5 Next-AI Service Offerings

 #1: Domain Ontologies as a Service (DOaaS)

 3-4 years ago, while at Wipro, we the AI (HOLMES) team were spending one of our usual weekdays with client teams. What was unusual about that day (the reason I still remember it  most vividly) was that- the client team was not just so much knowledgeable but also so creative & innovative about the AI application opportunities, in their organizational landscape.

Funny thing was- they were a traditional BFSI organization based out of EU, functioning successfully for decades, flourishing in one of the most mature & well-regulated markets and regions. They had every credible reason to be satisfied with their numbers and growth. Yet, the hunger for disruption using technologies were so prominent. 

The team that were visiting us were not core-tech or IT, but a combination of hardcore business services, business operations and CRM teams. There were 8 of them, senior directors and VP's, and ALL were equally comfortable in our geek-speak as well as their biz-speak.

No wonder that I was overjoyed. We started talking about why it took months, in some cases years, for these large banks to put even common AI usecases like KYC or anomaly/ fraud detection/ prediction, in production. At that time, I was pushing our HOLMES engineering teams to build domain-specific ontologies, lexicons, knowledge models and pattern bases. Key idea was that we could short-circuit the build/ validate/ test cycles of these mature & AI-starter usecases and didn't end up reinventing all wheels for every client, thus almost invariably overrunning their AI projects' go-live schedules.

While I was explaining how these prebuilt ontologies could make their AI use-case implementations more efficient, a senior client leader came up with a fantastic idea: "Why don't you monetize the domain ontologies themselves! Sell them as IP's/ knowledge assets, in an 'as a service' package, to us!"

Next morning, we went straight back to our strategy drawing boards, set up a small team of 3 senior AI engineers, and started building these "ontology boxes". Initially we chose to build 1 box for BFSI- one of our largest verticals- also most mature in terms of AI adoption, and 1 box for another very interesting corporate function- the Legal team- that gave a unique problem of contract validation, to us.

Given it was a brand new idea then, and we were building it from scratch, it took the small team almost 6 months to come up with a Minimum Viable Product i.e. a working ontology for the legal functional domain, to start with. The Legal team were so engaged with the small engineering team, throughout the MVS cycle, because their domain knowledge and experience and interpretations were the most crucial inputs for the engineers to capture and build upon. 

The prototypes in-prod proved that they could accelerate implementations standard AI usecases like contract intelligence, from years to weeks!

After leaving Wipro, many of us have seen tremendous readiness for these capabilities only in a handful of SPs- namely Genpact and a few IT SP BU's from the Big 4. This is such a huge blue-ocean AI service opportunity especially for the large SPs who have been delivering/ running vertical-specific tech & business services for 3-4 decades now. Large end-user client organizations love this capability simply because it speeds up their own AI adoption success cycles, while also indirectly giving them access to their domain best practices (i.e. in way of the comprehensive ontologies)- which is a huge value-add. Large SPs are also more well-placed to deliver these services way better than start-up's, simply because start-up's don't have access to that much of cumulative experiential domain knowledge! 

#2 Data Storytelling as a Service

Gartner has recently predicted that by 2025, data storytelling will become a multi-billion dollar AI-business activity in itself. Well, future is much faster than we think, as Diamandis & Kotler mentioned in their book with nearly the same name. 

Data storytelling has matured as a practice in top 1-5% early and most advanced adopters of AI & ML, e.g. in some of the topnotch financial services organizations. A handful of AI services start-up's have also started their strongly differentiated journey in this path less-travelled so far, in the past 1-2 years. The following trends are visible in data storytelling practices across the leading organizations:

• Data visualization technologies have become so mature now that it's not even considered innovative or fashionable enough to speak about, in the AI/ data sciences geek world. Even dynamic/ customizable data visualization have become relatively commonplace capabilities now. 

• Data Storytelling in its current form deals primarily with extending data sciences into the art of narrative generation or storytelling- presents data, analytics, AL-ML model outputs & recommendations etc., in easily consumable and relatable story form. 

