E13: Fernando Gamallo on AI

E13 Fernando Gamallo on AI

Ernst: [00:00:00] AI specialist Fernando Gamallo, shares how businesses of any size can use AI to streamline your business. This is episode 13 of the Tech Leaders Talk show

On the Tech leaders talk show. We chat with technology leaders. Who are experts in the field? Our aim is to help businesses understand and adopt technology to solve business challenges. Plus join our free Facebook group to join the ongoing technology discussion. Share your ideas, help others and post questions.

Search for tech leaders. Talk on Facebook and I'll see you there. So we have a fantastic show for you guys today. Fernanda was most recently featured in the Forbes magazine for his work and AI. Fernando helps us understand how businesses can use AI and how to get the leadership team on board. At the end of the show, we introduce our new competition.

So stick around, I start [00:01:00] to show, but asking Fernando how he's got started AI and why specifically he chose AI. So Fernando, how did you get involved in AI? Firstly, because you come from an infrastructure background, so there's quite a big shift to get involved with the AI. Can you tell us a little bit about that story?

Fernando Gamallo: [00:01:22] Oh, great. Interesting story. I think that the very beginning, as you already mentioned, I was pretty much interested in infrastructure. But after the few years, I start running business project, small business projects from warehousing and logistics. And after we complete successfully, a migration of a warehouse, from one location to another location, I figured out that there was a lot of data in, in that project.

And we [00:02:00] call the study using. Depth amount of data, you know, the, to understand times of movement of the warehouse. So I pretty much say that that particularly project changed me, change my mind from, from infrastructure to, to data oriented guy.

Ernst: [00:02:15] So, does that mean that you decided, okay, now I'm going to go ahead and start studying AI or was that a case of, okay. I'm going to, use my leadership skills as, and to implement AI projects.

Fernando Gamallo: [00:02:36] In fact, I would like to say that from my perspective, before understanding AI. I start with data and the analysing data, understanding data, getting highlights from data.

And that was my first move when I understood the complexity of data in general terms, [00:03:00] talking in general terms, because for each project, it could be changed dramatically, but in general terms, when you understand what data is and how it can be exploited, how can it be managed? Or enhance sometimes. But that was the very first part from me.

Then I start working with several applications for dashboards like business intelligence applications, or it's more business intelligence application within general terms is to. Gets data and represent perfectly did that. So I think that those are the main, basic steps of AI. When you, when you start doing that after some tries or some small projects, you start seeing that the data can be manipulated, that it can be transformed into patterns or into formula.

[00:04:00] Or into algorithm . And with that information, you start working with the very first steps of AI. So if I've made myself clear my transformation from a traditional IT guy to a AI guy goes through the use of data, that understanding patterns then. Business intelligence and then predictions, algorithms, and then more sophisticated tools  to predict, to understand or to learn machine learning, deep learning and that stuff from AI.

Did I make myself clear?

Ernst: [00:04:44] Yes,  you absolutely have. So you just mentioned about those different phases. And I remember when we chatted before the show, you talked about different phases before you get to AI. Can you talk a little bit through those phases and you know exactly [00:05:00] why those phases are important?

Okay. Well, it, it, it really depends because I would like to say that, first of all, what is intelligence from a human perspective? Intelligence is learning or understanding and using that knowledge to do something or to achieve some goals or to achieve something in general terms that is intelligence from a human perspective.

And from a computer perspective, AI is pretty much the same. We have to learn, understand and use exactly that you know that to achieve them. A very specific task. So in order to develop an AI project, I do recommend to, first of all, learn what kind of learning you need to understand prediction models, which are the most [00:06:00] basic ones we can use, even Excel to do some kind of prediction models.

And with that prediction models. And with that data, you are manipulating, you can learn. Then probably the second step is understanding, which I could make an analogy of, of machine learning, which is changing the rules of programming. You know, that you take the answers and the data and the programming is keeled.

Everything makes it, you know, that to generate automated models. That should be the second, the second step. So the first one is to do some predictions. The second one is to do some machine learnings and automated model to learn itself. And at the end, this suggests that probably AI or complex AI or advance AI solutions could be.

So I believe [00:07:00] these are in very general terms, three normal steps to go through the AI journey.

 Okay. So I think perhaps a, some examples would be really good to , cement that information you know, can you give us some examples that you've worked with in the past or common examples that's that are used

Fernando Gamallo: [00:07:25] For sure. Let's see. Let's do the following for example. I know that the predator sticker or a small prediction model, the budget, the standard business process, which is budgeting and you take probably results of one year.

