My journey with AI is progressing and as a great believer in active learning, I have been getting ‘hands on’ and ‘into the weeds’. I hope sharing and reflecting lessons learnt spreads the knowledge (and addresses the execution challenges that the technology pundits gloss over).

AI is not the silver bullet; the reality requires significant design purpose involving ‘sweat and toil’ to realize value. Rocking up to pitch AI without a solid use case or business proposition is like throwing mud on the wall and hoping some of it sticks.

The conclusion: there are no short cuts.

The Learning Journey – Execution Lessons

1. Quality of the prompt and question being asked is critical to elicit a complete and relevant response. A poor question leads to a poor and less relevant and ambiguous response. There is a excellent post from Anakin.ai blog [Mastering ChatGPT: How to Ask the Best Questions] which frames how to get the best out of ChatGPT:

• Be Clear and Specific
• Contextualize the Question
• Experiment with different Question variations
• Utilize prompts and examples

2. Format of the AI response is critical to structure the output insight. The construct of how the response is presented/ visualized is fundamental to comprehend the application results. This subject area seems to be underestimated in its difficulty and, arguably, its the advanced output revelation that you are seeking to gain by using AI.

For example, we are familiar with Netflix suggesting recommendations but without context relevancy and ranking, we end up with just another data list. Similarly if no consideration is given to how the output is structured, it becomes another mass of data that we struggle to make sense of.

3. Garbage In = Garbage Out. AI requires lots of good quality data. Where the volume of input data is limited, and its relevancy is questionable, the quality of the AI output will be similarly limited.

Furthermore, it is important to understand how AI derived the output; why did AI respond with that data or recommendation. Unfortunately this means diving into the detail. As part of the development and testing phase, this aspect is crucial to ensure confidence in a repeatable and reliable output. Configuring AI solutions will require several iterations as you experiment to determine an optimal context construct.

4. Engage Business User Subject Matter Experts. There is a direct correlation between the quality and validity of the output and the team engaged to support the AI development.

Despite what the data analyst or developer believe looks good, the accuracy and trust in the output can only be ensured if a strong team of experienced business SME’s are involved in the configuration and development process. The ‘devil is in the detail’ and the team allocated to support your AI initiative(s) need be at the top of their game (for the reasons listed above).

5. Business User Buy-In. As with any transformation; process, procedure, training and change management factors are key if AI is to be accepted and operated successfully. Open minds are required and there is a need to be receptive to an alternative way of working.

I am very optimistic that AI can be implemented successfully to accelerate and automate a number of front and back office processes. My experience of the journey so far essentially mirrors a transformation project; it entails typical business transformation challenges, and requires a steady and resolute focus on capturing value!

What has been personally surprising is the level of granularity and detail that AI examines and extracts – this has exceeded my expectations and has forced me to re-think how we can use AI to raise the bar even higher!

What is your advice on AI execution? What is your use case and what works and what does not work. Please share the knowledge.