You don’t get there by starting a significant company-altering digital transformation. You get there by knocking out some early, lower-risk projects. This approach lets you build know-how and competence, allows thoughtful assessment of your organization’s internal capabilities, and set the stage for identification of future projects (with, likely, more substantial paybacks).
In the beginning, you don’t know what you don’t know. You need to identify the high impact and practical projects but, by definition, have no company experience in doing so. You also have to identify the data science skills you need to execute the plan (and are those the same skills you need in the long run). It’s hard to know what your team skills need to be and to start building your team before you know what the high impact projects are going to be.
Hiring an intern, a recent Ph.D., or a retrained software engineer to be the lone company data scientist for that initial project can be a high risk. There are a million reasons why an inexperienced lone data scientist may fail. They can fall into a data science pitfall (over-training, for instance), tried to solve the problem with their familiar technique rather than some more appropriate technique, or lack the experience to integrate into a multi-disciplinary team. If you fail this way, where does it leave you? Abandon AI? Try again? In any case, the most significant risk is that time will be lost vs. the competition.
One fairly common solution is to use an outside (consulting) team to get the projects moving quickly. I propose that the outside AI team’s primary mission is to help to build the appropriate inside (full-time employee) AI team. Here are a couple of thoughts on how to make that happen:
You should, initially, access a more solid, experienced team that has access to a broad range of data science skills and has expertise in multi-disciplinary team integration (basically product development). The objective is to give the first project the highest chance of succeeding.
That outside team should actively help build the inside team that YOU need. Not every consultant can or will do this. Still, it is essential— data science is a core competency that benefits from a continual focus and that will require the company to have some inside resources. I don’t believe you can use consultants for 100% of your data science needs. The right decision as to what that an inside team looks like will become more apparent after the first project. The right choice could be to hire a larger inside team if the business case is healthy or maybe a smaller unit to make continual improvements would suffice, using an outside team to supplement new initiatives. Unless you are building a deep bench of AI capabilities, you probably don’t require a veteran Chief Data Officer, which are hard to find and the most expensive part of an AI strategy, until you hit a specific size of AI effort.
Maintain an outside team/partner for stability. The demand for Data Scientists right now is resulting in a lot of job-hopping. Good data scientists change jobs because they are motivated to learn more and solve new problems; smaller companies can’t always provide the continual learning and challenge that top data scientists require. We’ve seen multiple situations where a company loses its entire senior data science leadership in a short period (typically at the 12 to 18-month mark). Outside data science resources can mitigate that risk and retain institutional knowledge — allowing you to keep a smaller, more focused, inside the team.
AI will be a core competency for many (probably most) companies across most industries. It won’t be accessed episodically like legal help (call the law firm when you need them) but will require the thoughtful building of resources that are part of the daily execution of the company business. It is expensive and impractical to build a deep bench before that first project because you don’t know what you don’t know at the start. Find outside help not only on doing the first project but on how to design and assemble the inside team.
Note: I have mixed feelings about using the term “AI” for all NLP, Machine Learning, and other advanced data science applications. It seems like Data Science is a more broad term. I like data science because it reminds us that the best solution to every problem is not necessarily, or even likely, to be a complicated Deep Learning undertaking. Artificial General Intelligence (what many people think about when they hear about AI) is not a solved problem and is possibly a long way off. However, in the spirit of talking the same language, let’s consider AI == Data Science for the time being.