You have a budget and you create some. Cool graphics in ex Microsoft Excel for instance, but you can, there are several tools, some very complex tools, some very simple tools, MATLAB [00:08:00] or, or rapid miner or those tools can be really easy to use in order not to discuss complex programming languages. You can do some projections using you budgets.

So we use budgeting for budgeting purposes. We use a lot of projections, standard predictions, which are normal curves. From a statistically, it means which is a standard deviation or which is a linear regression. And you make a linear regression of a trend and you can figure out what it's going to happen.

That is one example, another example for machine learning, which is a little bit more complex. So imagine that you have your sales. Or on a quarterly basis, for instance, the last year, 2019, and this year 2020, and you have your sales on a quarterly basis. So you put your sales on an [00:09:00] algorithm and every day or every month you have new sales, you put those new sales into that algorithm and that algorithm is recalculated or rerun itself.

And change internally, the algorithm change the output, including new values. So there is machine learning it's stopped learning and learning from each new input. It uses the standard information previously. plus the new inputs and recalculate the whole process and give you another prediction. Now it gives you another proposition of solution, including new values.

So this is pretty much, machine learning. I don't know if I made myself clear.

Ernst: [00:09:54] Yes, that is, that is very clear. Okay. So a small, let's say a smallish business, [00:10:00] right? Or once who wants to be able to let's focus on the sales for instance, because it's obviously quite a common use case that small business wants to start implementing AI to get better predictions out of there.

As sales forecast, how would a organization start that? No, cause the first thing I'm thinking. There seems to be quite different elements. There's a data element. There's a programming element. How would you build a team and how would you kick off a project to start that journey?

Fernando Gamallo: [00:10:30] Okay. But a good question.

I think that two, two very important pillars, the first one is a very technical pillar, which is the data let me focus before on the second pillar, the second pillar is you need to identify your problem. Clearly, identify your problem and write it down and define the scope of the problem is very, very typical that in several AI project, that the [00:11:00] scope is so ambiguous that.

You can lost your project since the very first day, because it's not narrow objective. So the first advice is to identify clearly the problem second is to prioritize the value you are aiming to or to achieve from that point. So this is the second pillar is to understand what you're trying to solve specifically.

What I mean specifically is very very specifically. And the first pillar with disability technical pillar is in order to solve with AI a specific problem. You need data. So if you do have your data, you can solve a problem. If the data is missing or you need, or it's not complete data, you need  to create data.

And that make mistakes. [00:12:00] So the two basic elements you need in order to solve any problem in a business is to identify the problem. And then with the problem, identify, identify if the data you have will solve the problem. Okay. Hmm,

after that, after you have that, you can start building several prototypes of the solutions. And let me, let me tell you something. You can find several solutions to the same problem from different perspectives too. That is why I say you need to provide value because you can see that sometimes the answers.

Could, give you more highlights or more insights that you were expecting, and then you cannot handle that amount of answers. I don't know if I'm [00:13:00] making any sense.

Ernst: [00:13:01] Yes. That's a very interesting point that actually 

Fernando Gamallo: [00:13:04] I don't know if that answers the question of, do you want me to go with a little bit deeper.

Ernst: [00:13:08] It does answer the question. I think. So the first thing I heard . Firstly, we need to identify what is a problem we were trying to solve. So in the, let's say talk again in a sales context, the context yeah.

Is for my business, I want more accurate sales predictions. And well, hopefully. Be able to identify where I should focus my business on, for example, then the next is figuring out what data do you actually have available? Make sure that data is accurate. And then we start kicking off the project and doing some prototypes for that project.

But I also heard that you need probably two different types of. Teams right. One is a data person and another person would probably be your around your programming. Is that an accurate summary?

Fernando Gamallo: [00:13:58] Yes. Terrific. That [00:14:00] is great. You need somebody from, from data, which is in some cases that data could be a very specific data guy, but sometimes it's more related to the business that understand the value of data. From the business perspective. I always say that this is the guy that speaks business. So that $2,300, what does it mean for you? For me, probably nothing, but from a business guy that means particularly.

As a piece of answer of a question is asking, so this is what I'd say from, from the data perspective, we did the data plus business perspective, and the other part is a technical guys are the very advance and in programming skills that can handle that information, that data can manipulate it [00:15:00] using standard algorithms,  or complex algorithm. That depends on what kind of technology you are choosing to use.

Ernst: [00:15:09] Okay. So you could be looking at the various different types of programming guys. So either that, for instance, Python, what are the other common ones that get used?

