We talk a lot about AI at the national level. But what does “scale” actually mean? The number of people who have tried a tool? The conversations it can handle? The sophistication of the model behind it? Or something simpler?
I have been thinking about this for ten months, watching people use IremboAI inside IremboGov. Not a controlled demonstration, but a platform people already rely on for government services.
Before the numbers, it’s worth looking at why this matters now.
Rwanda’s AI ambitions have moved beyond aspiration; the focus now is what happens when AI goes to work. In June 2026, Cabinet approved a National Artificial Intelligence Agency to strengthen coordination across the country’s growing AI ecosystem. A month later, President Kagame became co-chair of the AI for Good Global Commission, putting Rwanda in a global conversation about AI expanding, rather than concentrating, opportunity.
At the recent Rwanda AI Partners Roundtable, that ambition turned into a practical question: how does Rwanda move from promising pilots to systems that work at scale? The Minister of ICT and Innovation was direct: “Rwanda does not need more demonstrations of what AI could do; it needs disciplined execution and measurable outcomes for citizens, health, agriculture, education, and service delivery.” In other words, beyond pilot purgatory.
This is where Irembo has a useful perspective. Over the past ten months, we have been learning what it means to run an AI support agent inside a real public-service environment with real citizens, real service journeys, and the languages people actually use.
Launched in December 2025, IremboAI is a multilingual conversational assistant that answers service-related questions, explains requirements and next steps, and increasingly helps users check application status.
Through IremboGov, citizens and residents can access more than 250 government services online, with IremboPay built into the payment journey. By the beginning of 2026, the platform had surpassed 51 million transactions. IremboAI entered a service environment people already relied on: they came to apply, pay, and check the status of applications, and some needed help along the way.
Since its launch in 2025, IremboAI has recorded more than 390,000 interactions. It has also grown beyond an informational chatbot into an agentic service layer that helps people complete transactions. The version introduced in September 2026 has powered over 30,000 agentic workflows, from paying traffic fines and bills to navigating driving tests and Mutuelle services.
People are choosing to return to IremboAI, with repeat use up by nearly 70%. Monthly interactions have grown more than 30-fold, but the real measure of progress is what happens after someone’s first visit. As IremboAI has moved beyond answering questions to helping citizens complete meaningful tasks, it has given people more reason to come back.
That matters because there’s a difference between curiosity and usefulness. Someone tries an AI assistant once to see what it does; coming back means it has become useful enough to be part of how they get something done
What are they coming back for? Mostly everyday services such as birth certificates, traffic fine payments, visa renewals, civil and social services. These aren’t abstract AI use cases; they’re the questions people already have when trying to access government services.
Perhaps that’s one of the biggest lessons so far: AI doesn’t always have to create new behaviour to be useful. Sometimes the opportunity is making an existing behaviour easier
This is the part of the data I find most interesting: Kinyarwanda usage has grown from 71% to 81% of all IremboAI interactions, with English and French together accounting for the rest. Why does this matter? We can talk about AI inclusion all day, but eventually we have to ask who can actually use it. For a country that communicates primarily in Kinyarwanda, that can’t be an afterthought; it has to be part of the product from the beginning.
For us, this is more than a localisation question. Language determines who gets to participate. A system can be technically impressive and still exclude people if it doesn’t meet them in a language they’re comfortable using.
These ten months have reinforced that getting an AI system to work in the real world is about much more than the model. Once people are actually using it, you start asking different questions: What are they asking, which services come up most, where do they get stuck, which languages, are they coming back? Those answers are what inform how the product evolves.
This is where responsible AI becomes practical. It isn’t only about principles like safety, fairness, and transparency; inside a public-service journey, it also means paying attention to what happens after launch, understanding where the system works, where it doesn’t, and continuously improving it. Once you understand that someone is using it to get something done, the standard changes.
There’s another lesson that’s hard to ignore: Building useful AI in Africa language models, voice data, annotation, domain expertise, evaluation, safeguards, and deployment experience is bigger than any one organisation, and this is already shaping how Irembo works with the broader ecosystem.
Our collaboration with Digital Umuganda focuses on Kinyarwanda voice models grounded in how people actually speak; and with Green Starz Impakt Hub, on data annotation. These collaborations produce capabilities in language infrastructure, data practices, evaluation methods, and local deployment experience that get reused across applications. That’s how an ecosystem compounds its learning.
After ten months, we don’t have the answer, but we have a better question.
As Minister Paula Ingabire reiterated, the next AI Partners Roundtable should be about sharing impact stories that prove AI is working, not pilots. I think that’s the right challenge.
I will leave you with a question to ponder: “As AI becomes part of essential public services, what should we be paying more attention to when building and deploying it?”
By Ifesinachi Eze
Engineering
Read also
Communications Product - 03 Jan 2025
We have upgraded the IremboGov platform to enhance your experience
The journey towards delivering seamless access to services starts with listening to you, our users. Guided by your valuable feedback,
CSW - 03 Oct 2023
Making Lives Better One Interaction At a Time
AS WE CELEBRATE Customer Service Irembo’s overarching mission to improve people’s lives meaningfully, our dedication through every interaction with our
Communications - 12 May 2023
Irembo, a digital services afro champion becomes a member of the Smart Africa Alliance
Victoria Falls, Thursday 27th April 2023 Irembo, the leading digital services partner of the Government of Rwanda, is proud to
Communications - 08 Jul 2022
Israël Bimpe joins Irembo as the new Chief Executive Officer
Kigali, July 8th, 2022 – Irembo announces today the appointment of Israël Bimpe as our new Chief Executive Officer effective
Product - 31 Dec 2021
IremboGov in 2021
In September 2020, we completed the migration of IremboGov to a brand new 2.0 version: a major overhaul of the
Engineering - 19 Nov 2026
Running AI at National Scale: What 10 Months of Citizen Usage Tells Us
We talk a lot about AI at the national level. But what does “scale” actually mean? The number of people
Impact Story - 30 Jun 2026
How IremboPay Is Transforming Payment Collections Across Rwanda’s Health Facilities
When payment is part of a patient’s care journey, it should feel like a natural step that helps them move