The Problem of Defining the Problem (I)

Why defining your business use case is key to building an AI solution.

When speaking to CEOs and business stakeholders, I find that while they want to invest in artificial intelligence, they don't know where to start. 

They have high level goals like "improve the customer journey," "make my supply chain more efficient," or "make timely interventions for chronic disease patients." However, for an AI solution, the stated goals are too vague. When goals are ambiguous, the build is significantly slower, the results are dissatisfactory, and it is hard to scale the solution.

The problem of defining the problem becomes the first roadblock to moving forward. Albert Einstein famously believed that the quality of your problem definition was directly proportional to the quality of your solution.

"If I had an hour to solve a problem, I'd spend 55 minutes thinking about the problem and five minutes thinking about solutions." - Albert Einstein.

Here are three steps to follow before you embark on an AI solution or, as it's frequently called, an AI use case. Defining a use case usually takes 1-2 weeks if all the relevant stakeholders are available for a discussion. 

1. Ask the right business questions
2. Investigate the data you have
3. Have (realistic) measurable goals

1. Ask the Right Business Questions

To understand business problems, you need to have your business stakeholders in the room (eg. from Finance or Operations). Then, you need to ask them to identify their three top pain points. More often than not, people have a legitimately long list of pain points, and the discussion can veer into a thousand different directions. Facilitating the discussion and driving it to the most high-value problems is an art in itself.

Usually, what I do is pick a journey, for example, that of an event, a person, a thing, which helps to unpack the problem. For example, here is a scenario that one business head of department presented to us:

"Our clients are complaining that whenever they highlight a problem, we take ages to resolve the issue, and they are incredibly frustrated. Honestly, I think we're going to start losing customers like this. It's ridiculous."

Now let's rephrase the question to dig deeper into the problem: "What is the typical journey of a complaint?" 

Quickly, you'll see that the problem starts getting more specific.

"Well, the complaint usually comes via an email or a call to the call center. Then we need someone to read it and put it in the right category - is it an outage, or a cable issue, or something else. The call center agent then puts it in a queue and the issue goes to an engineer. Sometimes the categorisation isn't correct and the engineer sends it back to the call center to understand the complaint better. The engineer then has to visit the site and figure out the problem, which can take some time especially if it's the first time he's doing it, and write a report. Then we send an update back to the client".

Now we're getting somewhere. Let's think about which parts of this process AI can make faster.

• Can AI automatically parse the emails, or listen to customer calls, and categorize the problem, eg. this is an outage problem?
• Can AI automatically assign an engineer who has worked on a similar problem before to pair the right engineer to the right problem?
• Can AI be an assistant to the engineer while he's on-site as a chatbot and recommend solutions to the client?

There are immediate benefits from this process automation: less manual labor, less human error, and faster time to resolution, which means higher net promoter score for customer satisfaction. If the AI model is unsure at any step, the AI will flag it to a human who will then take over. By observing the human's response, the AI keeps learning and improving itself.

By the way, 👆🏽 this is a problem my firm Addo AI tackled for one of our clients.

Now that you know the problem you want to address, the next step is to investigate the data you have to evaluate its technical feasibility, ie. do you have the data needed to build an AI solution for this.

In Part II, I'll speak about investigating the data you have. 

Thanks so much for reading. Comments welcome! 😊

Ayesha

Dr. Ayesha Khanna is CEO of Addo AI, an artificial intelligence solutions firm that builds intelligent data platforms and machine learning use cases for some of the largest enterprises in the world.

• Ayesha's website: www.ayeshakhanna.com
• Addo AI's website: www.addo.ai