The rise of the robots

Nabila El Sadek [N.S]: I'd like to start with the title of our panel, "The Rise of the Robots and What's Next in Tech." 

John, you're the CEO of Contact. When you hear this phrase, what's the first major opportunity or challenge that comes to mind for your industry?

John Saad [J.S]: The first thing that comes to mind is the myth that robots and AI will replace humans. So every time I think of that or I hear that, it puts a smile on my face. Because I believe exactly the opposite, I think AI — even as Ahmed [El Alfi] was saying in a previous speech [Where (and how) will we live & work in 2035] — is purely an enabler to make things faster, more efficient, more productive for humans. So when I hear that, I feel like: one, the myth is not true; two, I really look  forward to how we can adopt AI to really make our lives much better than today, whether in business or outside business.

N.S: The contrast between opportunity and challenges is a perfect starting point for today's conversation. 

Ahmed, at MNT-Halan, you're building technology for millions of unbanked Egyptians. Does AI represent a chance for radical financial inclusion, or does it risk creating a new kind of digital divide for your user base?

Ahmed Mohsen [A.M]: I think so far we've been very excited because it presents a radical change in the way we do financial inclusion. I think we're building products that we didn't think it was possible, or it would have taken us a very long time to build. Our risk and credit models are constantly becoming much better, and that's a great thing for financial inclusion I think.

N.S: Youssef, your company Wondercraft, based in London, is the rise of the robots. From your unique vantage point, what is the single biggest and perhaps most dangerous misconception that business leaders have about how GenAI will transform their industries?

Youssef Rizk [Y.R]: The biggest one is that it will do your job for you. I think, touching on what John just said, that we're all going to lose our jobs because AI will do it for you. If you try to ask anything moderately complex of ChatGPT that you know about, then you'll notice it does not give you the right answer. And so I think if you are a business hoping to integrate AI with the hope that you can, I don't know, instead of needing 10 people to do X, you now need two... I think that's true. It is a great amplifier, but not one that will replace, but one that will amplify rather.

At least as AI started becoming more prevalent, the conception has always been: give it the input, it gives you the full output. And if that's your paradigm shift, you're likely going to be unsuccessful. We’ll mention the MIT report that they did, right? 95% of AI pilots in big companies don't work. And I think the reason is fundamentally, you cannot… you need to have some ability to edit or refine the output afterwards.

N.S: That's  . Moving from the high-level view to the practical realities of implementation. 

Ahmed, let's start with the ‘build versus buy’ dilemma. How do you decide when to develop a proprietary solution versus when to partner with a tech provider?

A.M: We build our AI solutions when it defines our [economic] moat, or our advantage in the market, especially when it comes to credit and risk models. When the AI solutions are more commoditized, or it's much faster to actually buy it to go to market, we definitely go to that. There's no ego involved. We always like to go for the most practical option, unless it's something that is significantly becoming an advantage inside the company. That's when we try to keep it internally.

N.S: So speed plays a big role in your decision.

A.M: Definitely, speed always plays a great role.

N.S: Youssef, as a tech provider yourself, what's your take on this? When is it truly make sense for a company to build a solution in-house versus to give you a call?

Y.R: Great question. Obviously, as a tech provider, my preference is that everyone buys. Don't build yourself. But the reality is I find myself in two minds about this because obviously I am a tech provider, but I myself have other providers that we use. And we think a lot about whether we should essentially build what they're building or not. And I think what Ahmed just said makes a lot of sense, which is, it fundamentally comes down to what are you optimizing for?

If you are optimizing for speed and you're not super cost-sensitive at least for the proof of concept phase, don't build. Buy. Right? Someone has spent a lot of time figuring out how to do this, probably better than you can, at least with the initial kind of on-ramp period. And so it's 99% always going to be better for you, at least when you're in that POC phase.

After that point, when you're trying to scale, as in you're trying to get better control of your cost… I think anyone who's built anything agentic will tell you the costs are huge. And so you can't just rely on everything out there, because they kind of add their cost and then they add a margin. And so at that point it starts to become kind of tough. 

