TELL YOUR AI STORIES RIGHT FIRST TIME

---

Example of AI storytelling- The AWS way

https://in-resources.awscloud.com/the-art-science-of-ai-storytelling

AI Storytelling #1: Why, When, and Where to Start?

Effective AI storytelling is The Most Critical capability that can pivot an organization to the targeted and drastic changes in the ways their businesses and people work, with AI leverage. 

AI storytelling effectiveness is a multi-dimensional metric:

=> In terms of Time & Sequence aspects: AI storytelling is relevant both in pre and post execution of your AI strategies, e.g. to define business-relevant leading and lagging metrics & impact indicators -

• Pre/ planning phase/ leading indicators: In AI leverage opportunity identification, assessment, business case & usecase prioritization, tech & talent resource planning
• Post execution phase/ lagging indicators: targeted vs. actual CSFs and Change KPIs achieved, variance explanations, metrics-driven and proven success stories, value realization & business impact articulation & communication- internal/ external markets, customer & user feedback
• During implementation/ release & run metrics: The targeted AI stories keep the data science/ ML teams in the AI-ML-D&A CoEs and their business users and executive sponsors integrated in terms of shared communication. This helps in collaborative project tracking & management of conflicts & complexities & potential overruns.  Thus, the AI stories keep all key stakeholders on the same page, focused towards achieving the Big Picture target state in terms of impact. 

=> In terms of Impact, good AI stories cover:

• the most critical culture change and adoption@scale aspects of People (employees, customers, suppliers, partners- across entire enterprise ecosystems). As per our 2020 surveys, this soft-change dimension is perceived to be the toughest challenge faced by senior executives, in achieving their targeted AI leverage.
• the critical dimension of Process changes e.g. erstwhile manual decisions and actions processes getting hybridized & augmented in terms of speed, cost, quality, volumes, consistency and reliability, with IA (intelligent automation) or Autonomous AI leverage
• the relatively easier-to-achieve dimension of Technology changes e.g. autonomous agents-assisted migrations from SAP to S4/ HANA, or of the enterprise infra and apps stacks to IaaS & PaaS on hyperscaler cloud platforms like Azure/ AWS/ Google/ IBM and need-based combinations of their capabilities. 

AI Storytelling #2: Current state

As emphasized in subsequent sections and notes, AI storytelling practice in nearly 95% organizations, is nascent at best. Given that AI is still viewed as a Tech-led initiative as opposed to a 'Business Data-led' one, many organizations showcase their AI usecase pilots as success stories, even if majority process owners and used impacted by these stories have neither any visibility nor any say in it. 

Just having yet another tool in the organizational arsenal obviously doesn't guarantee its value, outcomes, or leverage. It's almost like saying- "I have bought the exact swimsuit that Michel Phelps wears, or the costumes that Nadia Comaneci had, hence I am a swimmer like Phelps, or a gymnast like Nadia" :). Of course it's commonsense that such claims are clumsy. But, sadly enough, in the context of AI stories of many organizations and/or their TSP partners, it reflects current scenarios. 

This is precisely a key reason why AI suffers from the huge Pilot-to-Prod gap, as reinforced by Andrew Ng. The only exceptions to this phenomenon are some of the early adopters e.g. a few US and EU BFSI co.s and surprisingly even fewer global ITSPs like Accenture and IBM.

AI Storytelling #3: Why, Where and How to start?

Q1# Why do the AI users/ developer teams need to hone their AI story-telling skills?

Gartner has recently predicted that by 2025 data storytelling is slated to become a multi-billion dollar tech-stack and practice business in itself. Data storytelling is already a big tech supply-side portfolio, in terms of combined capabilities from various related technology dimensions e.g.