Fernando Gamallo: [00:15:18] Well as I said, you can use a C plus you can use Python.

You can use R you can use MATLAB. You can use Rapidminer. I don't know if I say that. But also there's some business intelligence solutions that have embedded some small programming tools. You do do not lean to learn those, but for instance ClixSense or Tableau for instance, or even more complex like IBM solutions.

They have their own numbers. AWS. Amazon has also some solutions online, Google as well that you can start using [00:16:00] specific algorithms from their web services. In general. What I'm saying is from the very beginning from understanding the project. The problem, defining the team. You need somebody with AI expertise, but it doesn't need to be a very high qualified guy.

You need somebody to understand what is AI? What is that like and how to manipulate that when you probe is going on and growing and getting a little bit more mature, then you need to incorporate more complex teams. That in some cases depends on the information you are using. If it's confidential information, it's better that you hired those resources internally, but if it's not strictly confidential and can be shared with accordion legal contracts, [00:17:00] signed, then you can hire services, which is a little bit  more cheaper.

To, to hire some services per hour, per day, per week, or per project that having one guy turned on, it depends on the project or the complex of the project. But from the very beginning, I said, you need some guys with technical skills. Not necessarily. High qualified engineering programming and, and a master in data scientist, for instance. Yeah.

Ernst: [00:17:36] Okay. So you've decided on a a problem you're trying to solve, right? You have a idea of what type of team you need to build to solve that problem. What about if you've got a board that's not necessarily. Well, excuse the pun, but not odd board, right. Is convincing the board too, that you need to do this type of work.

What type of challenges have you hit in the past and how do you [00:18:00] address those  challenges?

Fernando Gamallo: [00:18:02] Okay. Interesting question. I think the most important answer the board would like to hear or to see is. How in terms of return on investment or increase in sales or increasing efficiency, the project will contribute.

So the way we can measure that, that's why at the beginning we were saying, we have to prioritize the value,  in a way that we can measure a specific return on investment or increase in sales or direct efficiency reduced costs or, or whatever KPI financial KPIs you would like to use is the fastest way to get an approval from, a board perspective.

For instance, I could talk about three years more projects, very small projects we ran [00:19:00] a couple of years ago. One is RPA robotic process automation, robotic process automation is a very simple and efficient way to start working with my finger and processes. Not creating anything beyond that is what ready software already created.

For instance, you can have on a monthly basis, certain of, of your organizations like sales or like finance or like corporations, it depends on each organization. It's analyzing a lot of information and it's very boring, but they have to do it every day on a normal basis in order to get one or two outcomes that can be automated.

Very simple with a robotic process automation. When a process is a continuous process, it follows a pattern. It can be automated using [00:20:00] RPA. This is a very simple, let's say machine learning protocol process from  AI that can be implemented is very cheap and you have huge return on investments because sometimes you identify that several members of a team organization are investing a lot of time working on something that can be automated.

Very simple, but it's a fact there is a small, a small piece of product that can be easily sailed to, to a board another one is for instance my maintenance, maintenance, repair, and maintenance predictions. If you have a fleet or if you have lots of boats or cars or, or, or trucks in your business, for instance, You kind of start loading a lot of information into a prediction system.

There is [00:21:00] simple prediction system is not going to be very complicated and you can predict when it's going to happen a failure. The problem is not to attend the fray. The failure. When you quantify how much sales you are losing because of fails. Because of not delivering to products, to customers or services to customers.

When you quantify the amounts of losses you have because of failures and you can predict failures, then I'll automatically, you can add those sales to you, to your income statement. Okay. So these are kind of very simple ways to express. Or to sell an AI project to, to a board from a business perspective.

Ernst: [00:21:51] And if I'm reading between the lines there is, is find those quick wins, which don't require a two year project [00:22:00] sell that quick win to the board demonstrate what the solutions can do. And then then start talking about your bigger projects and your bigger wins.

Fernando Gamallo: [00:22:11] Yes, correct.

Okay. So coming back to the RPA, I'm very interested about that.

So,  why is an RPA project so simpler than the traditional AI project, if you want to call it that? Okay. I would like to say that RPA projects are. The translation of routine airy activity or set of activities, because it could be a set of huge activities into a single pattern that can be automated by a computer.

For instance, examples , of quick, quick wins, as you said, RP A call center. In a call [00:23:00] center, you receive, let's talk about general numbers from 100% of phone calls you receive 17. The present are related. Let's say we do use sort of password for instance, or wave ballance  enquiry or very routine  question that customers would like to answer on a very fast way. And the huge proportion of the calls are related to that. Let's say this example, 70% are related to using a password change because I miss my user password. So what you are going to do is to automate that activity. So. If you were a customer Coleen and you say, okay, if you're calling for a user password reset, please press well and you press one.

