Subscribe
Share
Share
Embed
The FWDstart Podcast is a weekly show at the intersection of venture capital, startups, and strategic industries shaping the MENA region. Each episode features candid conversations with founders, investors, and operators behind the region’s most ambitious companies, from frontier AI and fintech infrastructure to climate tech, construction, energy, and space.
# Anis Harb, Founder of Algebra AI
*Spotify transcript, lightly edited for clarity and readability.*
**[00:00:00] Anis Harb:**
I reached out to Wil Shu and said, “I’d love you to be an adviser as I build this business in the Middle East.” He replied, “Just come and build it within Deliveroo.” That is how I ended up joining Deliveroo so early, before the company had even decided to enter the Middle East. Once I left, I started speaking to more and more business owners about AI adoption. There was clearly an individual productivity angle, which was very valuable, but adoption across the wider business remained low. I was trying to understand why. What it came down to was that nobody was really handling the workflow logic, helping with the data layer, or thinking through the exceptions and edge cases. We are there to ensure the system works, runs as intended and delivers business results. Historically, as a company grew, it added more headcount, which created more operational complexity and required more process. That was simply considered normal. I really hope we are entering a world where that is no longer the case, and I want to be part of it.
**[00:00:48] FWDstart:**
Where I’d love to start is with your time at Deliveroo. You spent almost ten years there, which is a long time. When you took the plunge and left last year, did you already have a next act in mind, or were you stepping into a blank canvas?
**[00:01:06] Anis Harb:**
It was more of a blank canvas. After I left last May, I wanted to take a breather and think about what came next. It was also a moment when AI was becoming genuinely exciting and really interesting things were beginning to happen.
**[00:01:19] FWDstart:**
Started to actually work.
**[00:01:20] Anis Harb:**
I left Deliveroo with a lot of observations, rather than a fixed idea of what I wanted to do next. The Middle East organisation alone had grown to three or four hundred people. As we grew and the complexity increased, we kept adding headcount. That was the normal way of doing business. Then AI came along, and we started adopting some of it. We began to see that it should be able to solve certain problems and automate legacy processes that consumed a huge number of hours. I also left thinking about agents and their ability to take action for you. Why were we still making decisions three days after something happened, rather than responding in the moment? Once I left, I started speaking to more business owners about AI adoption. Many were seeing gains from individual productivity tools, but adoption across the business remained low. The missing layer was often the workflow logic, the data layer, the exceptions and edge cases, and decisions about when human approval was required. Without someone handling those elements, projects tended to be dropped. We had seen some of that at Deliveroo too. Around the same time, I began speaking with the people who are now our investors at Infinity Constellation, BECO and Silicon Badia. Infinity had been one of the pioneers of the AI-services model in the US, while BECO had published a white paper on AI services.
**[00:03:08] FWDstart:**
So we're reading that.
**[00:03:09] Anis Harb:**
There was a lot happening at the same time. It was the perfect group to bring together and begin fleshing out thesis, particularly what it could mean for the Middle East. We all held hands and jumped through the door after that.
**[00:03:21] FWDstart:**
After that. And were there any sectors in particular in terms of. Those conversations that you're having? Because my understanding is Algebra AI. It's not sector specific. You can go into financial services like does it turn restaurant hospitality. Can you speak to that?
**[00:03:33] Anis Harb:**
We are not sector-specific. We have engagements across financial services, manufacturing and distribution, among others. Those are the more vertical solutions we are building with clients. We also work on cross-functional areas such as back-office operations and FinOps. So the opportunity is broad, but we are narrow in the patterns we focus on. We are not trying to solve completely random problems. We want to find similarities between businesses that can be repeated and transposed.
**[00:04:04] FWDstart:**
Kind of transpose.
**[00:04:05] Anis Harb:**
Exactly to.
**[00:04:07] FWDstart:**
Super wide label and custom work.
**[00:04:08] Anis Harb:**
That makes sense. Something I meant to ask earlier is about the origins of the business. I remember reading that you had considered starting your own food-delivery company around 2014. Am I right?
