How your small business can begin an AI project

By Paul Nicklin . Posted 29 January 2018
Small business starts AI project

Here’s a framework for how to embark on an AI research project when you’re a small business.

When Boots, the UK based pharmacy chain, pioneered the Advantage Card it set out to understand the customer and sell more. It worked. However, now artificial intelligence (AI) has burst on to the mainstream, the original data science Boots applied will look primitive in comparison to the possibilities we now face.

It’s no wonder then that Google, Facebook and Amazon are investing millions into AI research and development.

But, if you are a small company like ours (just 20 people), and you are completely self-funded and based in Derby, England then it can feel like the AI train will leave without you.

But it needn’t be that way. We have plans to really transform bookkeeping and all the tax credits that go with it. Not sexy admittedly, but necessary and anything we can do to make life easier for our clients translates into loyalty and growth so we know it’s really important to be on that train.

The limitations of being small

Being a small business, we’ve had to think more laterally about researching AI. Firstly, we don’t have huge sums to throw at research and if we did we’d need to be 110% sure that there would be a return on the investment.

Secondly, though we have skilled developers, they aren’t data scientists or specialists in neural networks. That’s not to say they aren’t learning about ML. They are, but it’s at a rate that’s not conducive to getting projects off the ground at the speed we need. After all there’s a day job to perform.

Hiring an AI specialist is an option, but competing with the likes of Amazon for candidates isn’t something we can do. Plus there are plenty of other technical skills we need aside from AI – so we have to prioritise the versatility of our next hire.

Our approach has therefore been to collaborate with Derby University and provide internships to research students. It allows us to test theories with someone at the forefront of research and provides MSc students with a practical environment within which to test their learning.

However, if you are thinking about doing it yourself then it’s not something you can just launch into as a small business.

It’s very important to set some parameters otherwise no one will gain anything. Here are the things we’ve learnt.

How to think big

The first thing to note is that good researchers are in high demand so you need to be very clear about what you want from a research intern, and be prepared to wait for that person to become available.

It may seem obvious but if you don’t get the right person working with you then you won’t get the intellectual capability interns promise.

Secondly, you need a meaty problem they can solve and you have to prepare yourself for failure. That’s important because failure is research. In our case we need to introduce software that can categorise a receipt into taxable or non-taxable, simply from the receipt image.

Understanding what data will get us to this point, and how much of it we need, is integral to the success of making it happen. It’s not something we can take a stab at, but equally it’s not something we have the time to assess.

So an intern is a great option for a small business. They can dedicate themselves to the problem, apply what they know and ask questions that you wouldn’t have asked yourself. As a result they will put in place tests that show where your plans will fail and succeed.

A good mantra for any research project is ‘fail and fix, fail and fix’. You’ll then have a much more robust end product based on research with real integrity.

Thirdly, you need to get over finding someone who is a domain specialist. By which I mean, someone who gets what you do, be that banking, retail or accountancy. They do need to know a few basics but frankly you can teach them that.

What’s important is that they understand what it is you want the data science to do and what sort of pattern recognition is important. They’ll then be in a better position to help you see which decision-based tasks could be performed by a machine.

Finally, and related to this, it’s worthwhile having an expert who can guide them. Interns will invariably be taking theory and applying it for the first time so they need help to navigate the real-life scenarios.

Embarking on a research project has moved us from a position of ‘unknown unknowns’ to ‘known unknowns’.

We now know what we need to do and how we should structure the way ahead in terms of scope, resource, money and time. It’s a big leap on and most importantly it’s a leap in the right direction. Not simply a leap of faith.

Paul Nicklin is Technical Director at inniAccounts which provides accountancy tailored specifically for independent professionals.