Making AI Adoption Visible: A Measurement Perspective on Kenya's AI Journey

The global conversation on artificial intelligence is increasingly moving from possibility to implementation. Countries are developing national AI strategies, establishing governance structures, investing in infrastructure, and promoting innovation ecosystems.

Yet one fundamental question remains:

How will we know whether AI is actually being adopted, and whether it is delivering measurable value?

This question has been central to my doctoral research and led to the development of the Three-Layer AI Visibility Framework (3L-AiVF™).

The framework emerged from a broader challenge facing policymakers, researchers and development practitioners: while many digital transformation indicators measure whether capabilities exist, they often do not reveal whether AI is being adopted by identifiable entities or whether observed outcomes can reasonably be attributed to AI.

From AI ambition to AI measurement

3L-AiVF™ was developed to address this measurement challenge by examining the evidence that policies and strategies provide about AI adoption.

It distinguishes between three levels of AI visibility:

• AI diffusion, where policies measure enabling conditions such as infrastructure, skills, governance, investment and institutional readiness.
• Countable AI adoption, where identifiable organisations, institutions or sectors using AI can be measured.
• Attributable AI outcomes, where measurable improvements can reasonably be linked to specific AI adoption.

The framework does not assess whether an AI policy is good or bad. Instead, it asks a narrower but important question:

What does the policy enable us to measure about AI adoption?

Kenya as a demonstration case

Kenya's draft National AI and Emerging Technologies Policy, and the monitoring framework it proposes, provides an important opportunity to demonstrate the application of this framework.

Kenya is among Africa's leading countries in recognising the strategic importance of AI and developing a comprehensive policy response. Its policy demonstrates ambition across governance, infrastructure, skills development, research, innovation and responsible AI adoption.

Applying 3L-AiVF™ to the policy is therefore not an exercise in criticism. It is an opportunity to explore how even ambitious AI strategies can make their implementation journeys more measurable.

The assessment demonstrates a pattern likely to be relevant far beyond Kenya.

Strong visibility of AI diffusion

If implemented as designed, the policy's proposed monitoring framework would provide strong visibility of AI diffusion.

Many measurable commitments relate to:

• governance structures,
• institutional arrangements,
• infrastructure development,
• skills development,
• research capacity,
• innovation ecosystems,
• enabling conditions for responsible AI adoption.

Approximately 70% of the policy's measurable commitments connected to AI adoption sit at this diffusion level. For a first-generation national AI policy, that is a reasonable starting point, but it also shows where the greatest opportunity for improvement lies.

These foundations are essential. Without them, AI adoption cannot scale.

Emerging visibility of AI adoption

The assessment also identifies emerging evidence of countable AI adoption.

Some indicators identify specific institutions, firms, startups and other entities connected to AI implementation.

However, many indicators remain focused on ecosystem participation or readiness rather than confirmed adoption of AI systems by identifiable users.

This distinction matters.

An organisation receiving AI-related support, infrastructure or training is not automatically equivalent to an organisation actively deploying AI systems.

Measurement must distinguish between creating the conditions for adoption and demonstrating that adoption has occurred.

Encouragingly, some of the policy's own proposed mechanisms already point toward a way forward. A national registry of deployed AI systems, a vetted marketplace of AI solutions for government use, and an audit process for high-risk systems each have the potential to link a specific system to a specific adopting institution, the foundation attributable measurement requires. Whether they deliver on that potential will depend on a single design choice: whether each entry can identify both the system and the institution using it, not just one or the other.

The challenge of attributing AI outcomes

The most significant finding relates to AI outcomes.

The policy contains important ambitions around improved service delivery, productivity, competitiveness and public value.

For example, the policy sets targets for improved productivity and service delivery linked to AI adoption across government agencies, yet its accompanying indicators do not yet specify how such gains would be distinguished from other digital transformation efforts happening at the same time.

Essentially, the current measurement architecture does not yet consistently provide for enough evidence to establish attributable AI outcomes.

This does not mean AI will not generate outcomes.

It means that demonstrating those outcomes requires stronger measurement systems capable of linking:

• a specific AI intervention,
• an identifiable adopter,
• a measurable change,
• and a credible attribution pathway.

A shared challenge for developing countries

This challenge is not unique to Kenya.

It is particularly relevant across developing countries and least developed countries, where fragmented data systems, limited interoperability and uneven institutional capacity can constrain the ability to track technology adoption and impact.

The Kenya policy itself recognises the importance of addressing data gaps and weak interoperability. This highlights a broader reality: successful AI transformation depends not only on technology adoption, but also on the ability to generate reliable evidence about that adoption.

A measurement contribution, not a MEAL assessment

It is worth noting that, on a purely technical basis, the policy's proposed indicators already correspond to the large majority of its own stated commitments, a sign of a genuinely thorough monitoring design, even where the resulting visibility into AI adoption itself remains limited.

This perspective draws from Monitoring, Evaluation, Accountability and Learning (MEAL) principles, but 3L-AiVF™ is not a MEAL assessment framework.

It does not evaluate:

• whether indicators are SMART,
• the quality of a Theory of Change,
• logframe design,
• or the overall quality of a monitoring system.

Instead, it applies measurement principles to examine whether AI adoption is visible, measurable and attributable.

My contribution is therefore not to evaluate Kenya's policy design. It is to contribute a measurement perspective to an emerging global challenge.

Looking forward

Pioneering countries help create the pathways that others follow. Measurement frameworks help ensure those pathways can be understood, evaluated and improved.

Kenya is helping Africa imagine what responsible AI governance can look like.

Through 3L-AiVF™, I hope to contribute to the equally important question:

How will Africa know when it has achieved it?