The one other thing I'll say is, anyone who's building on AI will notice a bit of a "rug pull" in a way, in that the floor shifts under you almost every month. There's a new model that comes out, suddenly that's the best thing that you should do. And so unless you have the capacity in terms of manpower to spare to actually dedicate to the maintenance of what you're building, you should buy. They have a full team dedicated to maintaining it.

N.S: Speed, efficiency, and expertise, of course, play into this question. Whether you build or buy, it all runs on one thing: data. 

John, data represents both internal and external challenges. Internally, you face the hurdles of data quality, quantity, and security. Externally, you must navigate a complex regulatory landscape, particularly around issues of data sovereignty. For a regulated institution like Contact, which of these presents the bigger obstacle for innovation, and how do you build a data strategy that solves for both the technical and regulatory realities?

J.S: I think the key challenge here is the data quality and structuring the data, and having the data ready to be used by AI. I think that's the key challenge. There are always solutions to solving the rest, with how you host the data, how you protect the data. But then the part around initially having good quality data has always been an issue in Egypt. Because digital data in the past has been very scarce. Recently, with the uptake of financial inclusion in Egypt, with the uptake of mobile wallets, for example, there has been much more access to data that is much more structured than before. So I think having a source of data that is much more structured is a breath of fresh air for the future.

I think the challenge in Egypt is a country-wide challenge. I think it is moving towards the right direction, especially with the financial inclusion progression that's happening and higher penetration rates across digital financial services. With that, I think the data we can get our hands on and utilize — and therefore be able to give better service to our customers — is going to be significantly better. For example, in the financial sector, in the insurance sector, in the investment sector, all of those. With the progression of digital penetration, we're seeing significantly better access to better quality data.

N.S: And that leads us to a question on returns. As Youssef mentioned, the recent MIT study suggests the vast majority of corporate GenAI pilots have yet to produce measurable returns. 

Ahmed, could you share a specific example of an AI initiative at your company that you took from idea to implementation and that delivered clear, measurable returns on investment?

A.M: Sure. I think the biggest example is we enhanced the platform to gather a lot more behavioral data, and then we used AI models to actually analyze this behavioral data and better target our customers and give them better credit risk scores.

And over the past year, this experiment has led to 3x —  even more than 3x —  in our approval rate. Our risk has never been better with the current risk models. So I think this has definitely provided a lot of return, this experiment specifically. But it took us a very long time from thinking about the product, thinking about the data, thinking about how we're going to use this data, and then experimenting with the models enough to actually reach a point that it delivers such results.

N.S: In terms of a timeline, how long would you say it takes to see a return on your investment?

A.M: This project took around a year to implement, and then we saw the results five or six months in.

N.S: Staying on the topic of investment, let's zoom out to the national level. 

John, Egypt's ICT strategy has traditionally focused on assets like call centers, which will likely be disrupted by AI. How should our national FDI strategy change, and what kind of industries should we be trying to attract now? What's your view on the timeline for this disruption?

J.S: I think the timeline… What's delaying the timeline of the disruption is probably the failure that Youssef was talking about earlier of big corporates globally, in terms of how to adopt AI. And I think the focus should always be on not trying to build the smartest AI tools, but the focus is to build the smartest AI users. And I think on that front, Egypt can leapfrog that development.

It's all about users. AI should be in the DNA of how people do things. There are many use cases that could be focused on but I think about having AI in the DNA of the new generation in order to be able to build smart users. And out of that, I think having the right ecosystem will end up generating something that could be meaningful for export.

N.S: Youssef, building on that, if you were given $100 million to invest in one area of AI in Egypt, where would you put the money and why?

Y.R: The way you improve these models is data. There's a lot of advancements in the field, like synthetic data used to train the models and so on. But ultimately, if you want to advance the frontier of the models, you need very bespoke, expert data.

Now, the problem with that is if you're training a model to solve, not even PhD level math, but you know, super complex math problems, the problem is there are not even enough humans in this world who know how to solve this. And so getting that training data is very difficult. Egypt is very interesting in that regard, in that there is a very... I'm just going to call it a "cost arbitrage", in the sense that you have a lot of extremely well-educated domain experts in a variety of fields here, where their cost — let’s just call it cost — is not comparable to what it would be in America or so on.