**[00:04:16] FWDstart:**
As far as I know, yes. How did you find out about that?
**[00:04:33] Anis Harb:**
That's so far I know. How did you how did you how did you find out about that?
**[00:04:36] FWDstart:**
I came across I came across this during my. During my research.
**[00:04:39] Anis Harb:**
Investigative journalism. When I left Amazon, my good friend Wil Shu was building Deliveroo in London. This was in 2014. I thought he would probably be busy with the UK and perhaps Europe, while the Middle East was still outside the company’s plans. I began preparing a deck for a food-delivery business I wanted to launch in the region, or at least use to gauge investor interest. I then reached out to Wil and told him I would love him to be an adviser because the Middle East was not yet in Deliveroo’s wheelhouse. We met, had a great conversation, and he said, “Just come and build this within Deliveroo.” That is how I ended up joining Deliveroo so early, before the company had even planned to enter the Middle East. It was a really special moment, and I was very happy to join Wil and the team. It turned into quite a ride.
**[00:05:39] FWDstart:**
Sorry it's such a tangent, but I just remember it popped into my head just thinking about it yesterday and I was like, oh, that's so interesting.
**[00:05:43] Anis Harb:**
It could have gone very differently. I could have built my own company, and who knows what would have happened. Instead, we brought the idea into Deliveroo, and it became a great story.
**[00:05:52] FWDstart:**
Was doing your own thing part of the reason you eventually left Deliveroo?
**[00:05:57] Anis Harb:**
After ten years, you naturally begin thinking about what comes next. The region had already gone through several waves, including e-commerce, quick commerce, crypto and fintech. I had always thought of myself as more of a B2C person, but when AI arrived it felt completely different. I wanted the time and headspace to dig into the space properly, which I could not do while running an intense business at Deliveroo. I still thought I might build something in B2C when I left. I did not expect to end up in B2B, which was a completely new world for me. But I felt it was the right time to step outside and explore it.
**[00:06:45] FWDstart:**
Yeah, food delivery is definitely having a moment at the moment.
**[00:06:48] Anis Harb:**
I'm happy to be on the sideline.
**[00:06:50] FWDstart:**
As it relates to Algebra AI, you mentioned that the company emerged through something resembling an incubation process. How did it evolve from that initial thesis into the business it is today?
**[00:07:04] Anis Harb:**
The initial thesis was very broad. We knew companies were going to need help adopting AI, but we still had to work out how the managed-service model should operate, how to do it properly and where to focus.
**[00:07:16] FWDstart:**
The model has received a lot of validation recently, with OpenAI building forward-deployed teams and other leading AI companies doing something similar. It feels as though you are riding a wave that is very much in vogue.
**[00:07:27] Anis Harb:**
It is interesting because we had been discussing and working on this for many months before those announcements. Each time one appeared, we felt it validated the direction we were already taking. As we have gone into the market and spoken to companies, we have continued refining our ideal customer profile and the workflow archetypes we want to focus on. That takes repeated conversations with businesses across different industries. You quickly see what resonates and what does not. We are not claiming we can solve everything. We are becoming increasingly specific about what works, for whom, and how we can build those solutions more effectively. It is still an evolving process.
**[00:08:17] FWDstart:**
Could you give me a tangible example of one of those workflows?
**[00:08:22] Anis Harb:**
A straightforward example is inquiry management. Traditionally, people might sift through calls, WhatsApp messages and emails containing questions from investors, LPs or other stakeholders. They would search across several data sources, cross-check the information, draft a response and send it back. That required a huge number of human hours. We streamline the inputs and create a knowledge base. When an inquiry arrives through any channel, the system finds the relevant information, interprets it and prepares a response for human approval. Hours of work become minutes. The business can scale without continually adding headcount. That is the cost-efficiency side. We are also doing a lot on the revenue side. One of my frustrations at Deliveroo was how slowly businesses sometimes responded to events. If a competitor acts or something changes in the market, why should you wait until the next business review to take action? Agents can help companies understand what is happening in real time and respond proactively, which can lead to increased revenue.