Okay. Please give me you is just an example. Give me you r user ID. Okay. I will. You will receive by email your password at this very moment, please confer. Thank you. And then it's a very [00:24:00] simple way to implement RPA. Internally. IPA has a complex programming  procedures to do that. But those programs had already developed.

So it can be very, very easily to customize by almost everybody. So it's not complex, let's say , another sample of RPA. You have you guys from manufacturing, let's say somebody from manufacturing as the planning programs and they work those planning programs in Excel and send it to the sales guy, sales guys, review those plannings and put the forecast in sales and item by item. Probably there are 100 items, so they have to do in one by one. And then send it to finance and finance will calculate or recalculate costs, margins, and some financial taxes and stuff, item by item [00:25:00] as well. And then send it back to manufacturing that they can review and adopt as well. Standard planning task for the month. For instance, you can now automate that using RPA. So it's going to be very fast reduce errors in 99%, for instance, normal human errors and make it very quickly to solve. And it's not going to be complex. That is another example.

Ernst: [00:25:28] That's very interesting that, cause it does, it does highlight that you really need somebody who understands, in that last example. Somebody who has to understand the process from the sales person all the way through to the finance person, which I think positions a for instance, a CIO who understands those data flows in a really good position to execute on a project.

Fernando Gamallo: [00:25:53] Yeah, that is fully agree fully, that there is a very smart way of [00:26:00] saying that is you understand the process.

Do you make , a flow of the process? And then you can see what, what can be able to make the work can be automated? How can it be automated? And then you can implement pieces of RPA in the whole process. Yeah.

Ernst: [00:26:17] Interesting. Okay. We, when we were chatting before as well, you, you mentioned there's a few things that's really driving the adoption of AI.

Fernando Gamallo: [00:26:28] What's your thoughts around that? Well, it's, it's quite complex because

it depends on the organization. Depends on the business impact. Depends on how you will like to achieve. I should've done we're living right now, a very complex CSUMB let's I was looking for the proper words to say this pandemic situation, but everybody's saying for them, you can, and it's [00:27:00] so nasty to say those all the time, but say we have important something that is called data.

Delivery and decision. These are the three D's in marketing that a deliberate decision to take advantage of these cloud economics. We are living right now, or it's exploded right now. This, cloud economics or this economic. Oh, using the internet or the cloud, or all the E services for saying in these particular words, we need to understand that that decision of delivery and AI projects or machine learning projects or whatever technology you are using in advance processing of data is going to allow your business to enhance, to make it faster and, to react, there are a lot [00:28:00] of small businesses boarding right now. Some, some great entrepreneurs there are having some ideas and saying, okay, imagine that we can do that. We can do this using this information because information is already there. We live in an information world.

So if you, I said again, the three DS data delivery of decision, if you can get better, if you can think and try to deliberate something with the data delivery, I mean, not only products, but also services, but also information. And with that take decisions, your business is going to run quite faster. That is running out.

So in general terms, not because of the pandemic we are living right now these three days could be a very good way to, to adopt AI in any business.

Ernst: [00:28:55] Interesting. Okay. What do [00:29:00] you think is next for AI ? You touched on RPA. Is there anything else that comes to mind

Fernando Gamallo: [00:29:05] From the perspective of examples for what is going to come next? Or where do you, where do you think AI is going?

Ernst: [00:29:13] Is there any sort of particular thing that you think is going to be almost like the next kind of important step in the AI? You know, whether it's technology or adoption or anything like that.

Fernando Gamallo: [00:29:24] I'm thinking that Jan said, because I, would like to share something from my perspective, being very honest from my perspective, AI is overestimated. And I would like to say why, first of all, I love technology, I'm a technology guy. And I love technology. The second I, my daily work, my daily job related to AI, but everybody, when somebody said artificial intelligence are the future, there is the answer is yes, [00:30:00] but we need to understand perfectly what we are aiming to achieve with AI.

From business perspective and being very honest, only from a business perspective, we don't need to invest huge tons of dollars or whatever currency you would like to use in order to implement a project. There are several solutions that can be adopted from the automation perspective that could transform your business. I would like to say that at the very beginning, for instance, and respect with, with trademarks, but Amazon or Netflix, they are using very simple AR models, AI models to, to suggest what to buy or what movies to see. So [00:31:00] it's not necessarily to invest in complex programs and, and data scientists to understand if you can act, if it, if it can measure the scope of your project, then you can really identify the needs.

