What Artificial Intelligence means to IBM SVP Rob Thomas – and what this should mean to you

In a far-reaching and candid interview after IBM’s #THINK2020 digital event, Rob and I discussed his insights into what’s happening with AI in the enterprise and where the biggest long-term impacts will be.  

My summary of Rob’s insights

Learn more from the interview itself, below.

    1. Too many companies confuse consumer-grade and business-grade AI.
    2. Companies have unrealized treasure troves of data scattered across the enterprise and various clouds. They should use that data, with proper analytics and AI, to enhance customer loyalty and raise business value.
    3. Today’s applications of AI technology have a long way to go. They’re about as advanced as car assembly lines were before World War I. The result? Businesses are spending more time making AI tools work than delivering new business outcomes with those tools.
    4. The “killer” business application of AI in the enterprise is the VCA, the Virtual Customer Agent/Assistant. IBM has drafted its NLU-Debater technology to enhance their VCA, Watson Assistant.

My perspective

  1. Google’s Meena – announced in January – and Facebook’s Blender – announced in April – advance the state of the research art for chatbot technology demonstrations, but they’re not ready for anything beyond consumer amusement. They’re not suitable for business use. They drop context and hallucinate.
  2. For the past four years, I’ve said that conversational platforms and applications like VCAs are the big game-changers.
  3. The combination of capabilities in IBM’s Debater appears unique. Future interviews with users, such as KPMG and other industry players, will explore that proposition.
  4. Specifics matter. There are many ways AI and analytics can impact business performance, but the number of significant examples demonstrating business model transformation within specific industries is limited.

The interview

Lightly edited for length and readability. Editorial comments are set off by square brackets [such as these.]

Hi, Rob. There’s one word that describes what I’m seeking here: insight. I am seeking your insights. I don’t care about corporate secrets or numbers or competitive positioning. I’ve looked at your blog posts and reviews of your books. I’m impressed. You have insights I want to share with all The Analyst Syndicate’s readers.

Let’s start with insights on business model disruption in particular industries. I’m tired of the broad platitudes I hear from consultants. Let’s get specific:

Where has AI resulted in a substantial business model disruption? Let’s focus on business models, not AI tools. How has AI already resulted in disruptions to business models in specific industries?


Here’s a good example that may surprise you. You probably know AMC Networks and some of their shows like Breaking Bad. AMC Networks is a public company.

We started working with them around their data and AI strategy. They produce a lot of content. And collect a lot of data. They have data from set-top boxes – what people are watching – and third-party data like Nielsen.

Their business has been creating content, then generating revenue off of that content – whether that’s from advertising or the content itself or otherwise. Here’s the insight they had – we helped, but they came up with it:

There was a huge, unexploited treasure trove of value in the data they were collecting.

Let me jump to the end: they ended up launching 2nd Party Media, a separate, spin-out business built on the analytical tools they created and data insights they had. 2nd Party Media sells this as a data product – infused across the board with AI.

So now, AMC Networks is in a completely different business. Nobody thought they would do that. They’ve done that because they understood their data, and they were able to apply AI. They didn’t have to hire an army of 1000 people to launch this business. The AI tooling is actually doing a lot of the work, so it’s a relatively small team.

This is also a good example of our strategy around the concept of the AI ladder, which is

Your AI is only as good as your data.

Companies that win in AI are going to be the ones that are best at collecting, organizing, and analyzing data, ultimately infusing that into their business processes – or into new business models or business approaches.


    • Built a hybrid cloud architecture for their AI.
    • Collected data on public and private clouds – federated those data sources
    • Build their AI model in one place
    • Trained it on data from all those sources

This is a good example that points to business model disruption. (It also illustrates how our strategy fits with business model disruption.)


This is a good example of creating a potentially low cost, high return business. It fits very well with the book “Infonomics: How to Monetize, Manage, and Measure Information as an Asset for Competitive Advantage” except you’re focusing on the end game: powering business model disruptions.

How about another specific example, other industries that have had their business models disrupted?


The impacts have been so profound and widespread it’s hard to pick only one.

The rise of FinTech would not have happened without AI being prominent and making it easier and more economical to do a lot of things that you have to do when you think about FinTech. A lot of FinTech has been around anti-money laundering, know-your-customer, governance, risk, compliance, and reg tech.

FinTech deals with a broad range of data sources. No human can possibly comprehend all the different transactions and monetary flows. It requires automation and AI to do any of that. FinTech as we know it today wouldn’t exist without AI coming to the forefront.

A different example: Retail.