**[00:10:06] FWDstart:**
What level of experimentation have customers or clients done with an AI in advance of you guys coming in and working with them? Like, is there a preference, do you prefer if it's sort of a blank canvas and they haven't done much, or are you untangling things and going, okay.
**[00:10:20] Anis Harb:**
We meet clients where they are. If a company has no technical team, we can begin with discovery and design the approach with them. If they have an internal AI champion but lack the engineering resources, we can support that person. If the company is further along, we can work alongside its existing capabilities. The key requirements are executive-level sponsorship and a serious commitment to AI adoption. We also look for clear operational pain points with a strong potential return. We are not interested in building something that is merely nice to have. The strongest use cases often sit between disconnected systems such as email, Slack, WhatsApp and an ERP. That is where this approach can create a great deal of value.
**[00:11:15] FWDstart:**
And what does the sales cycle look like. Because oftentimes with I've tried this people in the past and these human-in-the-loop AI workflows, they often get stuck in like a pilot purgatory situation. And it takes ages and it's like, oh, but let's see where the ROI is. And all these different things just push back and everything has become so custom. How do you approach that kind of go to market and when you're embedded? And also a separate question, I suppose, how hands on? Does the client need to be after the workflows are put in place as a situation whereby one internal person is taking ownership of it, or is it distributed across the team that all people have access and can tinker with this?
**[00:11:49] Anis Harb:**
I’ll answer the last question first. We operate as a managed service and remain embedded with the client. Part of our fee is a retainer that covers monitoring, maintenance and ongoing improvements. We are the accountable operator. That goes back to the original thesis. These systems often stop working or gradually die if nobody is responsible for them. We ensure the system continues to run as intended and produces business results. That is especially important with AI. A recommendation alone is not particularly useful, while building a system and walking away is not enough either. Things will break, datand knowledge will change, and employees will leave. Someone needs to remain accountable, whether that is an internal team or a partner such as us. During the sales process, we identify the highest-ROI use case, usually something that solves a substantial problem. We then conduct discovery, map the process from end to end, understand the systems and data involved, and design the proposed solution. We try to build around the way people already work. If a team operates through Slack or email, we do not want to force it into a completely new workflow. We embed the system into its existing way of working and add the intelligence layer on top. Once the client agrees to the scope, we begin building and testing alongside its team. It is not a black box. The client can see the system taking shape throughout the process. After launch, we stay embedded, monitor performance and continue improving it. The main work required from the client is giving us access to the operational process so we can understand it properly.
**[00:14:26] FWDstart:**
Is that a long timeline?
**[00:14:28] Anis Harb:**
It's not a long timeline. It's, you know, we can do this in, I would say, end to end about eight to ten weeks.
**[00:14:32] FWDstart:**
Okay. Okay. No.
**[00:14:33] Anis Harb:**
No, it is not an enterprise-scale timeline. A typical engagement can run from end to end in around eight to ten weeks.
**[00:14:36] FWDstart:**
Like what? That's like, what's the scope here? Yeah.
**[00:14:38] Anis Harb:**
Yeah.
**[00:14:39] FWDstart:**
And that's the benefit of having that more exact place as well. What does the actual outreach strategy look like? Who are these businesses? How are you finding these businesses? It's just like, talk to me about that.
**[00:14:49] Anis Harb:**
All of the above. Some companies approach us directly, some come through our network or referrals, and we also do outbound work. It is still early, but almost every business is thinking about AI and many need help. While we were in stealth, much of the initial pipeline came through word of mouth. Now that we are public, we are doing more outward-facing work to build awareness and generate leads.
**[00:15:18] FWDstart:**
What's the benefit of having Infinity Constellation? So it's actually a really interesting sort of holdco structure that they have in place there.
**[00:15:24] Anis Harb:**
What is the benefit of having Infinity Constellation involved? It has a very interesting holding-company structure.
