Login
Register
Safaricom’s Head of Billing and Core Applications explains how a new AI-assisted product design, configuration and recommendation tool, built using TM Forum’s Open Digital Architecture and Open APIs, has delivered immediate business benefits and is now feeding into its wider data ecosystem transformation.
Joanne Taaffe, TM Forum
Safaricom makes efficiency and CX gains with AI-assisted personalized offers
What: Faster, lower-cost design and configuration and recommendation of personalised offers, including for price-sensitive customers
How: Development and launch of ‘Idea-to-cash’ an AI-assisted product design, configuration and recommendation tool
Results:
Safaricom’s annual report for the year ended 31 March 2025 pointed out a challenge: its customers’ disposable incomes are under pressure, and they demand more value.
At the same time the company aims to become Africa’s leading purpose-led technology company by 2030. Growth of Safaricom's digital services portfolio, therefore, will depend on customers maximizing their use of connectivity at affordable prices.
To this end it has been investing in new systems and platforms that will use data and AI to more quickly and efficiently deliver on Safaricom’s culture of "customer obsession".
“Our current platforms cannot support our new business model because we are doing a lot more of work beyond connectivity. We are not just a standard telco anymore,” says Mark Oyier, Head of Billing and Core Applications, Safaricom.
One example of how Safaricom is improving both customer experience and efficiency is ‘Idea-to-Cash’. Launched in July this year, and co-developed with Huawei, the Idea-to-Cash platform builds on a recent upgrade to Safaricom’s converged billing system to enable product design and marketing teams to use AI-driven native language requests to generate and configure products, as well as recommend tailor-made packages to different customers at the most appropriate time
Previously Safaricom created a static bundle of generic products. “Whether you're a high data user or you're a low data user, you were seeing the same bundle on the platform, which doesn't make sense,” explains Oyier.
In addition, a manual approach to product development and delivery meant either spending weeks creating a product from scratch, or editing an existing product, which saved time but could result in errors.
“Safaricom was pulling all this data out of the billing platform, and then consolidating them out of the system and then running some machine learning ... and pushing it back for recommendation to the customer," says Oyier. "This was a lot of work. We are now able to do this directly on the billing platform."
The system, which incorporates generative AI (GenAI), allows employees to use natural language to describe a data bundle “and the system pre-generates for you a configuration file that you can just verify and publish and start testing,” explains Oyier. “So we moved from something that used to take almost five days to something that takes you an hour in terms of getting the best configuration.” Idea-to-Cash then dynamically matches users with the most relevant offers based on rich profiles and real-time context, so time to market has fallen to around five days, down from four weeks or more.
Crucially, business teams without technical skills can use the system.
“The person doing the design thinking – the product and marketing manager – is actually typing that into the platform and generating the product,” says Oyier. As a result, “I just need a central team that validates and supports in the testing. That lets me reduce my development costs significantly.”
And although the AI technology itself is not ground-breaking, admits Oyier, “the way it's used in platforms ... is definitely a completely new approach. The adoption of AI is significantly changing how the billing team work.”
It’s also transforming customer experience. It has an immediate impact, for example, on how customers consume one of Safaricom’s most popular data bundles, which is an hour-long data package, through a feature called the validity extender.
Previously if a customer encountered connectivity issues and “you only use the one hour bundle for 10 minutes, and the rest of the time … you are unable to use it, then you raise a complaint, saying we've robbed you,” says Oyier.
Now Safaricom is able to automatically extend the validity of a customer’s package, based on smart offer recommendations. This has helped reduce data bundle customer complaints by about 70 percent," explains Oyier. "Subsequently this is having an impact on our cost-to-serve. It has further also increased ARPU with doubled conversion rates, because many customers with good experience now trust the hourly bundles and buy more."
Integrating with ODA and Open APIs
One of the challenges the teams faced was integrating Idea-to-Cash into the customer's end-to-end journey. “That's where we we did struggle a bit. There was a bit of work that we had not planned for, that we had to do during the project to make sure the customer journey is not broken.”
Using TM Forum standards has helped Safaricom address end-to-end complexity.
“One of the pleasant surprises was that the platform … already had APIs that were TM Forum ready. And [because they were] already standardized … it was easier,” according to Oyier.
Safaricom also reduced fricition in customer journeys by building an Open Digital Architecture (ODA) wrapper to expose the solution to customers.
“We are on a journey to set up a proper ODA framework across in our entire ecosystem,” says Oyier.
Indeed, “a lot of it is already documented and available within TM Forum," he adds. "What we did was take a bold step to push our vendor to see what more can we do on our platform that would get us value for our customers and for our business."
Notably, the integration experience provided insights that will now inform planned upgrades to Safaricom’s customer relationship management (CRM) ecosystem to support sales and product management.
“We can see the constraint on the upper layers because of the data structures … integrations that are not standardized,” notes Oyier.
Addressing the data challenge
Idea-to-Cash fits into a much wider investment to support Safaricom’s TechCo ambitions.
“The introduction of AI in billing has been exciting because it preempts a lot of the work we were trying to do," says Oyier.
Notably, Safaricom has invested significantly in hiring data scientists and machine learning engineers to work on a larger big data project, says Oyier. These efforts include establishing a data lake house within our Data Centre in collaboration with Dell and integrated to AWS, Google and Microsoft for specialised use cases.
The operator has harnessed “some of those skills within the core platforms whenever we were stuck in terms of understanding what the Huawei team were doing, helping us to challenge the work that was going on there, and also to define how the requirements need to look, because we are coming from a product angle,” says Oyier.
Safaricom plans to extend Idea-to-Cash to more products, including in the enterprise B2B space. In tandem it would like to use data from the core network, and services such as its financial service, M PESA, which has approximately 38 million customers in Kenya, to generate integrated customer propositions.
“The work that we have done so far not only sets the base for that extra platform that we are setting up, but it also simplifies our AI journey in the sense that we've been able to achieve certain customer objectives almost immediately and in real time,” explains Oyier. It has “created an opportunity for us to expose this data that is being generated through AI billing to a much larger platform and build a lot more machine learning models that can be used for other types of analysis.”
As Safaricom builds out its big data ecosystem, for example, it is working closely with Huawei to generate Kafka streams from the billing platform.
“We can mix it up with other data in our ecosystem and use it for larger models, according to Oyier.
It is now discussing with Huawei how to expand usage while limiting investment in computing resources.
“How can they optimize their algorithms and their machine learning models, so that I don't have to invest in GPUs to achieve the same benefit?,” asks Oyier.
This is important because Safaricom, which has more than 50 million subscribers on its 4G/5G networks, is looking to extend personalized data offers and ecosystem services to all of its subscriber base, up from three million customers today.
To do so it is using its AI-based systems to help address the challenge of relatively low national smartphone penetration.
“We are actually using part of the work that was done to try and drive this [smartphone] penetration,” says Oyier. Based on a customer’s feature phone usage, the company can evaluate whether a customer would benefit from a low-cost smartphone offer “that that not only puts a smartphone in your hands, but it allows us to interact with you more than we are right now."
“It not only benefits customers, it also benefits the business because the revenue is growing on data, not on voice. So we need to move our discussion from the feature phones, where [customers are] only consuming voice, to a smartphone where they can start consuming data.”