Think about retail: the whole notion of retail, evolving from stores to direct-to-consumer models. Direct-to-consumer models would not be economical without the right practices around data and AI.

I don’t think that everything becomes e-commerce. There’s a big place for brick and mortar retail, but winning brick and mortar retail will present an integrated experience across online and brick and mortar, leveraging AI across both. As long as I give my permission, companies should know me online as well as when I walk into the store. And that’s what I see happening.

[A lot of the brick and mortar action feels defensive, seeking to close the gap with online.]

We hired the chief data officer in our division from Monsanto, where he had led their data and AI efforts. It’s something I wrote about in my book The Big Data Revolution: What farmers, doctors and insurance agents teach us about discovering big data patterns. It’s a good example of how this is impacting traditional industries like agriculture and farming.

But the impact is so profound that I think it’s time that everybody wakes up and sees this. Part of the reason people struggle is because of the role that AI plays in our lives as consumers, whether it’s smart speakers, or social media images, people start to think of AI in that consumer realm.

Too many companies confuse consumer AI and business AI.

AI for business is very different from consumer AI. AI for business is about automation, optimization, and making better predictions. It requires a whole different set of technical capabilities. That’s a distinction that not every company understands yet.


In one of your blog posts, you wrote about the evolution of transportation devices, from the 1769 Cugnot Steam Trolley to Teslas.

One of the points you made is that with business AI, we’re still primarily dealing with people hand assembling their own transportation devices the way they did at the end of the 19th century. We’re not at the point where there’s a clear understanding of what an automobile should be, who it should be marketed to, and what requirements it should deliver on. We may even be as far back as the middle of the 19th century, with factory owners trying to figure out how to replace water power with steam power.

With all your experience, you must have come across evidence of a new class of application – powered by AI, analytics, and data sciences – that’s beginning to emerge, one that’s likely to be commercially viable and disruptive. I’m thinking of the way ERP replaced a whole bunch of other applications starting between 1992 and 1995.

What is that new disruptive family of business applications, the killer application of the AI era?


In terms of adoption, two years ago, one outside survey pegged AI adoption at 4%. Last year, they reported an increase to 14%. Early this year, we did a survey that we published. It said it was looking like adoption was inching towards 30%. Regardless of who’s precisely right, I think it’s safe to say that adoption of AI in businesses is still less than a third of the way done.

[With no standards as to what constitutes AI adoption, most of these numbers are meaningless.]

If you think of that in the context of the development of automobiles, we’ve barely gotten to the realm of the assembly line. Back in the early days of the automobile, everything was custom.

What revolutionized that was [part-standardization and] the assembly line. And now we’ve gone well beyond that. If you use the automotive manufacturing maturity analogy, we are still barely at the assembly line stage with AI. We have a lot of custom approaches, a lot of people cobbling together different capabilities, different products.

People spend a lot of time kind of making the tool work versus working on the outcome.

That’s where we are today. So what is the killer application? I’m going to come to that point.

I have a pretty strong view of what the killer app is. It’s exciting for me. It’s revolutionary. You think about it, what does every company in the world have in common? Without a doubt, 100% of the companies have one thing in common, which is they have customers. Otherwise, they don’t exist. The other thing they all have in common is they want to serve their customers better. And AI is an amazing capability for better servicing customers.

In my mind, the killer app is what we call the virtual agent for customer service and care.  For example, IBM’s conversational AI application, Watson Assistant (WA), is used worldwide in multiple industries to help transform customer service. Recently, WA for Citizens has helped governments manage a large volume of inquiries during the global COVID-19 pandemic. Consider what MIT Technology Review has said:

As the coronavirus crisis has dragged on, understaffed government agencies, grocery stores, and financial services have all scrambled to set up similar systems for handling a new influx of calls. IBM saw a 40% increase in traffic to Watson Assistant from February to April of this year.

At this point, the adoption of virtual agents for customer services is, roughly speaking, somewhat under 20 percent of enterprises.

I have no doubt that virtually all companies will have this. I can’t predict if that’s in two, five, or ten years.

I have a conviction that 100% of companies in the world will be using AI as a virtual agent for customer service to better serve customers.

It’s not just about automating customer service agents. Certainly, you can use AI to deal with a lot of customer issues. But it’s also about scaling your customer interactions. If you use AI to automate 30 to 40 percent of your customer inquiries, then your customer service reps can focus on the most challenging problems. That’s how you drive up things like Net Promoter Score and customer satisfaction.