**[00:16:13] FWDstart:**
They are excellent partners. Francis, Brennan and the wider team have been working in AI services for several years, so we are fortunate to benefit from what they have already learned. We developed the local thesis alongside Infinity, BECO and Silicon Badia. Through the relationship, we can speak with portfolio-company CEOs and exchange ideas on go-to-market strategy, pricing and what each market is seeing. Infinity also provides support across operations, technology infrastructure and go-to-market. They have been a very strong investor and partner.
**[00:16:17] Anis Harb:**
The structure is slightly different from Infinity’s usual playbook because Algebra AI is a separate company. We have our own board and intellectual property, and we do not sit inside Infinity’s US holding-company structure.
**[00:16:29] FWDstart:**
Okay. Okay. That's interesting okay.
**[00:16:31] Anis Harb:**
Because a lot of the businesses are actually owned by Infinity.
**[00:16:33] FWDstart:**
That's what I was curious. Okay. That's really interesting Pricing. Talk to me about that. One of the hardest things to figure is as far as AI is concerned, whether to bill on usage. So you go the old SaaS model subscriptions. Do you do you mentioned retainer. Yeah. Talk to me about the discovery process as far as figuring out what makes sense there because it's look it's difficult. Yeah.
**[00:16:54] Anis Harb:**
I'll tell you a secret. I mean, no, no one.
**[00:16:56] FWDstart:**
Knows for sure.
**[00:16:57] Anis Harb:**
Nobody knows for sure because the market is so new. Companies are experimenting with pricing by outcome, usage, tokens and other measures. Our model currently has two broad components. The first is a retainer that covers monitoring, advisory support, security and the other responsibilities that come with operating a managed service. The second is linked more closely to the complexity and usage of the specific engagement. We are still refining the details, but those are the two main buckets.
**[00:17:35] FWDstart:**
Yeah. Because it's there's a couple different approaches I spoke to, I think it was Aria from applied AI before as well. And they did an interesting thing as far as like the human-equivalent billable hour, I think is their approach and trying to make it more intelligible to understand exactly how the usage looks like and what the ROI is. Yeah. What's what's the most resistance that you get as far as.
**[00:17:56] Anis Harb:**
The greatest resistance is usually around why the client needs us after the system has been built. People generally understand that customisation requires engineering time and that model usage creates token costs. The gap comes from the traditional SaaS mindset, where the customer often did not need an ongoing operator. AI systems are different. We are moving from the SaaS world into an AI-services world, and clients are still learning why the ongoing retainer matters. It is our responsibility to educate the market and help establish how this type of work should be priced.
**[00:18:46] FWDstart:**
Yeah. Exactly. Preparing the way to well, were there any companies that you took as inspiration as far as this was concerned? Like, you know, the one that springs to mind as far as I'm concerned, is maybe Palantir doing something relatively similar, probably derided for it for a while, and all of a sudden I was like, oh no, this far deployed engineer thing, this like, this makes a ton of sense.
**[00:19:03] Anis Harb:**
Palantir is an obvious inspiration. Its model is to embed with clients, build a customised system on top of reusable infrastructure and then remain involved. It is a very sticky model, and we have our own version of it. The biggest difference is scale. Palantir serves major enterprise customers, while we focus on the mid-market. It can place 20 engineers inside a defence department or the NHS for two years. Our engagements are shorter and often involve a business of around 100 people or a large family-owned group that wants to connect processes across several systems. The second difference is repeatability. Palantir can justify building almost anything from first principles if the contract is large enough. At this stage, we are more disciplined about the patterns we take on. But Palantir absolutely established the model.
**[00:20:15] FWDstart:**
What's the division like as you mentioned there the engineering side of things like what's the division of like a time allocation for an engineer working for Algebra AI at the moment, say in terms of an engagement.
**[00:20:25] Anis Harb:**
At the moment, an engineer is probably working across two engagements. That should change as we build more reusable components and can support more clients with the same team.
**[00:20:42] FWDstart:**
The localization side of things. How has that been? Obviously, the playbook that works in the US isn't directly transposable onto a regional context and also maybe speak to is it very UAE centric at the moment, the ambition beyond that?
