So you want to start AI? Here’s how (and how not to!)
41 percent of the 3,000 executives surveyed in a recent McKinsey Global Institute study are unsure about how to use AI (Artificial Intelligence) in their organisation. This is an article on where to start (and indeed where not to start!) based on my experience in the field for the past few years on a variety of AI projects.
Get Informed enough to be dangerous.
Business Leaders don’t need to know every AI technology and a set of use cases to match to each — the technologies can be overwhelming — let alone mapping them to your specific business challenges. A high-level understanding of the below is useful — these can help to build a picture of potential solutions to business challenges which can be validated in the future with others in the field.
There are loads of learning resources on the internet, but here are a few overviews to get started on some of the more common terms you’ll hear mentioned: (Machine Learning, Deep Learning, Robotic Process Automation, Natural Language Processing). The idea is to have a high-level idea about each capability. This will help to form a picture of what they can and can’t do and which group a given business issue might sit with.
Business First and the rest comes next
AI projects should be started with some sort of business benefit or value in mind. Sounds obvious? Less than half of the projects I have observed started with this in mind and the comparison to those projects that did were polar opposites in terms of their outcomes.
I was bought in after the start of one such AI project, where the customer wanted to do some ‘cool AI’ stuff that their users in the field (and let’s face it, the public) would see.
Of course, the idea was that it would make their field workers jobs easier, more efficient and more ‘tech’ focussed which all sounded great — in theory.
As my team discovered the real business problem, we found that in fact, the field workers were inefficient on the front line because they were spending so much time searching for information while back in the office. Their front-line vs back-office work days should have been at a ratio of 4:1 and it was the direct inverse. They didn’t need some flashy AI in the front of their business — the needed some powerful time saving AI in the back.
The solution selected was a Natural Language Processing capability — this meant they could ask questions of their large corpus of unstructured documentation and get the answers quickly. This meant they would only need to spend 2 of their 5 days in the office — much better than 4. While the new AI approach was not as ‘flashy’ or ‘visible’ to the public as their original selection (an iPad AI app essentially) — it has actually solved a real business productivity issue.
Start by: Focusing first on a high value business problem to solve and then worry about which AI technology (or cool flashy tech idea) to implement.
Data + Data Scientist = Success (only sometimes).
Just last week someone asked me ‘I have all this data, how much will it cost me do some AI on this and get an outcome using a data scientist?’.
The reality is, there are many data sets which return little or no value at all in the instances that they are (as I have commonly observed) thrown over a fence to a data scientist. Usually this is the approach when a project goal or business outcome hasn’t been defined and a purely data driven approach is being taken in the hope of finding a ‘diamond in rough’.
I witnessed this on one such project where a data scientist was instructed to ‘go find something’ in the data (yes, produce magic!). After several weeks (and expense) nothing was returned — the data scientist was highly skilled.
The issue was that the data sets did not contain enough of the elements that they needed to — so a reliable model could not be produced. It is a little like committing to baking a cake before allowing time to first check if you have all the ingredients.
We spent some time and found a more relevant set of data to solve the problem and while this came with Its own set of challenges to obtain, it was a far more profitable path to pursue than the original ‘dump and hope’ approach.
Start by: Performing due diligence on the data sets and make sure they relate to the business issue or proposed value before committing weeks and dollars.
Build the right team
After selecting the business problem to solve, and having identified the right technology to address the issue (or had someone help you do this) then it’s time to build the right team to get it done.
In my experience, small teams work best, around 3–6 people depending on the size of the project with a complimentary number of subject matter experts or team members who understand the business problem and domain knowledge around this intimately, these usually already reside in your business. This blend is critical to the success of the project.
Depending on the type of AI project you are taking on, you will need a different mix of skills. There is no point hiring a data scientist if the business problem you are trying to solve relies on an understanding of Natural Language Processing. Once the business problem is mapped to the right technology, selecting the required skill sets is fairly straight forward.
An effective way for business leaders to achieve a sense of direction on this might be to partner with experts or to build first pilots in partnership. Start small and use the process to inform an in-house capability down the track once the organisation has cross skilled in the new AI domain.
Start by: Partnering if you need to, it’s a huge domain and it will take time to learn which parts are and are right for your set of business challenges. In house teams can come when the capability matures.
I’d love to hear your thoughts on this topic or if you have any other questions or advice for others please comment -
About the Author: Lisa Bouari is an Owner and Director of outThought.co. Lisa has delivered roadmaps and solutions to a wide range of industries and executives to make Cognitive and AI technology possible in their organisation. With a focus on Natural Language Processing, Conversational Agents and Robotic Process Automation. Lisa enjoys the never-ending learning and excitement of her domain, meeting new clients and the odd salsa dancing night out.
