Should you build or buy AI? – VB Summit 2018 Boardroom Session Summary
At VentureBeat’s recent VB Summit event, I headed a session on whether enterprises should build or buy AI. Between comments from the panelists and a group of about 20 business leaders, a good decision tree emerged for how to answer this question. Given how important the question is, I wanted to share that decision tree more widely.
Here it is:
As you can see, at the top of the tree is the question “Do you even need AI?” I believe AI can positively impact any and all businesses, so the correct answer should always be yes.
The next question to ask is if AI is in your company’s DNA. If your company is AI at its very core and not an application of AI, then you should definitely build. In this scenario, your only differentiator would be the AI, and if you are buying from somewhere else (where others can also buy from), then by definition that cannot be a differentiator.
If you are a company that is using, or can benefit from AI, then we proceed by asking the question: “Are the current solutions enough to solve your problems?” This is not an easy question to answer. The difficulty everyone talks about in hiring great AI people comes down to two things. The first is the lack of talent, and the second, is the lack of expertise at the hiring companies to properly evaluate the candidates. This second reason also applies to evaluating software, thus creating the difficulty. The advantage in evaluating software vs. people is that there are more quantitative and discrete means of evaluation. Nonetheless, the topic of evaluating AI solutions is not a trivial one. You can get some help on that from this article covering the overall evaluation of AI solutions.
In short, when considering buying an AI solution, you need to have ideas on how AI can truly impact your business, and the main question you need to ask is whether the solution can get you to those goals. The solution could potentially reduce the time for AI projects, or reduce the requirements for the team, thus enabling teams to be built more easily and quickly. Or perhaps it improves the overall quality of your AI projects.
If the answer to this question is “no,” then you can build it yourself. But if your answer is “yes,” you can move forward and buy. The buy option is not, however, done in a vacuum. There has to be a team or group of people that will use the product, and another group that must get the results of such teams and place them in an actionable path within the organization so that these AI solutions can have a real impact.
We had an entire conversation at VB Summit about the tradeoffs between having a centralized AI team that serves all the different groups of an organization versus having AI-enabled people within every team. The latter was the consensus, with some caveats.
The idea of a centralized team has some obvious desirable traits to enterprises such as: more control, organization, and an easy marketing/visibility push towards innovation. The reality is that you end up with a group that under-serves the multitude of teams needing it, and the centralized team will inevitably be somewhat disconnected from the problems that need to be solved. This creates the perfect environment for delivering projects to some groups that never get used and delivering no projects at all to other groups, and as a result individual teams end up getting their own AI hires.
Having AI integrated into every team is the goal but can be challenging. The first challenge is finding people to hire into the separate teams within your organization; this can be alleviated by software that reduces the skills required, significantly enlarging the candidate pool. The second problem is having a consistency between data scientists across different teams. The conversation around this led to two solutions: One was to find someone to help coordinate all AI people across the organization, increase their communication, and standardize process and practices. The other was to use standardized software, since if all your AI point people are using the same software, their work will be standardized. This solution is a more feasible and efficient way of ensuring every group within your organization can get the transformative power AI can bring.
By the end of our discussion, it was clear the buy decision was the prevalent one, given that the percentage of companies that would land in the “I am AI” bucket is small, and the number of companies in the “Software cannot solve my problem” bucket is shrinking rapidly.
As the field progresses, though, organizations won’t need discussions or decision trees such as this. Choosing AI will be like choosing electricity. You certainly don’t need to build your own nuclear power plant and hire nuclear physicists if you want to manufacture TVs. You just get electricity and you know what to do with it. This is the future of AI. We are not there yet, but it might be closer than you think.