**[00:20:58] Anis Harb:**
Yeah. Um.
**[00:21:03] Anis Harb:**
We developed the initial thesis with Infinity bringing its US perspective, but the central question was always how to make the model work in this region. Nothing transfers directly from one market to another. I am very sceptical of assuming that something working abroad will automatically work here. You have to return to first principles and understand what local companies are actually struggling with. The labour market is also different, which affects which use cases create the strongest return. Saving human hours can be worth far more in certain US or European industries than it is here. Even at Deliveroo, moving food from one place to another looked universal, but the business had to be localised heavily in every market. Food delivery also showed that regional champions can beat global players. I bring that same perspective to AI services. We were born in this region, are backed by investors from Saudi Arabiand the UAE, and are focused on serving this market first.
**[00:22:19] FWDstart:**
When you went about building the team out, then in terms of first hires, who were the first hires, why did you make them? Why were those roles? What was the first role that you were like, okay, we need to fill this.
**[00:22:30] Anis Harb:**
A CTO was the first role we needed to fill. We went through dozens of interviews, with the whole founding group involved in the process, and landed on an absolute superstar in Fayez. He has hit the ground running and is building an excellent team.
**[00:22:52] FWDstart:**
Now I'm trying to contrast the experience of Deliveroo, which had turned into such a big beast like you mentioned. How many employees?
**[00:22:59] Anis Harb:**
Three or four hundred people.
**[00:23:00] FWDstart:**
Is that a sharp difference when you go back to a much smaller team than after building out a team to that size? Like, what's that experience like?
**[00:23:08] Anis Harb:**
It is funny because Algebra AI is a very AI-native company. We have agents working across the business, and that changes how you think about the organisation you need to build. My initial organisational plan was based on the old model, but you quickly realise how much of it can be removed. Five engineers can now do work that might previously have required 20, while a strategy and operations team can be half the size. It has been fascinating to rethink the company around those efficiencies. Historically, growth meant adding headcount. More headcount created more operational complexity, which required more process, and the cycle continued. I hope we are entering a world where that is no longer the default. We want to prove that inside Algebra AI and then bring the same model to our clients.
**[00:24:14] FWDstart:**
One of the knock on effects. We're starting to see this gradually in the market is redundancies. People losing their jobs as well. Like if we look to the US and look there's an argument to be made is there's actually efficiencies for some of these companies that are just massively or relentlessly hiring. Yeah. When, when money was cheap. How do you think about that or how do you how do you reconcile with that?
**[00:24:34] Anis Harb:**
Like you said, there's a lot of different narratives coming out.
**[00:24:36] FWDstart:**
There's a lot of doom as well, for the record. Yeah.
**[00:24:38] Anis Harb:**
I cannot comment on the specific circumstances behind individual redundancies because every company is different. What I do believe is that companies will need to hire less as they scale. In many of the conversations we are having, the motivation is not necessarily to cut the current team. It is to allow that team to grow the business without headcount increasing at the same rate. There is also a major revenue side to this. Some AI systems reduce costs, but others help a company become more proactive, respond to the market faster and generate more revenue. If that system makes the business more profitable and still requires people in the loop, the company might actually hire more. It is too early to know the overall outcome, but the revenue-driving potential receives far less attention than the cost-cutting narrative.
**[00:25:43] FWDstart:**
Given the service that you offer. Then you touched on there that you have multiple agents running. Obviously Algebra AI is effectively ground zero as far as what you'd be hoping to deploy into other companies. You effectively are under the hood, probably running your own version of this as well. Can you talk to me about that? Like where you're where you find the most value engineering ops, where we're utilizing agents most.
**[00:26:04] Anis Harb:**
Algebra AI is effectively ground zero for the systems you hope to deploy elsewhere. Where are agents creating the most value inside your own business?
**[00:26:45] FWDstart:**
We use them across product scoping, ticket management and related engineering workflows. They also help with legal turnaround, including MSAs, and support the bespoke nature of our sales process. Previously, the pitch, demo and messaging might have been relatively static. Agents now help us tailor the positioning and materials to the specific company we are speaking with.
