The Battle Between Enterprise-Representing and Consumer-Representing AI

gbekdash
Aug 10, 2021
8 min read

Do you want another company to always stick themselves between you and your customers? A titanic battle is looming in Conversational AI, which is becoming not only a technology to make business more efficient, but also an interactive storefront to project your products and services to the world. If you don't do it well, somebody else like Google or Facebook will use their own AI and their own consumer relationships to do it for you. Your agenda becomes secondary to theirs.

So, the battle will not simply between competing AI vendors. It will also be between enterprises who use AI to grab new verticals and established players. And also between some AI vendors and their customers and possibly governmental agencies. But the first battle is probably going to be between #Consumer-Side-AI and #Enterprise-Side-AI. Get ready.

A titanic battle is looming in AI-enabled automation. At stake is who will be the storefront or face of businesses and which ones will win with higher customer sat and lower cost. As AI becomes the “face” of all major businesses, enterprises must act to make sure they represent themselves otherwise they risk enabling somebody else with a different agenda represent them. Enterprises that wait too long will have some crucial choices made for them. The AI battle will not be only fought between AI software vendors, but also between some AI vendors and their prospective customers and possibly governmental agencies.

We’re going to map out a landscape starting at the strategy level. A technical comparison is of course also warranted in a future article, but strategic fit must take precedence over a feature comparison.

1 The Competitive Landscape of User-Facing AI

AI is so huge because it can intermediate almost any interaction and add value to it. For simplicity, let’s represent AI architecturally with an unstructured input whose meaning is not understood and a digitized output that includes an understanding of what needs to be done based on actual data. The “what to do” is what AI is about.

Let’s make a distinction between an AI system that can only determine the “what to do” from one that can determine “what to do” then “actually do what needs to done” (or simply “do it”). The second one is a much more complete automation platform that we will refer to as AI-Enabled Integrated Automation (AIIA, or AI2).

AI2 (in blue) is more than just pure AI. It has additional automation capabilities besides

intelligence, such as robotic process automation (RPA), orchestration and workflow automation (WF), and large-scale real-time feed processing to handle high machine to machine traffic. It also has the ability to dispatch work to “underling” systems and monitor progress as well as continual learning capabilities to recover from and learn from failures or escalations to humans.

Many expert analysts are now advocating this approach. However, note that AI2 is not common yet. IPsoft’s newest platform, 1DESK, is very close, but it has been many years in the making. New comers to automation like Microsoft and others announced plans to develop similar automation capabilities using various descriptions. Many automation vendors are likely to follow that faster approach of approximating AI2 platforms by marketing and stitching together various separate systems from multiple sources. But such stitching is unlikely to provide the benefits of a unified AI2 platform.

AI can intermediate almost any type of interaction. The figure below shows some of the more important ones but fundamentally interactions are between various groups of people and machines.

AI will be deployed and classified in many ways, but the most important strategic classification is where to put the AI and who will supply it and direct it. One broad way to categorize the options is show the position of AI relative to the relationship between customers and enterprises, as we show in the picture below. On the right side, the AI is provided by the enterprise and of course, it represents their interests. On the left side, the AI is on the consumer side. Since consumers are not expected to develop or buy AI, so it will be provided by an intermediary; let’s say for now it’s Google for illustration purposes, but it could be Facebook, Microsoft, credit card companies, or others. Consumer-side AI will present the wants of the consumer and will represent their interests –modulated by the interests of the AI owner. It is possible also that an enterprising company may provide a pure AI service to consumers without the baggage of a Google or a Facebook, but that is harder than it seems, and it is the topic of a different article.

2 Enterprise-side AI

Here’s a peek at an enterprise that shows the different interactions that happen inside it. AI alone is only about making decisions on “what” to do. To add value to every enterprise interaction, we have assumed that the enterprise will deploy enhanced AI that has the execution and continual learning abilities that are part of AI2.

Conversational and cognitive capabilities enable AI2 to converse with humans on any medium and deal with many types of documents. The ability to handle high-volume real-time feeds enables AI2 to extract meaning from machine-to-machine communication. Integration and workflow and robotic automation enable it to take action using the best out of many possible pathways. As shown in the middle leg, AI2 relieves the enterprise from using many internal agents or SMEs to “translate” what other employees want to action. In other cases, it also relieves people from doing a lot of grunt work themselves (as in the rightmost side of the figure). AI2 will also enable the enterprise to offer its customers and employees new services that are just too expensive or difficult to do with people.

