15th June 2017
By Dr Janet Bastiman
Chief Science Officer
It’s hard to discuss anything in technology without someone asking about Artificial Intelligence, Data Science or Big Data.
Established companies are desperate to get those in-house skills and startups are riding the bandwagon and hoping for that killer idea that will see them bought out after a few rounds of excessive funding. Even the investors are part of the hype.
All of this has created a supply and demand issue that is not easy to fill. A few years ago, graduates were leaving university with PhDs in machine learning and were expecting salaries that were on par with those of a mid-level developer. Right now I see bright people with unrealistic expectations who are poorly equipped for the world of business. This is just a problem waiting to happen. So how do you deal with it?
The first thing to think about is whether you need your own in-house team at all. In addition to the likes of IBM, Google, Amazon and Microsoft, there are a whole host of service companies providing AI on demand in specific areas. A partnership deal with one of these providers could get you a chatbot, visual tagging or analytics implemented faster and a fraction of the cost of an in-house team. This is possible because these companies use pre-built generic AI. For many companies, this could be enough – get up and running quickly and give the customer a great experience. So why would you even consider an in-house team?
Firstly there is the data. AI lives and dies by the data that you use to train it. Regulation around data ownership and privacy is being tightened in the UK under the EU GDPR legislation, and there are worldwide variances on what you can and cannot do. Handing this data off to a third party may not be appropriate for your business. Furthermore, some service companies I’ve seen include waivers that any data you upload to their service gives them the implicit right to use. If you don’t think your customers would be happy with that sharing of their data, then you need another solution.
Secondly and most critically is the problem you want to solve. If you have an idea and the services offered either don’t solve your problem or don’t solve it well enough then you need to build something in-house.
So you’ve considered the above and agree that you need a team. What do you do?
Assuming all of the senior team are in agreement, the most technical person in the company needs to educate themselves on AI. Do the TensorFlow introduction demos, go to some of the many conferences or meetups that occur and ask questions. The community is pretty open and helpful.
Find a good recruiter who really understands the community
This is essential. You need someone to help sift through the mountains of CVs to find the ones that are a good fit for the problem and your company. Cold emails won’t do this. Find someone who cares enough to get to know you and what you want rather than just forcing expensive candidates with “all” the skills onto you. If you can’t trust the judgement of the recruiter you use, then you should stop using them.
Hire someone to run your AI effort
The person heading up your AI team is critical. Depending on your company this may be Chief Science or Data Officer, Head of AI, Head of Data or a whole host of new titles that didn’t exist a few years ago. Pick a title that fits with your organisation’s existing structure. This person is critical – they will be the bridge between the AI team and the rest of the business, including the development team. They need to be a strong communicator, have a good head for business and a deep understanding of AI. More extensive technical skills are beneficial, but you need the business acumen, however, they can demonstrate that, to understand commercial pressure. Most importantly you have to trust them. Just like the trust you need in your recruiting partner, you need to devolve the strategy and implementation of your AI to this person.
Hire the team and invest in infrastructure
Work with your new AI head to define the size of team and infrastructure needed. As a rule of thumb you’ll want to make sure that each member of the team has a decent machine to use on a day to day basis (think high GPUs) as well as server to leave models running long term. Depending on the problem, you might need several servers per person or very large (expensive) machines, so plan on whether you’ll be buying these or using cloud services. The hiring process can be difficult. Try and build a team that’s a balance of academics (with business experience if you’re lucky!) and engineers with machine learning experience. This team needs to work well together, be open to learning from each other and able to challenge each other. You’re looking to create something cutting edge so the environment needs to be fast paced and stretch the team or your competitors will get there first. A great recruitment partner can help find these gems.
Overall, just like any development project, the team dynamic is critical. Get the team skills and characters correct, and you’ll have the advantage. Then it’s all down to the idea!
Get our latest articles and insight straight to your inbox