And so I think there's a huge opportunity here for essentially building kind of a data labeler that you can sell to the OpenAIs, the Googles, the Anthropics. These companies are paying billions for this data because they're essentially the ones pushing the frontier. So all the advancements we're talking about, everything we're building on is fundamentally underpinned by chips and data. And so I think we are probably a bit behind on chips, but we can be there on data.

N.S: We've discussed strategy and execution, but the human element is just as critical, as we keep mentioning. Let's talk about the impact of talent and leadership. 

Youssef, building a company from the ground up, what's the most futuristic or unusual job title or skill set that you're hiring for that simply didn't exist 5 years ago?

Y.R: The more interesting role we hired was a head of content. Now, I think for an AI company, that's a bit strange. People would have expected an engineer. But the reality is the cost of building a product, as in the resources, are essentially going to zero. In a couple of years, you can probably ask an AI to just build the whole thing.

So the head of content, what do they do? I think they're basically someone who is "chronically online" and able to push content on Twitter, LinkedIn, and all of these things.

This is a bit of a cautionary tale on AI. The reason we do this is, we as a company traditionally sell to enterprises a lot, and the traditional ways of getting enterprise leads don't work anymore. In the last five years, probably I can think of at least 15 companies that essentially automate outbound email sequences, which used to be the traditional way of doing it. Even cold calling, right? This is how you would get business back then or leads. 

The reality is I wake up now and I have 20 to 25 cold emails in my inbox every day. I'm not looking at them. And so there's a bit of a cautionary tale here, that AI does saturate certain markets to the point where the new roles that emerge… Talking about how AI will create new jobs, those new jobs tend to be very archaic jobs, if you will, which is someone who actually sits down and figures out what’s a creative storytelling way to get the word out. And basically, now that's the real way of getting leads.

And so AI will, kind of, break certain ways of doing things, but it will give rise to things that I would just argue require someone who can think critically and independently to actually execute these things.

N.S: On that point, Ahmed, given the challenge of finding university graduates with the right mix of critical thinking, as Youssef mentioned, and the ability to produce work using AI as a tool, how are you as an organization building the talent internally?

A.M: So, internally, we had to get a consensus on how we introduce AI into our workflow, and that's easier said than done. You would think that AI will just "enhance the way you work and it will increase your productivity," and that's definitely the answer when you're starting something from scratch. But when you're working with an existing code base or existing solutions, when you come to introduce AI in that workflow, it’s very challenging and has a steep learning curve.

Ideally, you would get a 30-40% productivity increase, but not the 80-90% we're talking about. Because the models now with the context length, you get diminishing returns. The more you ask of the model to do edits over time, especially when it's working with an existing solution or code base. You want to introduce it in a way where you build your own custom AI workflow, and this is something that we're doing internally. So everyone can plug in, start working with the tools, and start being more productive. I think that’s the main thing we are pushing for. 

N.S: Are your talent needs met in the market at this time?

A.M: Yes. It's a bit hard, but there are a lot of people who are talented in Egypt. Yeah. I think we're having no problems with that.

N.S: Excellent. 

John, how does leadership need to change in view of this new environment? What's the difference between leading a team of people versus a team of people who are empowered with AI tools?

J.S: I think a lot of companies today globally have not found the answer to this question.

Sometimes I laugh when I hear about a success story of a company being able to use AI to produce a smart chatbot. For me, that's in Arabic, I’ll have to say, “Tamahhada al-fiyl fa-walada faran” [Arabic idiom] — I think that we can do much more with AI than just something like smart chatbots. There’s much more to be done. 

Now how AI is developing, especially in the last 12 months, is beyond anyone's imagination. I don't think even if I ask people like Youssef in the field what's going to happen next month, I don't think anyone can predict. And I think we are getting surprised every month with the evolution of large language models, of what’s going with generative AI.

Hence, I think that evolution is much faster than what people or big corporations are currently using or adopting. So there's a catch-up game that's happening every day. Therefore, I think  — in my view, and my view stands today, it could change tomorrow — the best way to do it is that AI, or using AI becomes something in the DNA of the organization.

