If you are launching your first AI project, you should be part of it. Never give key responsibilities out of your hand and since AI will be one of your key pillars in your company, you should be part of it: from the start. You will have to be the “decision maker”, the person who understands how your company is creating value as a business, for employees and for customers. Don’t just allocate 30 minutes a week to this project. Take hours, take walks, take breaks from your daily routine and learn about AI. As AI deals with data, you need to know how to make decisions based on data. Data-driven thinking so to speak. It is fully fine to get a discrete AI leadership coach in the back who teaches you to become that data-driven decision maker, but in the end you need to in this team. As a tip, get a master of psychology graduate on your side as a personal assistant. They have year long experience in data-driven decision making and are used to make the unmeasurable measurable.
#7 Project Manager specialized in AI
It is funny that AI projects often have no project manager. Many try to use SCRUM for AI projects, but this often fails. This AI project manager plays, next to you, a super important role in this whole endeavor. For your first AI projects, this role should be bought-in.
The AI project manager knows business, but also data, process management and project management. He or she will always make sure that everyone will stay on track, keep the troops together and will be go-to person for every team member. AI project managers are rare and cost more than normal project managers. Do not take existing project managers from your company and just hope that “they will be able to do it”. Saving money on this role always will come back to you in a negative way.
This sounds like a fancy term, but basically this role can be filled by every team member in your company. The role is to look at existing data and get excited by what it might reveal. Just imagine following scenario: You have an excel sheet with 50 columns (each column showing one data source). People in your company, probably domain experts in their field, will very shortly form an opinion what problem you have in your company (e.g. need of forecasting, prediction) and how you can use this data to solve it. A data scientist might be able to build you a model, but he or she will never understand the bigger picture of what these data mean for you, how they are created and how they relate to each other.
Most time in an AI project is spend on so called data engineering. This role helps to get, pre-process and process the data, before the data gets analyzed or modelled. If you have small data sets, the data engineer might just change the commas to dots to merge European and US data sets, but if you have terabyte of data, this job can get complicated pretty fast. This role is absolutely key in your project. And do not expect any data scientist to be able to do this “additionally”. On the one hand they might feel offended, but on the other hand they might not know what to do. Furthermore just imagine you want to solve real-time optimization problems. Then the data engineer needs to know a lot about databases, data base management, search and sort strategies and everything surrounding this topic.
# Applied Machine Learning Engineer
Cassie Kozyrkov, one of the leading AI voices in Google, suggest to put this role on the list, too. An applied machine learning engineer does not create algorithms or knows how they work in detail. This person knows all the software suites surrounding AI models and knows how to apply them and make them work in production. They will write a lot of code and try to make the whole system work. They will know their whole Machine Learning Pipeline and will be later the person who helps you bring the amazing models to life. According to Cassie, look for people with a high tolerance for failure, as his/ her job is mostly a black box. Others call this role also MLOps.
We all hope that data scientists are statistian, applied machine learning engineer and also know how to build deep learning models, but in reality this is seldom the case. Furthermore the qualifications of the person you are looking for is dependent on the problem you want to solve. People specialize on certain areas, e.g. natural language processing, self-driving cars, strategy optimization in banking. So before you hire a data scientist, develop with your team a few ideas what problem you actually want to solve.
There are of course many other roles, but these roles above are the key roles you need in the beginning. One further note: Do not waste your time on getting data science researchers for your first project. Your first project should use well documented and well-studied models. Do not start too fancy for your first project. Remember you want to reduce complexity to reduce risk. If over time you discover which kind of data you have and how well the existing out-of-the box algorithms work for you, then you can start looking for highly specialized researchers.