Enterprise AI Implementation: Theory vs Practice
The current pandemic will forever change our lives. The digital, contactless, and automated society is now an inevitable future that will come much faster than anyone could’ve predicted. The shift to eCommerce and the adoption of AI-based tools is becoming a top priority for businesses around the world. However, different organizations have vastly different maturity with incorporating AI into their business operations. How far along is your organization in this process? And where is your company along this AI adoption maturity curve?
This presentation provides an AI adoption roadmap that could serve as a maturity curve for you to see where you are on the journey to an AI-augmented society. Although each step along this maturity curve is a natural transition, this is where theory and practice diverge. Each step of this AI adoption roadmap has its own practical and implementation challenges. We will highlight some of these challenges and help you overcome them with best practice advice. This will ensure you have the greatest chance of success with your AI implementation, so we can help transform our economy into a more robust and resilient digital economy.
About the Speaker
Dr. Michael Wu's research spans many areas, including customer experience, CRM, online influence, gamification, digital transformation, AI, etc. His R&D won him recognition as an Influential Leader by CRM Magazine along with Mark Zuckerberg and other industry giants. Wu believes in knowledge dissemination, and speaks internationally at universities, conferences, and enterprises.
Full Transcript
Dr. Michael Wu: Hello everyone. I'm Dr. Michael Wu. And I'm the Chief AI Strategists at PROS. Today, we are going to talk about some practical aspects of implementing AI in enterprise. And we all know that theory and practice can be very different. And this is especially true in the enterprise where a lot of people are involved. But before I begin, I just like to take some time to go through some context to explain to you, why is this so important at this time? And the reason is because the Coronavirus pandemic have really greatly accelerated adoption of digital technology. So, McKinsey actually did a survey earlier in April this year. And their data basically shows that, it took many, many years for the US to build up a pretty modest digital user base. And with COVID, it only took a few months to catch up almost halfway. Now, what's not shown in this data is the relative speed at which things happen....
So if you actually add up all the users in this two population here, what you see is that the new digital user base here, so these are the ones in pink. So these are the user who are using digital channel for the first time because of COVID. They are only 46% of the regular digital user who had been using digital channels before the coronavirus pandemic. So, what's not shown here, is that it took about 20 years for us to build up this population here in blue, but it only took about two months for this new digital user base to catch up almost 46%. So this pink population is actually growing very, very rapidly. Now, moreover, this is actually a global trend. It's not just a US phenomenon. So, McKinsey also look at the online purchase growth across 10 different countries around the world.
So in this data, darker blue means there's more positive growth, whereas darker pink means there's actually more negative growth. So if you just glance at this data set, I mean, it's not too hard to convince yourself that a lot of this is actually blue, right? So there's generally an increase in online purchase across these 10 countries. So, now these are obviously consumer behaviors, so they are very relevant for B2C business. What about B2B? So this is Patrick Dupin. So I actually had the privilege of sharing a stage with him at a global conference in Switzerland. So Patrick runs a 300 year old glass manufacturer in France called Saint-Gobain. Now, Saint-Gobain told us that before COVID their digital sales was pretty modest, only at about 10%, but after the onset of COVID, it actually increased to about 80%. So, that just says that, whether you are B2B or B2C, it doesn't matter.
This digital trend is here to stay, and COVID is going to only accelerate digital transformation. So now, why is AI so important for e-commerce? So, if you actually think about this for a minute, if you are a brick and mortar business, right? Your business and your market is constrained geographically, but your competition is also constrained to those close by you as well. Okay. So now what happens if you actually go online and become an e-commerce? Okay? What happens when you go online become and e-commerce is that, well, the first thing is that your market and opportunity becomes much bigger. Because digital is borderless. So now your market becomes global, right? But your competition also grows as well, because now you'd be competing with every single online business on the planet. So how do you stand out?
How do you actually differentiate yourself? So for this, I think it's always instructive to look at what the leaders actually doing, and how they differentiate themselves against competition around the world. So, let's take a look at Amazon. So, one of the tools that Amazon leverage is artificial intelligence to help them differentiate. So which type of application do you think Amazon is using to help them differentiate in this case? Well, Amazon's actually use all four of them. So let's take a look at each one in this category. So first, let's take a look at internet AI. So internet AI is also called personalization AI, because they are basically recommender systems. Now, Amazon actually has a very famous recommender system called, customer who bought this also bought that recommender system.
So they basically recommend product throughout their e-commerce experience. Now, Amazon also uses these types of AI to personalize a lot of their customer experience online. Amazon is actually like controlling what you see, how they actually rank your search result to everything, to how you like to pay and how you like to receive your product. So internet AI, basically can help you personalize your customer experience. So next, let's take a look at perceptual AI. Now, perceptual AI is also called cognitive AI, because they basically are trying to mimic the higher cognitive function of a human being. And that includes vision, speech, and language. So from a technology perspective, they basically consists of some kind of computer vision system or some kind of chat bot or digital assistant.
So for Amazon, I'm sure you all have heard of their digital assistant, Alexa. So, with Alexa, now you can actually purchase online anywhere at any time with a simple voice command. So basically that makes shopping very, very easy. So, if you actually make shopping easy, people actually shop more. And as a consequence, they buy more. So now, what about computer vision system? So, if you give the machine the ability to see things as we do, if machines actually have an eye, what can you do with that? Well, the application that comes to mind immediately is probably face recognition or optic recognition. So Amazon actually take that to the next level and build a store that doesn't actually have a checkout line. You simply scan yourself in, and you pick up the item that you want, and you can just walk out. Okay? Because now this store actually have this sophisticated computer vision system that monitors everything, is tracking who picked up what, and will know to charge what item to the proper account.
This is actually a very sophisticated vision system because there can be many, many people in the Amazon Go store. And moreover, there can be more than one person going in under the same account. So you can scan your account, go into this Amazon Go store with your friend, and each one of you pick up some items, and after both of you have walked out, it will actually charge all the item you and your friend picked up to your account properly. So pretty amazing. So basically this perceptual AI, Amazon is using it to simplify shopping. Okay. So next, let's take a look at autonomous AI. Now, autonomous AI is the AI behind the self-driving car, but it can also be used to drive other machinery or other drones and robots alike. So now, Amazon is not building a self-driving car yet, at least not yet, but they are actually making use of this type of AI heavily in their fulfillment center.
A lot of the warehouse operation in their fulfillment center consists of robot that's working alongside with humans. So these robots consists of, I will say, robotic arms that have only rudimentary amount of intelligence. So they probably have some camera system on there that tells them that they should probably pick up stuff that looks like a box. They probably should not grab something that looks like a human. But there's also these newly intelligent shuttle robots, that's actually shuttling their shelf and inventory around the warehouse. So these shuttle robots are pretty much like administer self-driving car. They can shuttle the products to the right destination in the warehouse so that human can actually work on them. And they know how to drive in their designated lane. And they can even avoid collusion with other robots and other humans. They can even avoid traffics and congestions in this warehouse. So they are basically like a miniature self-driving car, except that they operate in a much more well controlled environment of the warehouse.
So now if you could shuttle products around the warehouse, why not shuttle them to people's house? So, Amazon is also experimenting with delivery bots so they can actually deliver their product, say from an Amazon truck, to somebody's home. The AI that's powering the delivery bots and shuttle bots are pretty much similar, very similar to this one that actually powers a self-driving car. And if you are able to shuttle these items to people's home, why not fly them there? So, Amazon is also experimenting drone delivery as well. So, as you can see, these type of AI is able to make a lot of these, what traditionally thought of dumb machinery a lot smarter. Okay. And they can automate a lot of physical tasks around the world. And Amazon is basically using this to automate their logistics. So lastly, let's take a look at business AI.
So business AI is also called decision AI, because their primary use case is in business decision automation. So, decision AI actually is very, I would say, diverse, because there's so many different industry, different business function that specialization out there, but they're very, I would say, obscure, they're not very well known to consumers. Because the users of these business AI are typically very specialized people in a specific domain. So, they're not very well known. So, Amazon also has a lot of use cases in using this business AI or decision AI. Now, I'm going to talk about two use cases here. The first one is fraud prevention. So, Amazon used to have a pretty sizable team of fraud investigators. So these fraud investigator will look at potential fraud transactions, and they will determine whether they are truly a fraud transaction and decide what actions to take to mitigate future fraud.
So now this actually generates a lot of data. So Amazon is able to use the data that they generate to train an AI to basically mimic this fraud investigators decision. So basically now, Amazon basically have this AI system that does automatic fraud detection, automatic recommendation of actions to deter future fraud. This free up the human so that they only have to review the action that's recommended by these AI system. And so that they don't have to do all the investigation that takes a lot of time themselves. So, the next application that I want to talk about is, dynamic pricing. So there's a report in 2013 that basically says that Amazon actually priced 2.5 million times in a single day. So, on average, that's about one price change per every item, every 10 minutes. So, at this scale, there's no human, not even a team of human could perform at that scale.
Okay. So Amazon basically leverage their rich transaction data to train the AI to help them price optimally. Now this AI, basically, it does pretty much what a human would do, right? It will actually analyze the market demand, the inventory level of that market, competitive pricing, and also a lot of other factors to come up with the optimal price and maximize the margin every 10 minutes for every item. Okay? So this is actually such a large scale that even with human experts around, they can't review every single price changed every time that happens. And frankly speaking, they really don't need to. They basically only need to review the outliers, where the price change result in very, very dramatic price differences. And even if they actually miss it, it's actually okay, because 10 minutes later it would probably be corrected anyway.
So by doing this very, very fast, Amazon is able to match the consumer's willingness to pay and actually drive more sales that way. So, these are both use cases that are business AI. And so, when you're actually using business AI, they can help you optimize your decision for your business. So obviously, AI is actually a very powerful tool and Amazon is leveraging it heavily to differentiate themselves in this digital world. But this doesn't happen overnight. Okay? The adoption of AI actually is a long journey. So, what does this journey actually look like? So, I've created a six stage AI adoption maturity curve that could show you what this journey would look like in the future. Because it could also serve as an AI adoption roadmap for you as well.
So, the first stage of this journey, basically is what we call, the digitization of work. So this involves turning our work into some kind of digital process so that they can be tracked and recorded as digital data. So this is all about generating the data which is needed later on to be consumed by AI. Okay? So evidently, the first step of digital transformation is basically, digitization. So when you're actually working on your digital transformation, you'll probably be going through this step already. So, if we're actually able to collect these data, and have our work being digitized and turned into data, we can already do a few things that can actually help us move onto the next stage. The first thing we can do is typically what we call descriptive analytics. We basically summarize this data to see how people work, and where the bottleneck is to help us reorganize our workforce, to make better decisions about the business.
Then we move into what we call predictive analytics, where we start to use this data to make inference, and make estimate, and make forecasts about something that we don't know. And finally it moving to what we call, prescriptive analytic, where these systems are even able to prescribe actions for you to optimize some kind of outcome. So now, if you're actually able to prescribe actions for people to take, right? Then basically you're ready to move on the next step. So, the next stage is what we call, exchanging data for automation. So when that happen, what that means is that, you are exchanging data to ask the machine to help you automate something, right? So if you can already prescribe some action for people to take, to optimize some kind of outcome, why not just do it for them? Right?
So instead of telling you how to go home with the GPS, right? Why not just drive you home with autonomous vehicle? Right? So this is the beginning of automation. So as we are actually collecting more and more data about how we make decisions around our work, we will have more and more data that will allow the machine to learn, to how to mimic our work. And that is at the essence of this stage. Okay. So now, obviously, these machines are actually, I would say, they're not very good yet, right? So we are giving the machine some kind of data to ask them to learn how we make decisions at our work, but they're not able to perform at the human level yet. Right? So the next stage is stage three, is what we call learning from humans.
So these machines are learning to refine their model of how we work, right? Every time the machine is actually making a recommendation of actions, basically we can either choose to say that, "You did a good job so execute this action," or we can say that, "No, that's actually not what I would have done." And then you take over and perform the task yourself. Right? Now, because through the stage one, we are already able to digitize your work, right? When you actually take over and perform the task yourself, the machine, the AI, is also collecting data on what you do this time that's different than what they recommend, and how you do it, and why you do it. Right? So it's collecting this data to help you refine the model of how you would have worked in this situation. So, if actually repeat this over and over again, eventually this AI will be able to automate your decision pretty much in all the different kinds of scenario that it could possibly have seen. Right?
So now, you are ready to move into what we call the fourth stage, what we call the full automation and job shift. Now, if AI is actually doing such a good job, right? In automating your job, why not just let them take over? Right? But you may say, what do we do as human then? Right? Well, the answer is the same answer as if you were training at intern or new employee, right? What happened if they do a really good job? You hire them to do your job, right? And what do we do then? We basically become managers of these people, right? So essentially, we would need to become the manager of these AI. Right? And let them take over some of our job. But the nature of our job also change. Right? We become managers. We take on more different work. We take on more work and different types of work. And we take on more responsibility as well. Okay?
So that's a full automation and job shift. Now, everything that we talk about so far, stage one to stage four, these are generally internally focused. Okay? They focus on the internal operations of the company. Although there are some that are actually starting to be focused on external. These can sometimes start as early as stage three. And clearly when an AI is able to perform at human level, to actually have full automation and create a job shift for us, right? Then they're also ready to face the external world as well. But the next two stage is primarily focused externally. Okay? And these are focused on other customer or other companies. So the fifth stage is what we call, the delivery of last mile. Right?
So obviously, after you're able to automate people's job, we need to go into this next stage where we deliver to the end consumers. So digital and AI is great, but there are a lot of things we can't do digitally yet. Right? I mean, we can't eat a meal digitally. I mean, all the item that we buy online, right? We could configure them, we could buy them, but a lot of them still need to be delivered physically to your home. Right? Not everything can be downloaded to your laptop. Right? And it can even be some kind of services that need to be performed by a human, for example, cleaning a window or something like that. Right? So basically, this stage involves a lot of robotics and machineries that are basically automating the delivery of this product or this service to you eventually, to the end consumer. Okay?
So they could be delivery bot, could be actually automating some people tasks such as, making you a coffee or something like that. So, in San Francisco here, we already have these robotic barista that will actually make your coffees for you. Okay? So these are entering the fifth stage. Now, the last stage is what we call AI augmented economy. Now, when most company are actually going through, stage one to stage five, basically we will start to have an economy that is augmented by AI. Now this is a very different world from the world that we're familiar with, right? This world will be actually, probably human will be working alongside with these machines. And we are there to actually help them deal with situations where data doesn't exist yet on how to deal with certain problems. So we will actually work a lot less, but we focus on the non repetitive nature of our job.
And sometime we may need to solve a problem once or twice, or maybe solve it several different ways to generate enough data for the AI to learn from us so that they know how to solve these problems in a different scenario also. But the important thing is that, because we work less, we actually have more time to focus on things that are more important to us. And so, the society, this is likely to have a lot of very important, very profound societal and economic impact as well. It may enable us to support a program such as universal basic income, so that people would be able to, I would say, work on the things that they're actually passionate about and not work because they need to survive. So, this is actually a very different world. It's a world that we can all look forward to.
So now, if you actually look at this journey, obviously we not there, right? We're not at this AI augmented economy, clearly. Right? And because there's actually a lot of challenges along this journey, and these are all the different challenges that could arise when you actually go along this AI adoption journey. But throughout different stages, there actually may be some that are actually more significant. So remember, so even though these are actually challenges that you will see all along this journey. Right? But they are actually more severe in certain stages, right? So now, where do you think most company are along this AI adoption roadmap? Well, not surprisingly, most companies are actually still early in the stage, right? Of this digitization work, this digital transformation stage. But if you are a PROS customer, you are already at stage two and sometimes at a stage three already.
So, because all of the AI solution that you're using from PROS, we are actually taking your transaction data, your booking data, and basically automating some tasks of analysis of pricing for you. Right? So, we are actually exchanging your data for some kind of automation by this machine. So because this is a PROS conference, when you're actually in stage two, basically, you will actually experience a lot of adoption challenges. So adoption would be the most challenging thing that you will see, right? Even though they may exist throughout this entire journey, right? It will be most prominent in this stage, in stage two. So let's see how we can address this adoption issue. So, adoption, there's actually a strategy throughout my career working to help company through various technology adoption, I developed a three phase adoption strategy that works fairly effectively.
So, if this was your entire employee base, right? As with most company, you will have some stellar employees. So these are marked with a star. So meaning they are stellar, they're top line performers, right? And there's also, I would say, different levels of enthusiasm for artificial intelligence in your company, right? So these are the people who are really excited about AI, and eager to try them, eager to use them. But within your company, there probably are some they are not so excited about AI. Some may be moderately lukewarm, some may be skeptic, some may be naysayer. Some may even be frightened by the fact that there's AI coming into the workplace, right? So now, the phase one, the first step of this adoption strategy is to create a secret pilot. Okay?
Now, the secret pilot is designed to be a successful pilot. Because you select the people who are top performers but are also excited about AI to be in this pilot, and you must recruit these people secretly. Now this secrecy is actually important because it keeps this team, whether you select after you find the top performers who are also excited about AI, this tend to be typically a very, very small group. And because they're small, it's easier to kick off and there's less disruption, right? And because it's secret. So, if something goes wrong, you contain the damage, right? As you know, there's no perfect technology drawer ever, right? Every technology drawer is going to experience some kind of problem somehow. So, you contain the damage, but you also allow you to give it enough time to perfect this rollout process.
So now, once you've perfected this rollout process, you move on to the next stage. Okay? And that's recognition and socialization, right? Because you already perfected this rollout process. Right? And these are already top performers already. And they're also excited about AI. Now they should be able to achieve something that seem like humanly impossible, right? Because they should be able to realize the performance gain from using this artificial intelligence working with them or for them. Right. So now your job is to basically, you need to make sure that everybody in this company will know that these are the superheroes, they are able to achieve something that's pretty amazing. And so, you need to recognize their performance and socialize this assess across the company. Okay? But you also need to facilitate the curiosity, the conversation within the company, because this rollout was a secret, right?
So this is what actually tests your communication skill as a leader. You want to give it a hint, but you don't want to give it away too early either. Right? So that you create this curiosity. So people actually want to know more about how they are able to achieve something that seemed impossible. Right? So this creates curiosity through this social facilitation. And when this happen, this is actually a good time for you to roll out some educational program to educate people more about AI. Okay. And this actually help them reduce fear and understand how AI will be able to help them do better at their work. So now as you do this, people's enthusiasm basically grow. Right? I would say the naysayer and the skeptic, even they could become curious sometimes. Okay? And they will actually like to learn more about AI.
Now, when that happens, you're basically moving to phase three. Right? This is what we call opt-in gate. And the opt-in gate is basically a program where you can opt into, where anyone can opt into, but is actually a pretty narrow gate by design such that, not everybody can opt in at the same time. Okay. So, now you basically want to create a little bit of friction to create this kind of exclusivity, such that people drive more desire [inaudible 00:29:12] that only special people who could actually go to this program to be trained to use this AI to help them work better. Okay? So you create this exclusivity. Now, but this not only creates exclusivity, but this gate is there to ensure that people get properly trained and mentored so that they know how to use this AI effectively and efficiently.
So when you do this, basically, how do you decide who gets into this program, this mentoring program? Right? Basically you do the same thing as you do to create your pilot team, right? You pick the people who are most excited about AI, who are also top performer, right. But if you actually don't have a top performer that's excited about AI anymore, right? Basically, you lean more towards the people who have excitement about AI. Right? So in this case, because you already have a pilot group who's actually properly trained, they are actually very excited about AI, but you can actually accept some people who may not be the top performer, and even they themselves their performance can be improved by this AI. Okay. And basically, each round of this training and mentoring, you're doubling the size of this group of mentors that they are there to train the next group of people who opt in to participate in this program. Right?
And this basically build up a pretty large community of practice that can support your entire company. So whenever people have questions about how to use this AI, they know who to go ask. Right? Now, every time you do this, this doubling every time, right? Basically it grows very fast. And eventually, you basically can tip the balance towards some critical mass of this adoption scale, right? When you actually tip this adoption scale, right? Basically everybody will try to adopt even the naysayer and the skeptics, right? Even those who are frightened, they will have no choice but to convert because they don't want to be left behind. So now, I just like to say that, throughout this three process, there's actually a parallel phase that needs to happen at the same time. Right? So remember I said in phase two, this is actually a good opportunity to roll out the education.
It really, really helps to start this education program early. Okay? Because understanding of the technology actually reduces fear. And so, that people will not have such a resistance to adopt. And if you actually start this early in phase two, right? Remember in phase two you're recognizing people's performance and socializing their success. Right? So if actually you to start this early in phase two, people could actually see a convergence of theory and practice, and that will actually typically drive trust. Right? And basically people will be able to see, how what they've learned from this education program translates to real results that their peers are able to achieve that's real. Okay?
So, basically that is how you drive mass enterprise AI adoption. Right? So what does the future entail for you? So we know that due to the coronavirus pandemic, the future will definitely be more contactless, but what does that mean? That means, business will also need to operate in a more digital world. So the future is all digital. And to differentiate in this digital world, business need to learn to leverage AI. So the future will obviously be AI augmented. What that means is that, more and more of our work will also be automated as well. But now you know how to drive adoption of AI and overcome one of this major roadblock along this adoption journey. So basically, this future is actually here now. And this future is actually now in your hand. Thank you.