I think that’s the killer app. I’m not sure why that’s not well recognized yet. Maybe that’s something you can help us with, Tom. But it’s pretty clear in my mind, that’s going to be the killer app.


There’s a lot of synergy between what you said and what I’ve been writing regarding customer service agents, conversational platforms, conversational technology, virtual assistants, and so forth. I’ve been writing about this since 2016 as the wave of the future.

Let’s dig deeper into what’s slowing it down. Most virtual customer assistants today are intent-driven systems. Most decision-tree based systems do not do particularly well with multi-turn conversations and retaining context from one interaction to another.

In January, Google came out with a multi-turn conversational technology known as Meena. And in April, Facebook exposed Blender, their multi-turn conversational technology. They put on cool demonstrations that just aren’t realistic yet for commercial deployment.

Blender, for example, hallucinates.

How good a job is the IBM Research Project Debater technology going to do in terms of fitting into the conversational world for customer interaction in the future?


Watson has grown from research to experimentation to a range of AI offerings with over 30,000 customers. And, early last quarter, we announced that we were going to bring technology from IBM Research Project Debater into our Watson AI offerings.

Back in time, Debater was fundamentally a massive data ingestion and understanding engine that could then be used to develop arguments. And it could somewhat sort signal from noise.

We were asking ourselves, What would the application for Debater be in business?

As we were working on many different efforts around conversational agents and the like, we realized about a year ago that businesses are really a product of language and communications, whether that takes the form of an email, or a PowerPoint or Word document, or chat, or even a voice conversation. Businesses are a product of language.

We realized that we had something with Debater that could understand language in a way that no single human could, and do it at a rate and pace that not even an army of humans could.

Here are the three specific capabilities that we’ve taken out of Debater and brought into Watson to better enable customer service:

Creating summaries or briefs.

I ask myself, can I read a million documents and create a summary of what I found, and do that nearly instantly? With Debater technology, you can do that. And you can imagine how that would really help in a customer service setting.

Understanding topic clusters.

If you’re looking at a bunch of different documents, what are the things that fit together? How do I know which pond to fish in if I’m trying to solve a particular problem? It does a really good job of creating different topical clusters. If you could imagine a customer service environment, then you can point people where to look for the metaphorical needle in the haystack.

Working with context and idiom.

It has significant strengths in how it works with the rich fabric of human language. As it looks at a paragraph, it’s not just looking at the words in the sentence it’s focused on, but it’s also looking at the sentences around that sentence and what do they mean and where is all of this in the larger context of the document. Context is everything when you think about language. And idiom and shorthand are everywhere.

I think what we’re going to find – like with everything in this realm – it has to be trained on the data of the customer to maximize effectiveness. From that point forward, adoption should take off.

For example, we’re working with KPMG. They built an application using Watson capabilities like this for tax preparation. With it, Watson can scan, hundreds, thousands, millions of tax documents, and make the best recommendation to a client. That’s serving their customers. You could never do this without the kind of technology that I’m talking about.

Look at somebody like Royal Bank of Scotland who’s using this for their customer support, or Credit Mutuel or Regions Bank.

Some people might think they understand why the Royal Bank of Scotland and Credit Mutual would do it –  they’re big global banks. But Regions Bank is a small regional bank. And they’re using these types of capabilities, actually not to serve customers but to serve their employees. If you can do this for your customers, you could probably also do this for your employees because it’s similar data sets. So that’s why I’m so optimistic about this area.

I think the applications are wide, and I think what we have with Debater gives us a unique advantage.


Rob, thanks for your key insights here. To wrap: the two top takeaways are

  • Virtual customer agents are the killer app
  • Debater technology is helping power the transition from decision-tree intent-based systems to deeper, more capable conversational agents.

Actions for you

    1. Take note of the insights, not the omissions. This was a 30-minute interview.
    2. Now that you’ve read the interview, reread the summary material at the top, and revise it to match your own observations and needs.
    3. Contact me via to discuss this content.
    4. Send me suggestions on additional industry executives to interview. I’m particularly looking for executives who can share
      • Examples of “Disruptive industry-specific business model transformation enabled by technology”
      • New business-transforming applications or application suites exploiting AI that have been significantly disruptive

Reference Material

Rob Thomas is IBM Senior Vice President, Cloud and Data Platform. Here are Rob’s IBM bio, formal IBM blog, informal blog (with books published), and LinkedIn profile.




The views and opinions in this analysis are my own and do not represent positions or opinions of The Analyst Syndicate. Read more on the Disclosure Policy.


Leave a Reply