• The process of turning facts and figures into contextual and relatable stories, can engage the targeted audiences esp. the business users much better. 

Data storytelling typically includes:

1) creation of a narrative structure for data/ model outputs/ inferences, 

2) generating contextually meaningful narratives on data & analytics output, 

3) exploring stories from datasets in multiple facets / aspects, 

4) tuning the data-driven narratives to a form that the targeted audience understand best. 

Data storytelling can lead the target audiences progressively e.g. through multiple layers, based on their exploratory curiosity, requirements, and expectations. Data storytelling, in addition to being on the output side of several AI-ML models, can also leverage AI on the input side, e.g. as a dynamic data storytelling service offering. It can leverage generative AI techniques and natural language generation modules (e.g. Quill from Narrative Science), to generate timely relevant and contextual stories based on data and ML model outputs. 

Start-up's in the AI space, specifically in the NLG/ NLU space can create dynamic and interactive data storytelling as a Service offerings, that can quickly enable end-user organizations to leverage these capabilities to transform their business interactions completely, both within (e.g. between functions like marketing & finance) and outside (e.g. in B2B or B2C conversations and negotiations e.g. on price vs. value).  

#3 Knowledge Pattern-bases as a Service

This is again a blue-ocean opportunity both for large global end-user organizations and their service provider partners, as this one also is based on integration of their experiential knowledge-bases, gathered through decades of operations. 

This is NOT a start-up type opportunity, for same reasons as stated in #1. In line with what 5 of the largest and otherwise fiercely competitive US banks did in their TruSight* initiative, pattern-bases especially for fraud detection/ prediction, or cyber-attacks and attempts, are extremely valuable and highly monetizable intellectual properties that the large organizations are sitting on. This is primarily because the "Cost of Not Doing It" is very high for the end-user organizations as well as their service partners. Any new fraud/ anomaly, or even attempts to fraud or cyber-attack, that are at least 50--70%+ similar to previous fraud or attack-patterns, can be predicted and hence preventable, if these pattern-bases are made available as a service. 

Again, only one of the Big 4's have demonstrated capabilities witnessed in 2020, where they could show what they built in way of knowledge-pattern-bases, as an offering somewhat similar to this. But even that one organization is also not delivering it as a Service. That is- it isn't monetizing the knowledge-patterns as consolidated stand-alone revenue opportunities. This offering is of great value for large end-user corporate clients- who run high risks of financial frauds or legal liabilities due to potential cyber attacks on secure financial and PII data. 

[*  TruSight was founded by a consortium of leading financial services companies, including American Express, Bank of America, Bank of New York Mellon, JPMorgan Chase, and Wells Fargo ] 

#4 ARMaaS - AI Risk Mitigation as a Service

This blue-ocean opportunity can be realized as an extension of the previous design/ idea, further enhanced by the unique AI application potential that AI is ALWAYS ON. Obviously, machines never sleep and can be programmed in a way (e.g. using RT unsupervised or reinforcement learning) that they will NEVER STOP learning and will keep themselves constantly updated and will predict risks/ threats based on their latest learnings/ most recent & up-to-date models. 

So, it's the best of both worlds- combining the knowledge of all pre-existing risk patterns (as mentioned in #3), plus adding the always-on real-time learning capabilities, so that the risk profiles are constantly updated and prediction accuracy is not data-staled or affected by latency. Hence, it's only an obvious logic that all risks that are high-impact, high-urgency, and require continuous monitoring and alert actions, are top candidate usecases for autonomous AI applications and systems. 

These risks can be business risks, operational risks, technical (e.g. hybrid cloud performance vs. opex optimization) risks. Irrespective of their classification, the AI capabilities built for Risk Management as a Service will be horizontally applicable to all types of risks. The business case for end-users for this AI offering is to be based on the size and time-sensitivity of the risk/ threat and their subsequent mitigation actions. 

