Putting Artificial Intelligence to work

Here at Deloitte we're starting to see some fascinating trends and terrific use cases emerge in our work deploying AI and Cognitive technologies for our clients in their day to day businesses, working across a wide range of sectors including financial services, healthcare, telco, retail and beyond. I've captured a quick summary below, hopefully of interest and don't hesitate to get in touch if you've got any questions or need support.

Think of AI as ‘data science lego bricks’ - The term Artificial Intelligence is very broad. In Deloitte we use the definition that AI is any computer program or system which does something that “we would normally think of as intelligent in humans”; a definition coined by the computer scientist Kris Hammond. However even this definition still encompasses huge swathes of technology, and it’s useful to chunk up a little. The term ‘Cognitive’ helps us do this, and is a way to present AI as technologies that mimic (but I stress don’t replicate) the mind – think image recognition, analysing/reading unstructured text, machine learning, deep learning etc. Overall, AI represents a set of advanced data science technology 'building blocks' that have rapidly matured over recent years (or even months), that when combined with the mass availability of compute power and data mean we can build solutions we simply couldn’t before, and they are getting better all the time.

Take a ‘data science first approach’. There are a lot of off the shelf AI tools and solutions emerging, often industry specific. Some are great, some less so. At Deloitte we take a data science first approach. We look at the specific problem and work out what from our ‘AI toolbox’ we need to try out. And, yes, AI is pretty data hungry, but you’ve probably already got all the training data you need, for example in your CRM systems, email logs etc – and AI is the perfect opportunity to really use it.

AI deep learning technology is great for understanding unstructured content, like emails. For example whether a customer email is a complaint or is a warranty enquiry. Using an AI technique called 'deep learning', we’ve built great solutions to ‘read’ emails automatically and identify what they are about (without complex language training), even when these are highly complex (as complaints often are). We then combine with a robotic process so we can automate these emails being sent to the right team member, or even automatically answered. A terrific use case, many of our clients receive thousands of unstructured emails a month, and this is a great way to speed up the process. 

Chatbots are getting WAY better, MUCH faster to train, and they can teach you a lot about your customers. The ability to combine natural language understanding with speech to text technologies and deep learning mean we can build chatbots that get it right most of the time, and thanks to new technology we can get them operational really fast. You also learn a huge amount about what your customers really need (as opposed to what you think they want). Most of our clients also find out that their FAQs need updating ;-)

Automating repetitive reports/commentary is now possible using AI technology. Think about high volume reporting where the structure doesn’t change but the commentary does – “this went up because of this, this went down because of that’ – there are now some great technologies that can dramatically speed this up

Don’t start with the hardest stuff, start with the boring. People often try to automate the most complex difficult things with AI. In reality that’s what you want your people to do. Instead focus on simple, boring, repetitive things that your teams will thank you for automating (like reading thousands of emails and forwarding them to the right person, see above)

You can get started in AI with less investment then you think, and you don’t need a super-computer to do it. AI projects DON’T need to be career limiting investment decisions. At Deloitte we take an evidence-led Proof of Concept approach, moving to production pilots and full roll-out in a proportionate manner when the technology looks good and the numbers add up. We use a blend of open source tools as well as those from the big technology companies, and it can often run locally (which means no sending all your data to an external cloud)

And a few final thoughts: 

• Consider how team roles & profiles will change - have you got a data science strategy and team?
• Try to move fast and work seamlessly across IT, Business, ecosystem - AI needs agile, multidisciplinary teams that bring the best of business, process, data science, digital and IT to the table
• Innovate, build labs, combine datasets - find ways to spin up experiments, and allow things to fail so you can learn and try the next thing

Don't hesitate to drop me a message if you would like to discuss further, and in the meantime some further reading here: