Running Your First AI Project (Guest: Emmanuel Lai)

Andreas Welsch: 0:21

Today we'll talk about running your first AI project. And who better to talk to about it than someone who's just done that? Emmanuel Lai. So Emmanuel. Hey, thanks for joining.

Emmanuel Lai: 0:33

Thank you. It's a absolute pleasure to partner with you today in educating the world about what artificial intelligence is capable of doing and what are some of its limitations as well. I think oftentimes people I really like your motto turning hype into reality. Because often, oftentimes there's a lot of unrealistic expectations of what AI is capable of doing. But if you apply AI in the proper way, it can act, it can far succeed whatever expectations you had. The most important thing is realizing what it's good for, what constitutes a good use case and what's not a good use.

Andreas Welsch: 1:11

Excellent points. Hey, maybe can you tell us a little bit more about yourself what you do, what you have been doing how you've come to automation, and AI.

Emmanuel Lai: 1:21

Definitely. So I'm currently a vice president at a large Fortune 500 financial institution. I manage a I manage a RPA program for the cards and Consumer Services division. But my previous role was managing a automation program for a financial and global shared services organization. And where I led a team of developers and PMs to implement robotic process automation solutions. But I have always been a big fan of entrepreneurship and what do I define as entrepreneurship? Intrapreneurship is when you are a entrepreneur within your own organization. One of the things that I did as an intrapreneur at Mars Incorporated when I was serving as their regional automation expert was that I instituted and stood up a analytics center of excellence entirely from scratch. I took people who had no, these are all finance and accounting professionals with very little technical experience on, on how to develop alx workflows in automations. We gave them licenses and I sat down with them and I taught them how to develop these workflows, and we achieved a lot of success with Alteryx in that organization. Every license we achieved at least 200% ROI on in terms for, so basically for every dollar you invested, you got 200% back, and that as a very hands-on leader. I have never been one to sit at a ivory tower while other people do the work. Technology is my passion. One of the things that I did while running that Al Center of Excellence was that, I created a machine learning model. And this machine learning model actually predicts the outcome of whether or not a credit order was worthy of getting approved. And this is the first time that anyone in the organization had ever thought of using such a tool for doing something that was very highly manual. We already had some RPA on it that I personally led the team that implemented that. But we took it a step further. The RPA would automate a lot of the steps and put the credit order and all the financial details consolidated from all the different systems involved and put it in the face of the credit representatives who were responsible for making those decisions. That alone was able to save 7,500 hours per year. Huge lift for the business. Exceeded all time saving expectations because little did we know that during Covid there would be more credit orders than ever before. But no one has ever thought about automating the decision making step. And that's exactly what I did with Al. Cause I built a script that can actually read all the different criteria for releasing a credit order and then it will output the confidence interval that we had for that order being credit.

Andreas Welsch: 4:31

Fantastic. Hey, so it, it really sounds like you've done quite a lot in that space and also bringing people to that topic of AI and automation who might not necessarily have been familiar with it in the first place by way of being finance professionals and experts of that process. So maybe quick shout out to the audience. If you're just joining the stream, drop a comment maybe where you are on your journey and what questions you might have for Emmanuel. Maybe one question to start with. I'm really curious in that context of what you shared here doing this in finance, what would you say is one of the things that people shouldn't be afraid of doing?

Emmanuel Lai: 5:17

I know this sounds very cliche. That is what I have always told everyone that I've mentored in the past and present. And that's what I will tell people in the future. You should not be afraid to fail. It is better. It is better to fail quick. And discover that your idea isn't valid in its current shape, or form. Because oftentimes ideas that don't work today doesn't mean that they won't work tomorrow or, a few months from now with a little bit of tweaking. It is better to have a working prototype up sooner or later. It does not have to be perfect. Perfectionism in digital transformation will not get you very far. Because perfectionism oftentimes is a deterrent to getting stuff done. And I'm a huge fan of perfectionism. I hold myself and the people I work with to very high standards when it comes to a final product. But getting that POC done, the fear of failures should not deter you, number one. The second thing that people need to keep in mind with artificial intelligence. Another tip that I would give to any leaders who want to implement that in their organization is that you have to get the buy-in of operations. One common concern that I see in many organizations is that the people who are performing the work on a day-to-day basis, the operational managers, the operational executives, and the digital transformation leaders and executives, there's oftentimes a disconnect between the goals of operations and the goals of digital transformation. For a initiative to succeed, you need the buy-in from both parties and their goals need to be adequately aligned so number one, make sure you fail. Make sure you fail fast, move quick. Break things. Number two, make sure that you can convey and articulate the value of artificial intelligence to operations, management and executives. There's no need to do AI for the sake of saying, hey, we did AI. We need to make sure that it fits in a integrated, holistic transformation.

Andreas Welsch: 7:29

Fantastic. I think that sums up very nicely what we see and what we hear from others as well that have successfully rolled out these kinds of capabilities. Now I remember you said you specifically chose the credit type use case. I'm curious, what was the reason for choosing it in the first place? How did you make that decision and what were some of the factors that have influenced that decision?

Emmanuel Lai: 7:59

Great question. This actually leads to the background of some previous transformation initiatives that I've led. One of my biggest wins that year was actually putting all and consolidating all the credit information into one unified platform. And the reason why we built that with RPA was that the cool thing about RPA is that RPA is very rarely the perfect solution. It's very rarely the optimal, clean, perfectly optimized solution. And that goes into my philosophy of perfectionism is not always the answer, but RPA did a very adequate job of getting people the information and lifting 7,500 hours per year. And when I saw that automation at the end of the quarter, I did some analysis and I saw, hey, this is lifting a big amount of the non-value added work. But how can we potentially expand this automation something that's delivering 7,500 hours per year worth of savings. I'm not satisfied with that. Yes, that's a lot of hours, but I'm always striving to not just hit the targets, but exceed the targets as well. I did a deep dive of what kind of metadata exists and was being generated by my automation, and that gave me inspiration. I looked at the metadata of the RPA solution and I saw that one thing that we were always missing, which was history of whether or not these credit orders were held or released, that was all being captured by my RPA solution. Because it enabled analytics and tracking this actually served as the basis for my machine learning model because all that data was being generated by the RPA. Now I had sufficient data to train the model and teach it. What are some traits of a order that's credit worthy? What are some traits of orders that were not credit worthy? Without this automation being in place, we wouldn't be able to easily extract all that data. And because we did the work, we already planted the seeds previously with the order release automation, getting the machine learning part stood up, was actually a walk in the.

Andreas Welsch: 10:29

That sounds fantastic. Especially when we typically talk about AI it's, hey, it's custom projects. It takes a long time. There's a lot of risk here. And you say it's a walk in the park. I think that's very encouraging and to me. It's also great to see how you frame that problem more like an evolution or like a step-wise problem. Automate first with RPA. Collect more data. Have some hard facts and evidence and then see how you can take that with AI and elevate it even.

Emmanuel Lai: 10:59

As I said earlier, getting consensus, getting buy-in that is sometimes more difficult than actually deploying the solution itself. I saw someone posted a question. Have you any tips on what format or presentation you have used to best communicate the value of AI to the C-suite spreadsheets, peer-to-peer discussions where you mostly listen? That's a really good question, and can I just take a minute or so to answer it?

Andreas Welsch: 11:25

Absolutely, that'd be perfect.

Emmanuel Lai: 11:27

Cool. So I have participated in many of those presentations before. And oftentimes what the C-suite wants to see or, what the executive leaders want to see is they want to the data. We're not just doing AI for the sake of doing AI. We're doing it because there's a tangible dollar value that can be returned to the business. There's a potential reduction of risk. There's a potential improvement of the customer experience. So when I made that presentation, I outlined number one, anytime you can potentially save, I had a rough estimate of how much time people were spending to run these calculations, run these reviews. But the main save is actually enhancing the employee experience. And also having a extra layer of controls bec because oftentimes we, we think that humans are really good at checking things. That's actually not true. Sometimes your worker didn't have enough coffee for the day. Maybe he could've fallen asleep that night. Maybe he's a little bit stretched out because his commute to the office took a little bit longer than usual, got stuck in bad traffic, etc. Whereas a automated mechanism of checking always runs the same exact way, and having two layers of checking the human and the machine makes for better controlled than just one layer.

Andreas Welsch: 12:56

Fantastic. Great to hear you talk about the two working in tandem. Have automation take care of the repetitive stuff but still have a check with the human in the loop as well to make sure nothing is missed.

Emmanuel Lai: 13:12

And I quantify all this in a framework that I call the three E's. I go from most quantitative to a little bit more qualitative. And that's not saying that the qualitative side is any less important. It's very important. However, when I present to senior leadership, I always want to stress the more quantitative items first. And then walk over the more qualitative items after. Efficiency is how much time we're able to recover for the business. That's really easy to quantify into a dollar value. So let's, so assuming you have a, assuming you have a accounting or finance specialist, you value their time as at around 40,$50 an hour. Let's just say$50 an hour. So it's easy to round. You recover 200 hours for the business. Now you have saved yourself 200 hours times$50 an hour. That would be$10,000. So efficiency, really easy to quantify. Effectiveness is your ability to hit a objective and key results or a key performance indicator. Also not as hard dollars as efficiency, but still very tangible. I can point at a KPI that I'm improving. Thirdly, it would be the user experience. How draining is it for the user to perform this task? There's a lots of tasks that are very rote, very mundane. It might not return your money today, but it will return you money to your organization through better employee morale, better motivated employees and associates, and then lastly, less turn turnover as well. I think this is always my philosophy as a leader. If I can make the jobs of people who work for me a little bit more enjoyable. And I can make them enjoy going to work a little bit more every day, then that's how you lower turnover. And as Andreas, I'm sure you've dealt with it before as a executive.

Andreas Welsch: 15:17

Absolutely. And great way to frame it. On one hand, the personal impact that one can make as a leader, bringing more more exciting parts of of the work actually to the employees. But at the same time also connecting that closely with the business goals. So fantastic. I really love the three E's. I think it makes it very tangible. Alright, so I see where we're getting close to the end of the show. So maybe let's just summarize the three key points that I heard you say. One was really around taking steps in your automation program. Start with RPA. Get the hard facts. Look for additional automation. I think the second one that I heard you say was bring those facts then also to your executives to show what kind of value you add. And thirdly, frame it around the three E's to make it tangible, considering the efficiency all the way to the experience and the impact you can have.

Emmanuel Lai: 16:25

Absolutely. The last point I want to make is that if AI is not the solution, if RPA is not the solution. Oftentimes, I view RPA and AI as a solution of last resorts. And you do it only after you fully optimize your process. Only after you're able to exhaust the kinds of savings that you don't need to spend money on, you don't need to spend effort to do. If you can fix a AI use case or RPA use case with a process related fix, that's gonna make your senior leaders really happy, cuz you're spending$0 versus some dollars. So ultimately try to solve the problem without technology. And then if you need technology to solve it, you can't solve the problem with human behavioral changes, then let's talk technology.

Andreas Welsch: 17:21

I think that's a perfect note to end on. Fix the problem first and then see how technology can help and not the other way around. That's beautiful.

Emmanuel Lai: 17:30

Excellent. Andreas thank you very much for having me on this presentation today. It's been an absolute honor and I really look forward to working with you again in the future.

Andreas Welsch: 17:40

Perfect. Thank you so much for jumping on. Really appreciate you sharing the learnings with the community.