Seeing through the AI razzle dazzle

Everything seems to be AI these days, which can become overwhelming when you have to think about how it can be used in your industry or business process. All whilst the market is screaming from all directions with their AI message.

The AI razzle dazzle will continue. It will probably accelerate. Expect a ChatGPT every one or two years. Although let’s also hope that one day, AI will be less focused on dazzling AI capabilities, and more focused on how these capabilities will help companies own (or redefine) valuable enterprise workflows.

To survive (and hopefully thrive) in this environment, I think you need to focus on two things. Firstly, you need to maintain a learning orientation. It is your responsibility to continuously stay up to date, do your reading and research of WHAT AI can do, not necessarily HOW it does it. Don’t parrot what’s popular or the latest fad. Do the work, to see through the razzle dazzle.

Secondly, simply be a master of your business process. Then, with the knowledge from your own research, you’ll realize that in your business process, not all AI is the same!

Master Your Business Process

In my short career as a mental toughness coach, one of the big lessons I learned is the following: You don’t have to be a master of the content, you need to be a master of the process.

You can coach a CEO without being one. You can coach an elite athlete without being one. Sure, it helps if you have an understanding or affinity with what your coachee does and understand relevant jargon, but it is more important to understand the coaching process and guide your coachee through it.

A similar approach and mindset can be applied to the use of artificial intelligence in your business. You don’t have to understand the details of how AI works, you do need to have understanding about what it can do, but most important is to identify where it can add value to your business process. Or even better, make up new business processes or operating models where AI’s capabilities can shine.

So be a master of your businesses processes first, then understand relevant AI categories then think through the benefits AI can bring to your business process.

Relevant AI for a Decision-Making Process

Let’s take an example with the below schematic, a best practice decision-making framework, which was used for decision IQ research. Let’s see how we can apply AI for every step.

• Alignment, Framework, Stakeholders

A good decision needs proper framing. One of the biggest biases in decision making is wrong framing. In a business, we need to find stakeholder alignment around framing the decision before we start working on it.

Generative AI can create new content and has many capabilities to enhance productivity. It can code, make your presentations, write an article in rhyme, a letter as your lawyer, create a new brand logo, or make a forecast for you. GenAI (Large Language Model) techniques can be used on your internal enterprise data, so a user can prompt it to get some insights with some nice graphs.

But don’t let it make your decisions. It hallucinates (makes up stuff), it shows drift (gets dumber in some areas), it struggles with negation. It needs to get trained, so it doesn’t have real time data. For now, it can’t possibly make operational short-term decisions. If you want to use today’s weather in your decision making, or the latest customer orders? Don’t ask GenAI!

In decision making, generative AI can’t be used for short term, operational decisions, and for high impact, longer term decisions, it is best used as the whacky brainstormer to provide guidance and align around how to frame the decision. Ask GenAI; How should we approach this, what actions can be taken, what unforeseen consequences can there be? Using ChatGPT together in a brainstorming session can actually align stakeholders around how to solve a problem and framing a decision.

See here an example where Lorien Pratt, Decision Intelligence co-inventor, uses ChatGPT to frame a strategic problem. And here is an example where Erik Larson uses ChatGPT and three questions to improve a decision.

• Assessment & Speed

Larson mentions, “decision-makers need insights and recommendations from reliable, highly informed domain experts with deep knowledge of their business and their specific situation.” Here, AI can’t be a whacky brainstormer but needs to be a trusted advisor. Especially when the stakes of a decisions are high. In business, you wouldn’t risk hallucination before making a million dollar decision. In life, you wouldn’t want a hallucinating health diagnosis that informs a decision to operate you!

Specific or Narrow AI is good at solving one problem (and it is pretty dumb for other problems, which it can’t solve). And for that problem it can become your trusted advisor. We can use ML algorithms to forecast, predict, categorise, find patterns or correlations in structured or unstructured data, recognize pictures, or calculate probabilities of different options and business scenarios. Using computing power, it can do so at speed (and at scale).

But what do you do with this output in your business process? How does it lead to a decision and an action. Any forecast, insights, or business intelligence (using AI or not) without feeding into a decision or an action, can be considered assessment/analytics waste.

So, be a master of your business process here. Work your way back from the problem you are trying to solve, the goals you have, the actions you can take, the decisions you have to make, the options and trade-offs that are possible. Only then it becomes clear where you would require AI/ML for assessments and analytics.

As Lorien Pratt, co-inventor of Decision Intelligence says; “Data, technology and AI must take a back seat to diligent understanding of the decision that they support.”

• Communication

We’re all experts of what AI can do here. Alexa has been on the market for about a decade, we talk to our phone, so most of us have experience using Natural Language Processing (NLP), let’s call it CommunicationAI. When we make decision with a machine, we want to have a frictionless interface, like voice (ConversationalAI), or text prompting and search. If we need to communicate outcomes to stakeholders, the AI should be able to automate this, translate this, and protect it. ProtectionAI has been used for a decade to detect spam in our incoming email communication. We use facial recognition and fingerprints to unlock our phones.

• Decision & Execution

This is where I start thinking about Intelligent Agents. An intelligent agent (IA) is an agent acting in an intelligent manner; It perceives its environment, takes actions autonomously in order to achieve goals, and may improve its performance with learning or acquiring knowledge (Wikipedia).

Based on the trusted advice of the analysis, AI can now even make a decision, and execute on this decision. High impact decisions we would still want a human in the loop. However, smaller operational decisions, or micro decisions, can be automated, within the boundaries and goals a human sets. At my employer Aera, we call them Cognitive Skills.

Intelligent agents will require multiple capabilities, like gathering data, analysis it, make a cognitive trade-off between options, decide and execute. IA’s therefore require smart workflows to orchestrate all these steps, as well to execute the final decisions.

• Learning

If we digitize our decisions, the decision steps, the decision itself, the decision maker, decision value, decision impact and the decision context are all data points that can be digitized. Using a decision memory, machine learning can be applied to estimate the likelihood/probability of decision success and impact to augment a decision maker accordingly.

The decision memory can be used to learn from and continuously improve decisions, hence improving its quality. Although this can simply be machine learning calculating probabilities of success or failure, let’s call this LearnAI.

• Improving

If we want to improve, we want to understand what went wrong. In supply chain, humans often assign reason codes of why something failed or didn’t perform as required. These are classification problems which can be solved by machine learning algorithms. We can apply these classifications/reason codes across our decision process, understand them, and start addressing them to improve. Examples are LIME or the SHAP value. This is AIX or explainable AI, which also can help to build trust between human and machine.

• Optimisations

Many optimisation techniques, like linear programming, originate from operations research (OR) and date back to the 1950’s. These techniques are still valuable to make better decisions, for example in multi-echelon inventory optimisation, production scheduling, or network optimisation. In artificial intelligence, AutoML can be used to finetune parameters of the machine learning algorithms in a closed, automated loop, in order to continuously improve and therefore optimise the outcome of the algorithm. AutoML can be applied to many types of machine learning you have identified in support of your business process. OR and AutoML can be integrated.

Conclusion

Like any big project or problem we’re taking on, it is useful to cut them down in smaller problems and start understanding and solving those. Breaking down a decision-making process to assess where AI capability can be supportive in every process step is not unsimilar, as I tried to show.

If you master your business process, I’m sure that with a bit of effort, you can create an understanding on where AI could benefit. Just promise me you won’t do it the other way around and try to solely fit in the latest AI fad into your business process.

Featured picture credit: created with DALL-E : “A human being confused by Artificial intelligence in a desert digital colorful art”