Again, this is a huge opportunity potentially and visibly untapped by large service provider organizations, many of which have mature risk management service offerings but haven't explored their AI-empowerment in an integrated, full-blown manner. Only EY has demonstrated capabilities in this space, thanks to their highly mature risk management knowledge-base and frameworks and their progressive and innovative approaches to leverage this risk knowledge to deliver highly differentiated value to its clients.  

#5 Pre-trained models/ Verticalized AI as a Service

Genpact has initiated, led and mastered the pretrained accelerator space. So has Accenture - for certain domains and functions, with similar approaches but slightly different execution and communicational narratives. But, offering the accelerators themselves, like the ontologies, lexicons, APIs, even AI-infra sandboxes, all bundled up by verticals/ functions, in "As a Service" models, is something of a blue ocean still now. AWS Healthlake is a pioneer offering in this applied AI solution for the healthcare industry, as IBM Watson Health was, in terms of the curated training datasets of millions of scan images etc. 

Till 2019, we could understand why. 90%+ of AI projects, irrespective of verticals, were in POC/ pilots. If they were running in production, then obviously getting the pretrained models tested and validated for performance in the real world, wasn't possible. But, 2020 onwards, due to rapid adoption of autonomous technologies for remote work support and service delivery (the COVID effect), productionization of AI-ML in business and IT services has become a mainstream reality. Now is the time to build tried and tested pre-trained accelerator bundles by industries, sectors and functions, and offer them as services. This will further speed up implementations, bring in overall best practices and SP's experiential knowledge by the client domains, and will speed up AI's Time-to-Value curves exponentially, especially for large clients with economies of scale at work. Their RoI and breakeven from large AI programs will become much better and faster. The business case is right there, for the large end-users and their service partners. 

Amazon AI Conclave, and the Union Budget Next Monday Morning: Emergence of a Perfect Equation?

Amazon concluded the 4th edition of Amazon  AI Conclave in early 2021. Because of COVID guidelines it had to go virtual this year.
While I was writing this blog on my key takeaways from the Amazon AI conclave, I was also listening to the live telecast of budget, presented by our Finance Minister.  She was clearly listing down the financial and economic priorities of our country, to push it towards a growth path that was hopefully faster than the ones some of our smaller and hence nimbler Asian neighbors and friends have already started enjoying (e.g. Bangladesh and Vietnam).

A key benefit of my highly practiced multi-tasking skills suddenly dawned upon me- Connecting these dots. The key messages from the Amazon AI Conclave and the Union Budget priorities are so much in sync! Whatever the honorable FM was saying, I heard the technical equivalents of those initiatives and narratives, just 2 days before, in the Amazon AI Conclave.

Directionally, both were spot on. However, as all the experts were mentioning in their budget analysis,  the devil lies in the details, and in execution. Great strategies have failed far too often, due to unimaginative, conservative implementations. It’s one thing to say, and completely a different thing to actually do, and achieve demonstrable and proven outcomes.

Bridging this mission-critical gap between ambitious visions and average/ half-baked execution, is the task at hand- both for the country and for AWS, for India and the world to become the connected, ubiquitous, global powerhouse of AI, data, cloud- the three key pillars of the Brave New World:

1. What stood out for me from the very first session of the AWS Amazon AI conclave, is that the leadership has already started stretching their goals and execution plans in the right direction, in terms of moving beyond just tech-speaks and talking consistently about Customer Outcomes i.e. proven execution and measurable, demonstrated results. Everything that were spoken, had proof points from case studies that were already successfully executed and delivered in real customer landscapes, and had already generated proven results, e.g. bringing down ML model-build (hence productionization) times from years and quarters to 1-2 weeks!

2. The global AI-data tech supply-side shifts from labelled data and supervised learning to unsupervised and deep reinforcement learning algorithms were explained with brief, specific, relevant contexts of client applications and realized business outcomes like retail shrinkage prevention, and not just outputs in terms of model performance, AUROC and accuracy parameters.

