Inside CBK's AI strategy: Machine learning, mobile payments data and economic nowcasting
The Central Bank of Kenya (CBK) is implementing artificial intelligence applications across its operations, with nearly 300 use cases identified across 14 departments, according to the bank’s Head of Research. The CBK AI strategy includes machine learning for inflation forecasting, economic nowcasting using alternative data and tools to analyse mobile payments and online supermarket prices, giving researchers additional information to support monetary policy decisions.
The Head of Research said the bank was working to implement several applications in 2026 while assessing the opportunities and risks associated with the technology. The applications range from automating routine tasks to using AI to analyse large datasets and exploring agentic systems that could perform tasks currently handled by people.
Research is a key focus, particularly as economic volatility complicates forecasting and delays in official statistics make it harder to assess current conditions. The bank is exploring how AI can complement conventional economic models and provide more timely indicators of inflation and economic activity.
CBK identifies nearly 300 AI use cases across 14 departments
The Head of Research said CBK had identified almost 300 AI use cases across 14 departments and was implementing a number of them during the year. The applications include routine automation in banking, payments and research, where tasks previously completed manually can be performed more quickly.
He described four levels of AI application, beginning with routine or intelligent automation, followed by machine learning, generative AI and agentic AI. Each offers different capabilities, from reducing the time spent on repetitive work to analysing complex information and enabling systems to carry out tasks with greater autonomy.
The bank is also piloting an AI-based chatbot in its Financial Markets Department, according to the Head of Research. He said CBK was assessing opportunities across its operations while paying attention to the risks associated with implementing the technology.
The different applications give the regulator options for improving internal processes while developing more sophisticated tools for research and economic analysis.
Machine learning to complement inflation forecasting models
CBK is exploring machine learning to improve economic forecasting, particularly during periods of volatility when relationships between economic variables can become less stable. The Head of Research cited the energy crisis that began in late February 2026 as an example of the conditions complicating inflation forecasts.
He explained that conventional forecasting models typically organise the economy into different blocks, including the real economy, monetary conditions and the external sector, before modelling the interactions between them. During periods of disruption, some of these relationships can change, making it harder for traditional models to produce reliable estimates.
Machine learning can help by allowing models to learn from previous forecasting errors, according to the Head of Research. If a forecasting exercise underestimates second-round effects, for example, the model can be trained to give those effects greater weight in subsequent projections.
CBK is working with a combination of models intended to minimise forecasting errors and improve projections, particularly during volatile periods. The Head of Research said AI would complement the bank’s traditional forecasting methods, with economists continuing to apply their judgment when interpreting the results.
He emphasised that human judgment remains important because forecasts must be assessed against economic conditions and the relationships between different variables. AI can help researchers identify patterns and refine their estimates, while people remain responsible for evaluating the information used in critical decisions.
AI could help CBK track economic activity before GDP data arrives
CBK is also exploring nowcasting, a method used to estimate the current state of the economy when official statistics are not yet available. The Head of Research said the approach was important for monetary policy because the Monetary Policy Committee needs a timely assessment of economic conditions when making decisions.
He cited the release of GDP statistics as an example of the challenge. According to the Head of Research, second-quarter GDP data became available at the beginning of October 2026, roughly three months after the quarter ended, meaning policymakers need other indicators to understand what is happening in the economy at the time of their meetings.
The bank already uses high-frequency indicators, including revenue estimates from different sectors, and is exploring AI methods that can supplement these sources. The Head of Research said satellite observations could provide additional information about economic activity through changes in night-time lights, electricity consumption, vegetation and atmospheric conditions.
Night-time illumination can help researchers assess patterns associated with industrial and commercial activity, while vegetation changes can provide information about agricultural conditions. AI methods can process large volumes of such observations and help researchers extract indicators that complement conventional economic statistics.
The Head of Research also cited mobile payments as a potential source of information for estimating economic activity. He said CBK had observed a relationship between GDP and the value of mobile payments, estimating that the latter could be roughly 2.5 times GDP.
With daily, weekly or monthly payment data, researchers could use that relationship to develop a rough estimate of economic activity between official GDP releases. The Head of Research stressed that these measures would complement existing methods, with AI helping the bank work through the large volumes of data involved.
Mobile payments offer a potentially useful source of high-frequency economic information because they generate transaction data more frequently than quarterly national accounts. Analysing those patterns could help CBK assess changes in economic activity while it waits for official statistics.
Online supermarket prices could improve inflation monitoring
CBK is also exploring web scraping to supplement its existing methods of monitoring inflation. The Head of Research said the bank wanted to assess price pressures during the month rather than relying solely on the monthly release of official inflation figures.
The bank already uses agricultural surveys, inflation-expectations surveys, CEO surveys and market-perception surveys to inform its analysis. Online supermarket prices could provide another source of information for researchers assessing price movements across consumer goods.
According to the Head of Research, the Kenya National Bureau of Statistics has 330 items in its consumer price classification. AI methods can help match products sold online to the relevant categories, accounting for differences in brands and quantities, while large language models can assist in validating some of the collected data.
The resulting information could give researchers additional insight into price movements as they prepare for monetary policy discussions. Online supermarket data would supplement the bank’s existing surveys and other indicators, helping it assess inflationary pressures before the next official release.
Agentic AI raises new questions for payments and governance
CBK is also considering potential applications of agentic AI, which the Head of Research described as systems trained to carry out tasks that would otherwise require human intervention. He identified payments as one area where such applications could be useful, while acknowledging the opportunities and risks associated with greater autonomy.
The distinction matters in financial services because AI can be used to analyse information, identify patterns and recommend actions, while separate controls govern whether those actions can be executed. TechTrendsKE’s previous coverage of M-PESA’s approach to AI agents examined the importance of retaining control over financial authorisation and settlement, providing a useful framework for understanding the questions raised by agentic systems.
CBK’s broader regulatory work also addresses AI use in financial services. TechTrendsKE reported in October that the bank’s non-deposit-taking credit provider regulations include requirements concerning explainability, human oversight, accountability, privacy and security in automated lending decisions.
The Head of Research said human judgment remained essential when evaluating AI-generated forecasts and other analytical outputs. He also stressed the importance of AI governance, particularly the risk that staff could inadvertently expose sensitive information when using automated tools.
The bank is assessing opportunities and risks on a case-by-case basis as it implements AI across its departments. According to the Head of Research, this approach is intended to allow CBK to benefit from the technology while managing the risks associated with its use.
Go to TECHTRENDSKE.co.ke for more tech and business news from the African continent and across the world.
Follow us on WhatsApp, Telegram, Twitter, and Facebook, or subscribe to our weekly newsletter to ensure you don’t miss out on any future updates. Send tips to info@techtrendsmedia.co.ke