So the question as a leader is, how can you encourage the organization to embed AI into that DNA? If you think of what AI can do across all domains, and across different functions in any organization, I think it can help everywhere. And hence it's all about creating the right use cases, jumping on them, making sure they're being adopted right now. And that cycle is an ongoing cycle for the time being, until we see where all of this evolves.

N.S: Question for the panel. What is the single most durable skill, technical or otherwise, that a young person in Egypt should be developing right now in order to thrive 10 years from today?

A.M: I think it would depend on the profession. If I would say a general answer, I would say ask better questions, because I think that would be the gateway in the next coming part. But I think it really depends on the profession. Like, if it's an engineer, I would give different advice than someone else.

N.S: John? 

J.S: In my view, the winning formula is to be a subject matter expert somewhere, with product skills and AI knowledge. I think the trio, the three of those, is probably where the future is going. How can you actually form a career or experience that way? I don't know. But I think that trio is a magic formula that I've seen work very well at this point.

N.S: Youssef, what's your view?

Y.R: I think you have to put in a tremendous amount of effort now to maintain your ability to think.

I think with the rise of ChatGPT, there's a very fine line between using it as a replacement for a textbook — "Hey, how do I do this?"— whether you're a programmer or…. Actually, I don't know if I agree with the subject matter expert thing, because in theory, that knowledge can be then on ChatGPT, you can learn it and so on.

But there's a very fine line between "show me how to do this thing" and "do it for me," which is all the agentic stuff. And the more you lean on the latter... I found myself doing it, right? I will ask the thing to code a new feature for me, mostly because I care about the speed and efficiency of how fast we ship. But the reality is that you do notice your ability to think about these things independently decreases. But the reality is that that's really the most important thing in a future where AI does everything at the execution layer.

N.S: Looking to the future, what is one area where you believe collaboration among Egyptian tech companies will benefit the entire ecosystem?

A.M: I think creating better Arabic models. That's one thing. Also maybe collaborating on having anonymized data sets — local data sets — to enhance different models. It's very easy to say it. It's very hard to get the data out anonymized. I don't think it's an easy thing to do, but I think it's worth a try.

N.S: Interesting. This has been an incredibly insightful discussion. To conclude, for the business leaders in the audience today who are feeling behind, what is the single first step that they should take tomorrow morning? Ahmed, from a tech perspective?

A.M: From a tech perspective, I think I mentioned that before: introduce AI into the workflow in a correct way. Because if it's not introduced in a correct way, it can get actually very bad, almost most of the time.

And revisit all the engineering practices, because there are things today that you can afford to do faster and then fix certain scalability issues later because it's cheaper. So you need to rethink the way you think about engineering and introducing AI into the engineering workflow. That's number one.

Number two, I think introducing AI in your products is something that is non-negotiable anymore, whether it's thinking about the product or in the ideation or in the gap analysis phase, or actually introducing AI features in your product. We're going to see that more and more, and I don't think it's a luxury anymore.

N.S: John, from a business strategy point of view?

J.S: Again, I repeat the same thing. I think being in the DNA of every single person in the organization is a very important thing. Choosing a use case across every domain in the organization — even if it's one small use case, but making it a success story so everyone can rally behind and believe in being able to create more and build more. I think it is very important, and I think it should cut across every single unit in the organization.

N.S: Youssef, from an innovation standpoint, what moves the needle?

Y.R: Everyone's thinking of AI transformation as, "I'm going to build an agent that does this," or whatever the new paradigm is that people are thinking of, now it's agents. I think the thing to start with from an innovation perspective is not to focus on, "let me boil the ocean at once" and rebuild my entire business, but rather focus on a small bit of the product or the business or whatever it may be that you can automate.

The reality is that the much more scoped a task is, the better AI will do it. The more open-ended — i.e., if you're automating your whole platform — that's where, at least with the current performance level of these models, you'll start to see some hiccups and it'll lead to a lack of adoption and so on. So the smaller that you can scope the initial problem, the better.

N.S: Perfect. Thank you very much.