**[00:26:56] Anis Harb:**
It keeps changing.
**[00:26:58] FWDstart:**
That's kind of what I was getting at. I was speaking to Bayzat as well. They're like, yeah, it was Claude. But then yeah, we still use cursor as well. And it's like often they seem to have multiple and it seems very dependent on the engineer as well. Like certain ones will prefer Cursor Devin or then, you know, someone used called code.
**[00:27:13] Anis Harb:**
Yeah, yeah. And then yesterday, of course, the new Fable model was released and it was like this, you know, which apparently is getting great reviews by the team. So let's see, let's see.
**[00:27:21] FWDstart:**
It keeps changing.
**[00:27:25] Anis Harb:**
I think more like the Chinese models. Yeah, I think the team does look especially I think it's good for you know, it's I think cost is costs are lower and then also good for testing in-house especially. But yeah we're using we're using everything. And I think we have to also you know we also factor sort of where, where regulations and what needs to be sort of local versus which is okay being in the EU or the US. And so there's a lot of kind of dynamics around where the chips and processing actually sit.
**[00:27:54] FWDstart:**
We have experimented more with Chinese and other open-source models, particularly where costs are lower or where we want to test something in-house. We use a broad range of models. We also have to consider regulation and data location. Some workloads need to remain local, while others can be processed in Europe or the US. Where the chips and processing sit is an important part of the decision.
**[00:27:57] Anis Harb:**
It depends on the client and the industry. We are flexible, but there are always trade-offs. The more flexibility a client has, the easier it is to access the strongest infrastructure and models. When there are stricter restrictions, compromises may be necessary. People do not always think about where the chips and processing actually sit, but it matters. Regulation will continue evolving, and that will determine how easy or difficult it is for local companies to use the best available systems.
**[00:28:34] FWDstart:**
Something I forgot to ask earlier about is actually how proactive you are. Once you're already embedded with a client, you set up those initial workflows, how proactive you are with regard to recommending other things that they could do once you've observed that over a particular time horizon.
**[00:28:47] Anis Harb:**
Very proactive. That is another advantage of the managed-service model. Once we are embedded and can see what is working, we can identify additional opportunities and continue building agents across the business. The client does not need to choose one use case and stop there. The first implementation requires the greatest effort because we have to establish the datand orchestration layers. Once those foundations are in place, adding further workflows becomes progressively easier. We still assess every recommendation on return. Agents have a real running cost, and companies can burn through tokens quickly. Almost anything can add some value, but the important question is whether the value justifies the cost.
**[00:29:50] FWDstart:**
I've been talking to somebody recently. I keep bringing up this idea of token maxing and like, you know, certain companies having leaderboards and, you know, which is just nuts. But how do you guys think about that actually from like a token cost. Do you have like a cap in place, like how do you how do you measure that even like an engineer basis? Like, I know certain companies are like. Yeah, it's like 4000 or 5000. Are you thinking, okay, what would the cost of an engineer like a single engineer be on a monthly basis? And that's sort of the tokens are the cost of an attribute.
**[00:30:14] Anis Harb:**
You're saying you mean for our in-house engineers?
**[00:30:15] FWDstart:**
Yeah. In-house.
**[00:30:16] Anis Harb:**
We have not introduced many guardrails yet because it is still early. I am sure there will come a point when we look at the bill and decide we need tighter controls.
**[00:30:21] FWDstart:**
Hang on.
**[00:30:21] Anis Harb:**
Okay, we need something.
**[00:30:22] FWDstart:**
Like AWS, but when you're not paying attention to it, it's like, oh, okay. Oh, God. We need to introduce the gate.
**[00:30:27] Anis Harb:**
Ramp has even introduced a feature that allows companies to track token expenditure alongside their other expenses. It is a strange new world. Taxi receipts, dinner receipts and token usage are all becoming normal day-to-day business costs.