Now, not talking about business, talking about why do you say AI, what's going to happen every day. We're going to have more AI solutions around us. From a very simple Alexa or Cortana or c ri we will have more AI solutions surround us today. You can turn on your lights or turn off your lights. But if you days you can have a whole eco-friendly systems are calm.

That controls everything inside your home, since water, electricity, and heat, and that [00:32:00] stuff using AI or, even. Your cars , or in the health business or in the health world, we can find there is project bringing in some universities in us that they answered really mean printed hard in 3d, in a 3d printer.

So they already created a melt, animated heart. With electronics, plus a processor, a single processor that is programming with AI pieces of AI. Let's say deep learning processes in order to work as your heart, but it's a piece of electronics. So that is what we're going to face. So in 1980, or probably 1990, There was a movie called the Terminator.

Everybody has, [00:33:00] some has watched that movie. We are reaching those times right now, but not from scientists, , or it's schools or academic programs. We we're reaching on a daily lives.

Ernst: [00:33:15] That's fascinating. And it's also, such a important part. Right. You know, obviously the heart, so. Yeah, that's fascinating piece of information. Okay. Final question. If you've got a young technologist that wants to get involved in AI, what's your recommendations? How should they go about, doing that?

You know, if they want to learn that they want to become an expert in AI.

Fernando Gamallo: [00:33:42] Okay. First of all, I suggest to do kind of a data mining. Data understanding there are several courses and trainings regarding data data governance manipulation of data, make it semantic data. I mean, first [00:34:00] of all, they have to understand data, what is data and helps you understand data.

And , the second part is to understand linear programming from a linear perspective, like stand up. Engineering, but topics like regressions, like elder models, these kind of mathematical processes for analyzing data. And you can be surprised how many problems can be solved with the statistics, traditional statistics.

In order to program, you need more complex mathematical experience, but if you are strong in a statistical. Plus data, you can solve a lot of private problems. So I suggest that choose that, understand the data that mining and statistical then basically on what kind of problems they would like to solve.

They need to specialize. It's not the same [00:35:00] learning from imaging AI for, from imaging or AI for Driving cars, self drive cars. It's completely different way of thinking or a kale or c emical solutions, all the different processes and different equations that they, they need to understand.

But in general terms, starting with data and the statistical, they can pretty much understand what is, what is happening with data and how to focus and move on.

Ernst: [00:35:34] Okay. That's interesting. So I'm sort of hearing, okay. Get those two first elements in place, get some experience underneath your belt. And then potentially start deciding to go, where do I want to focus on? But having those two first building blocks in place will help you get your, your first job. And then you can build on top of that.

Fernando Gamallo: [00:35:53] Yeah. In fact, it what's the thinking your question right now. And I said, [00:36:00] I have a black belt, black belt certified lean six Sigma guide. So I got my black belt in , 2010, probably 10 years ago.

But the more complex, the more difficult part of being a black belt is to understand process. To understand data and to statistically find data and create a process that help there is the mid basics of, of AI. Okay,

this can help, but probably somebody could have a very fast track of lean six Sigma. Understanding could pretty much understand if they are willing to continue on that way or not. Yeah, because going deep on programming, they will lose their head, trying to find a purpose. And the purpose is the problem is the [00:37:00] solution.

You're looking for a problem, not learning programming in order to, okay. I already know how to program in R or Python. And what else? Well, you can use marketing. Marketing on digital marketing, for instance is quite complex, different. It's quite different of, of chemical solutions for instance, and both can be solved using AI.

Ernst: [00:37:28] Fascinating. Fernando, thank you very much. I was a fascinating conversation. And yeah, I've well, personally, I've learned a massive amount out of that conversation. Thank you very much.

Fernando Gamallo: [00:37:39] Are welcome.

Ernst: [00:37:42] So here's what grabbed my attention from the show. First, find a core business challenge that you need to improve first.

Otherwise you'll lose your way and spend time and resources. Second, solve a small problem first, so you can build business conf.

[00:38:00] Third, you don't need to invest heavily in developers. At first, they are off the shelf solutions that can help in the early stages. Next stand a chance to win a a hundred dollars gift card by telling us what actions you've taken from the show and implemented in your business.

Let us know in Facebook, on tech leaders, talk all, leave us a voicemail on our website by clicking the little mic on the right-hand side of the next couple of days, we'll announce the next guest by Facebook. So keep an eye out for that. Thank you for joining us.