How to Succeed with AI: Decisions First

AI is a hot topic and we get asked a lot by clients how they can succeed with AI or cognitive technology. There’s often a sense of panic – “everyone is doing AI and we’re not!” – and a sense that they have to start a completely separate initiative, throw money at it and hope for the best. In fact, we tell them, they have some time – they need to keep calm and focus on decisions.

The folks over at HBR had a good article about adopting AI based on a survey of executives. This is well worth a read and makes a couple of critical points.

1. AI really does work, if you use it right. There’s plenty of hype but also plenty of evidence that it works. But like all technologies it works when it works, it’s not a silver bullet.
2. Not everyone is using AI – in fact hardly anyone is doing very much with it. Most regular companies are experimenting with it, trying it out in one small area. Despite what you read there’s still time to figure out how to use AI effectively in your organization. Stay Calm.
3. AI works better if you have already digitized your business. Of course AI is a decision-making technology, so what matters here is that you have digitized decision-making.  Focus AI on digital decision-making.

To succeed with AI we have a concrete set of suggestions we give to customers, many of which overlap with the HBR recommendations as you would expect:

• Get management support
  The best way to do this is to know which decisions you are targeting and show your executives how these decisions impact business results. Being able to describe how improving a particular decision will help an executive meet their objectives and exceed their metrics will get their attention.
• DON’T put technologists in charge
  Like data analytics, mixed teams work best for AI. Make sure the team has business, operations, technology and analytics professionals from day 1. For maximum effectiveness, use decision modeling with DMN to describe the decision-making you plan to improve as this gives everyone a shared vision of the project expressed in non-technical terms.
• Focus on the decision not AI
  You will want to mix and match AI with other analytic approaches, explicit rules-based approaches and people-based approaches to making decisions. Most business decisions involve a mix:
  • Rules express the regulatory and policy-based parts of your decision
  • Data analytics turn (mostly) structured data into probabilities and classifications to improve the accuracy of your decisions
  • People make the decisions that involve interaction with the real world and poorly scoped or defined ones
  • And AI handles natural language, image processing, really complicated pattern matching etc.
• Make sure you focus on change management
  Change is always a big deal in Decision Management projects – as soon as you start changing how decisions are made and how much automation there is you need to plan for and manage change. AI is no exception – it will change roles and responsibilities and change management will be essential for actual deployment (distinct from a fun experiment).

AI is a decision-making technology. As such it is a powerful complement to Decision Management – something to be considered alongside business rules and analytics, and integrated into a coherent decision model. Here’s one example, for a company that needed to automate assignment of emails. This depended on who it was from, what it was about and how urgent it was:

• Deciding which client an email was from involved rules run against the sender and sender’s domain.
• Deciding on the subject of an email involved rules about senders (some automated emails always use the same sender for the same subject) and rules about subject lines (some are fixed format).
• This left too many unclassified, however, so the subject and body of the text were analyzed using text analytics to see which products were mentioned in the email to identify them (analytically) as the subject of the email.
• Urgency was hard too. Historical data about the client was analyzed to build customer retention model. This analytic score was used to increase the urgency of any email from a client who was a retention risk.
• Finally AI was used to see what the tone of the email was – was the email a complaint or a problem or just a description? The more likely it was to be a problem or complaint, the higher the urgency.
• Each of these sub-decisions used different technologies but were orchestrated in a single decision model to decide how to assign the email.

AI is certainly new and different, but success with it requires the same focus on decisions and decision-making. Put decisions first.