DECLUTTER YOUR AI STORIES: THE POWER OF 3

---

The key purpose of any effective AI story or data story/ narrative is to reduce your cognitive processing workload while presenting a high-impact message in the most palatable and easily consumable form.

But, given that AI and data analytics are heavily driven by data processing, the outputs often tend to become overwhelmingly cluttered and therefore generate a high volume of human cognitive processing load or efforts. Even if not for the ML engineers/ data scientists/ model builders, a story full of complex diagrams & crowded charts & multi-thread narratives, definitely distracts the business actors/ decision makers from their key focus areas. 

There are several indicative measures of clutter in an AI/ data story e.g. 

• Signal-to-noise ratio: Signal being the crux of the message and noise being all the peripheral elements that forces the key points to lose clarity (Ref. Duarte's Resonate)
• Data-ink ratio: the proportion of ink/ pixels to represent the key message/ data/ pattern in focus, vs. the total amount of ink/ pixels available
• Data-pixel ratio: similar to data-ink, the higher is the data-pixel ratio, better is the clarity of the story, given that the key message is more in focus, than matters that don't matter much

Even if these metrics are more directly relevant for data storytelling/ visualization design with graphics and data-plots and charts etc., the semantics apply equally well to narrative-based stories be it on AI or data. Any effective AI/ data story will have a strong narrative in focus, supported by a few visualization elements- simpler the better. 

How to declutter and improve the signal-to-noise ratio of your AI/ data stories

1. Focus on max 3 key target messages in the narrative: Follow Steve Job's golden rule of the Power of 3 in AI storytelling. Your teams may be doing a lot of experiments with AI capabilities. Not all of them are high impact. 
2. Choose either the top 3 highest strategic impact stories, or integrate high to medium impact narratives by stitching multiple related stories into ONE STORY OF THE STRONGEST AI CAPABILITIES AT YOUR ENTERPRISE: In principle, this is similar to the core logic of boosting techniques in ML models/ ensemble-based machine learning, where the focus is on combining strong leaners/ models with weak learners/ models and get the best of both worlds. 
3. If your story has too many threads/ problems/ plots addressed simultaneously, remove each of them one by one and evaluate how much impact is getting reduced. This way we can identify the relatively less impacting or contributing features/ threads/ plots i.e. the noises, and can reduce/ remove them subsequently. This is again quite similar to feature engineering and determination of initial weight vectors, or in sensitivity analysis, in various ML and XAI techniques. 

Example applications:

In the stories of successful E-KYC or E-KYX (X= customer/ supplier/ employee)- the most commonly deployed AI solution bundles, often the story-telling and subsequent discussions quickly degrade from the key business process challenge in focus, to the component/ task-level technical challenges. 

For example: The solution architects focus on excessive coverage of common tech/ dev issues such as:

• challenges in collecting data through massive web crawlers,
• checking input validity and data quality, 
• classification of positive/ negative news about the customers based on social media and web postings, 
• problems of variability of formats/ language/ currency/ recency etc. of the verification documents
• problems with sematic extraction of data from noisy/ blurred/ sparse/ bad quality images and scanned documents, etc. etc.

These are all valid tech challenges no doubt. But, what would the business leaders and process owners do with these information clutter? And, most importantly, what was the key business problem that the AI-powered E-KYC solution is supposed to solve?

An AI-powered E-KYC solution is supposed to exponentially improve the core KYC process TAT, costs and QoS. For example, a robustly designed and well-executed E-KYC solution can 

• reduce one KYC-process TAT from an average of 8-10 hours, to less than 10 minutes, thereby also improving cost efficiency (e.g. reducing cost per process instance), and freeing up process owners to do much more in terms of volume, with less efforts in same no. of hours
• It can reduce manual data capture errors, missing data/ values, other omissions, by 80% at least, and 
• can improve new customers' experience with the onboarding process by 20-30% at least. 

These are the Top 3 Signals- the key messages to convey, in the AI story on KYC. The underlying tech-stacks selection, effective applications and implementation challenges and details are just noises/ irrelevant plotlines in the story. 

Applying any or all of the 3 techniques mentioned above, in combination, we can declutter our AI stories and tell the most impactful ones, using the Power of 3!