**[00:30:51] FWDstart:**
That is my main takeaway from using Fable at the moment. I keep hitting my usage limit much faster than before, and I am not entirely sure I can justify the price. I also wanted to ask you briefly about food delivery.
**[00:31:08] Anis Harb:**
Of course. I am still a keen observer.
**[00:31:10] FWDstart:**
What do you make of the current food-delivery market and its increasingly competitive nature? You might have experienced the prospect of Keeta’s entry more acutely than most.
**[00:31:23] Anis Harb:**
I had already left before Keeta launched.
**[00:31:26] FWDstart:**
Right.
**[00:31:27] Anis Harb:**
I left Deliveroo around the time the DoorDash acquisition was announced and completed.
**[00:31:31] FWDstart:**
What do you make of how the competitive landscape is developing? What do you think the eventual outcome will be?
**[00:31:38] Anis Harb:**
It is incredibly interesting.
**[00:31:39] FWDstart:**
I have been spending an inordinate amount of time on it.
**[00:31:42] Anis Harb:**
I saw your newsletter. It always made sense to me that the sum of Delivery Hero’s parts was worth more than the company’s total market value, so the recent activity does not surprise me.
**[00:31:57] FWDstart:**
It should no longer be surprising, but somehow it still is.
**[00:31:59] Anis Harb:**
From the outside, Uber appears to have made some smart moves in building its position. I do not know whether that gives it leverage over what it ultimately chooses to acquire, but the Careem transaction made sense. Careem was operating primarily in one market, so there were likely substantial efficiencies available from becoming part of a larger platform with shared fixed costs. The same logic applies on a larger scale to Deliveroo. I do not know the Saudi market well enough to comment in detail, and what ultimately happens to Talabat will be particularly interesting. I am also not sure how seriously noon should be considered in the equation. It remains privately held and closely connected to government-backed capital, so the picture is relatively opaque. Talabat is likely to be the most interesting prize in the region.
**[00:33:15] FWDstart:**
I am going to satisfy my own curiosity here. How meaningful is the uplift from paid subscriptions? When a customer joins a membership programme, does it materially change order frequency and engagement?
**[00:33:36] Anis Harb:**
Without getting into specific numbers, paid membership programmes are absolutely meaningful. Deliveroo Plus was a very strategic decision because customer behaviour changes once someone joins. The challenge is getting customers into the programme. Once they are there, order frequency and engagement can shift significantly. Uber has seen it with Uber One, DoorDash has seen it, and we saw it at Deliveroo. Subscriptions are also one of the few ways a delivery platform can create a pricing moat. The other major lever is exclusivity, which was especially important for Deliveroo in the region and the UK.
**[00:34:25] FWDstart:**
It certainly seems to be working for Deliveroo. I believe 61% of its GMV last year came from paid subscribers. HungerStation Plus is also reportedly at a very high level.
**[00:34:41] Anis Harb:**
I had actually forgotten that. It is a huge number.
**[00:34:44] FWDstart:**
It certainly seems meaningful.
**[00:34:45] Anis Harb:**
I am not surprised.
**[00:34:46] FWDstart:**
At the same time, some platforms have been using subscriptions as a very aggressive promotional tool. HungerStation Plus, for example, may have been offered for around 10 riyals for an entire year rather than as a normal recurring subscription.
**[00:35:02] Anis Harb:**
When so much is being given away for free, the market can become ugly. Keeta’s entry undoubtedly increased the pressure on platforms to offer stronger incentives.
**[00:35:12] FWDstart:**
Deliveroo was something of an exception, though. It did not appear to compete through the same level of aggressive discounting. Would it be fair to characterise it as a more premium offering?
**[00:35:23] FWDstart:**
You were not trying to engage in a race to the bottom.
**[00:35:26] Anis Harb:**
That was always a deliberate strategy. We did not want to compete primarily on price. Instead, we focused on creating value elsewhere, which is why we invested heavily in kitchens and restaurant exclusivity. At the time, a significant share of our GMV came from exclusive partners.
**[00:35:43] FWDstart:**
I need to stop asking you about food delivery. Sorry.