to organically build a team from nothing or divert resources from other products.

The second element of an acquisition is product, or the core tech and algorithms, in the case of AI startups. Oftentimes, acquirers have existing data, but have no idea how to utilize the data they have; on the other hand, AI startups have product and algorithms that effectively utilize data and can generate significant accretive value. The leverage is understanding what data is needed to generate value, find acquirers that have that data, and target acquirers that will benefit significantly from a product integration.

The third element of how an acquisition is priced is the team. The team can be comprised of the following components – leadership, sales, engineering, and/or operations. Usually engineering and leadership components command the higher premiums, but this is not always the case. The leverage is that the more rare or specialised the skillset is to the acquirer (for example, Machine Learning talent), the higher the premium will be. This was the main pricing element for some of the early AI acquisitions, like DeepMind and Wit.ai.

The fourth element of an acquisition deals with the users/customers. Some acquirers are willing to pay a certain price per user/customer, since organic customer acquisition cost may be even more expensive and take skillsets and time that the acquirer does not have. The leverage is that the higher customer acquisition cost is, the higher a premium an acquirer will pay. This element is a common method used for ecommerce startups where the acquirer does not have an advantage in a certain target market.

Finally, startups should look upon this framework as a guide to start thinking about how to mediate what is generally an obscure process. But obtaining guidance from someone who’s gone through the process is also suggested, both from an experience perspective and as a way to bounce off questions that are specific to each deal.

Thank you Wendy Chen, Rodolfo Rosini, Jonathan Savoir, and Alexandre Winter for comments and edits