• data and model visualization platforms with canvas & toolboxes: Features range from simple charts and scatterplots to regression models, time series and moving average plots, to multi-dimensional visualization
• Use-cases of NLG for autonomous report generation from data and model outputs
• Easily interpretable and comprehensible model output descriptions in mixed forms combining graphs and narratives, targeting business users and decision makers who do not have any math-stat background to be able to decipher and interpret the outputs of a model during the inferencing phase

These tools and tech-stacks have existed for quite a few years now, and have been adopted by several organizations that were early starters in the analytics and data technologies space. Even then, adoption of AI@scale, across the enterprise, has still remained a distant dream. In more than 90% organizations, AI initiatives have failed to move beyond POCs and pilots. Data storytelling has definitely worked well within specific application contexts in some pockets / siloes e.g. specific tasks, activities, decisions, processes. But it has not yet been able to make an organization-wide transformational impact of strategic relevance. 

Q2# Are data story-telling tools and techniques enough for creating & delivering effective AI stories? 

NO. 

Andrew Ng highlighted this major gap in one of his most current interviews, that the POC-to-Production gap in enterprise AI-ML-data science applications, is only increasing, no matter what tools and platforms are used. This is happening primarily because of 2 reasons, one feeding into the other:

• Model accuracy values and relevance achieved during the POC/ pilot stages are going down the drain the moment enterprises are trying to scale them up beyond the controlled experimental environments of the POCs, and stretching them into the real world. 

• Due to the rapid erosion of model accuracy and value when deployed in-prod, they're soon becoming unusable and redundant. Naturally, as a consequence, the business users literally at the receiving end of these model outputs, are not trying out and relying on the models any longer. This is triggering the vicious cycle of decreased adoption, even before the adoption had actually hit the targeted usage-scales in-prod. 

Q3# Are the current AI storytelling practices, as mere extensions of data story tools, delivering targeted strategic value?

For no fault of the excellent data story-telling tool-stacks, the adoption of the models in production still remain abysmally low. Our surveys over the past 2 years (2019-20) have shown that percentage of AI and ML usecases moving from POCs to production and remaining successfully deployed and used, have been hovering only around 8-10%. 90-92% models never see the light of day in real operations. 

Interesting to note: These usecases are often NOT the high-risk, never-tried-before, disruptive and super-innovative ones. Most of them are the me-too usecases, ranging from customer churn predictors to loan default prediction models to revenue forecasting to claims fraud detection kind of mainstreamed models. Showing real AI leverage in terms of strongly differentiated competitive advantages, with these red-ocean AI capabilities, is an impossible feat to achieve.

Q4# How is AI story-telling much broader and bigger in terms of scope, objectives, capabilities and strategic relevance, than vanilla data story-telling tools?

Based on the analysis above, it can be deduced that, machines are learning to learn better thanks to newer algorithms and techniques being invented everyday. But, we the humans who are designing these learning machines, aren't learning much from our own past mistakes e.g. failure to convert AI applications from POCs to production assets of key strategic relevance for a company.  

This is the huge void in AI-ML adoption, rather a chasm, that AI story-telling needs to address and fulfill. As Andrew Ng pointed out: Most of the pilot/ POCs that enterprises conduct in early-stage adoption of these relatively new technologies, use homogeneous training and test datasets, often drawn from controlled data environments. The diversity, heterogeneity and richness of the contexts that businesses (and their representative models) face in the production environment, are not captured in the POC versions. These critical but missing dimensions are the AI strategic value levers that good AI story-telling practices enable, from the ground 0 of AI-ML adoption in any organization. 

Q5# What are the top 3 dimensions of a good AI storytelling practice?

Good AI story-telling practice makes a world of difference, where:

1. During the inception, ideation and design stages of AI CoE, strategy, practice and usecases, the AI story-telling STARTS from the BUSINESS PROBLEM and NOT from the tech solutions that are available. 
2. AI storytelling covers multi-faceted aspects of the target business problem- described in and covering the most holistic and heterogenous scenarios possible =>
3. Good AI stories eventually lead to rich data stories e.g. training and test data sets that represent the diversity and richness of the model application context: The business problem. 

Net net, the business problems and the direct & indirect business stakeholders affected by it are the REAL HEROES of a well-built AI story.