3. The other key take-aways were multiple reinforcements from the Amazon leadership team-members on the importance of harnessing an “ML culture”. This is why the patented framework of AI-SWIT‘C’H includes ‘C for Culture’ as a key dimension, for operationalizing AI and intelligent automation in organizations. The leaders explained with ample examples both from within and outside their organizations, across clients and partner ecosystems, on how certain culture change levers and initiatives are bearing fruit now, e.g. ubiquitous access to quality-validated ML training programs & self-learning materials, the AWS ML University, and democratization of access to data and tech-stacks for AI.  [For details: ref Amazon Machine Learning Solution labs <https://aws.amazon.com/ml-solutions-lab/ ] Zomato, Edelweiss Tokyo life Insurance and Jubilant Food works (Dominos India) are some exemplary instances of ML deployments where Amazon ML Solution Labs led the deployment effort for ML model building.

The 6-step approach and the importance of prioritizing on the right projects absolutely resonated with the “Minimum Viable Strategy (MVS) for AI” guidance that AISWITCH has published in 2021 (https://aiswitch.org/ai-practice-tsp).

While the proven success stories were many, there are obviously even greater opportunities to explore and expand on:

Having worked very closely with the AWS tech teams during the early days of setting up the Wipro HOLMES AI & Automation Ecosystem, there are so many wonderful assets that we tried hands-on from the AWS ecosystem, which did not get adequate mind-space of the analyst audience. This was probably 1- for want of time, 2- also due to the known fact that the analyst community usually don’t have much hands-on tech exposure to the “Deep AI space”:

1. Sagemaker examples were mentioned, DevOpsGuru, CodeGuru were touched upon, for the hybrid “geeky analyst” folks like me, I guess. Healthlake – the healthcare datalake construct was briefly explained. These were like taking a ride in the Tomorrowland’s Carrousel in Disney Orlando, but the carrousel got stuck mostly in the current scenario.
2. AWS is the global Father Figure of Cloud, being the First Mover in bringing the cluster/ grid computing technologies out of the labs and making them commercially feasible in making businesses agile, elastic and digital (thanks to GPUs usage in a novel way, EC2 and the storage clouds). There are great stuff in SageMaker e.g. Ground Truth and Data Wrangler- these capabilities must be flaunted and brought forth so that business users i.e. the target citizen developers of ML get early exposures on what the right questions are. Of course these topics were extensively covered in Day 2 which was the technology edition by design. But, quick exposures to some of these, at least in terms of demonstrable "Ease of Learning", could have benefitted the business leaders and analyst folks who could then enthuse the citizen builders in their respective functions, in a more informed manner. 
3. AI-readiness is a leading indicator, not a lagging one. Hence, all the conversations, be it in customer, partner, analyst or tech research-space, should be feasibly ambitious & forward-looking. Most of the AI tech capabilities are already here, except in spaces like generative, super AI, and the 'Catch 22' scenarios in language corpus, with OpenAI's GPT-3 & Google's SMITH and BERTs & their domain-specific versions, and BIGGANs' semanticity problems, etc. 
4. Discussions must also include topics like Quantum ML (e.g. IBM's Quantum cloud equivalents- for which ML problems do quantum techniques make sense?) and Green AI, meta-learning, zero-shot learning, RT AI, XAI and data bias reduction techniques. Knowledge exchanges on AI should aggressively promote innovative lateral thinking, e.g. how the unsupervised/ reinforcement learning techniques produce indirect "labelling bias reduction" benefits, beyond just improving performance and precision.

Net net, using India as the “Most Complex Problem” statement and context, Amazon has more than sufficient deep tech first-mover capabilities to build and test these prototypes for our immediate future here, and then scale up quickly, for the world.

Technology-wise, AWS and other hyper-scalers are the Super Integrators of elastic data-algorithms-infrastructure, in the way of cloud and federated & edge AI. As the global IT services hub, India has a huge pool of service knowledge & curable datasets (can be masked for PII), and trained tech manpower that is agile i.e. that can willingly and quickly switch to AI-ML capabilities.

If this is not an example of a Perfect AI demand-supply equation, then what is?