One key interaction is between the enterprise and the external customers. For simplicity, we showed that the customers are individual consumers but the analysis is similar if they belong to enterprises. The consumer-facing AI can combine and leverage all the knowledge the enterprise (and possibly their partners) has about the consumer to provide the best possible service.

Historically, many enterprises deployed automations, such as Interactive Voice Response (IVR) or equivalent chatbot systems that save it money but at the expense of the consumer. Consumers gripe and joke about it, but they can do little about it. But soon, there will be an opening for unhappy consumers to have automation that represent them, and enterprises that pursue automation that does not add value to their customers might find that their fate will be determined by the businesses that provide automation useful to the consumers. That brings us to the next topic.

3 Consumer-side AI

Here, consumers will use an intermediator like Google or Facebook to interface on their behalf with the enterprise. As an illustrative example only, let’s say that consumer John would ask Google Duplex to make a restaurant reservation at Chez Henry. Google will contact Chez Henry and make the reservation either through a robotic conversation with somebody from Chez Henry, or by directly putting the information in a back office system like opentable.com, which is now used by many restaurants.

It is quite likely that Google would also provide Chez Henry with additional information not included in John’s request. Google knows a lot of information about John from various Google platforms. From Chrome browser and Google Search, Google knows he queries high end foods and Italian wines and that it’s his wife’s birthday. From his Android phone, Google knows that John spends a lot of time at high end restaurants and that he frequently travels often to Boston, where there is another Chez Henry. The analytics may be very valuable to Chez Henry and they will probably pay significant money for it so as to provide a unique experience to John. John is happy, Chez Henry is happy, and Google is very happy.

But now John wants to move some money into some investment and he asks Google to get it done. His bank is not happy that Google (or whoever) is now mediating between them. Google might point John to another bank or investment. It’s no wonder that Google’s Duplex demonstrations so far were with small restaurants and hair salons. Large enterprises are unlikely to be enthusiastic. For that matter, even Chez Henry might notice later that Google told John Chez Paul is better. John himself found that Google is giving too much information about him, which he feels compromises his privacy and bargaining power, and he’s not so happy either. So the bank, Chez Henry and John will ask their senator to protect them and their senator may push the government for anti-trust or other regulatory relief, possibly to make the consumer information either generally accessible or completely private.

Consumer-side AI by parties that know a lot about the consumer can be quite powerful. Google, Facebook and possibly Microsoft and Amazon are well positioned. Other parties, like credit card companies know a lot of information (potentially) about consumers, but do not have the expertise on how to mine it or monetize it and are not naturally positioned to intermediate without some fantastic effort. But it is still doable.

We can see a sneak preview about this approach in China. Tencent (China’s Facebook) is already doing something similar, possibly with the government’s encouragement because China wants its own AI companies to grow rapidly and because it is not queasy about privacy. It’s no surprise that Facebook actions hint that they want to reproduce that strategy in the rest of the world.

In addition to the business issues above, Consumer-side AI suffers from an architectural limitation. It does not do anything for the enterprise internally. So, while this approach does offer the enterprise cost savings (as it reduces the effort it takes to interface with customers) and top-line benefits related to the analytics, it forces them to work behind an intermediary with their own agenda and it does not offer them any savings for internal work. Many enterprises will see such external AI as an existential threat.

4 What does this mean for enterprises and companies that provide services for them

There is a race underway that is becoming about more than saving money or increasing efficiency. Amazon, Uber, Airbnb and Netflix are examples of amazing companies that mastered the use of end-to-end automation. They now completely dominate their market segments and have rapidly revolutionized the business model of entire verticals. In fact, as Amazon shows, mastering the art of automation and process and customer experience design (all related to good AI practice, by the way) enables a company to move to new market segments relatively easily. This part of the race is about efficiency, providing additional services for less cost and an exponential accumulation and application of learning.

The other part of the race is more existential and is about who represent your company to your customers. Many companies have already concluded that since customer experience is increasingly dominated by various types of AI, AI will become the “face” of the enterprise or a big part of it. If that AI is provided by somebody else, many enterprises will lose control over their own image and their fate will become secondary to somebody else’s agenda.

The means to win both parts of the race is the same. Start the journey towards AI-enabled integrated and end-to-end automation. Dabbling with low-end chatbots, using RPA alone or relying on any single purpose or dead-end automation tool is becoming increasingly dangerous even though it might provide some feeling of security.

One common hurdle to starting a transformational journey is evaluating ROI on a project by project basis. When you’re starting the journey, no project is likely to have compelling hard ROI. Business cases are important tools, but consider that part of the ROI of early projects is training yourself and your own employees on how to use AI and end-to-end automation.