Can an AI-Approach to IoT Fix Adoption?

IoT was meant to transform the physical world. A decade on, adoption is still slower than the industry promised.

In this episode of IoT and AI Leaders, Nick Earle speaks with Afzal Mangal, Founder of Hello Things and author of IoT: The Hype No One Knows About, to unpack why IoT scaling is still hard, even when the tech “works” in pilots. The conversation lands on an uncomfortable point. Most IoT projects fail for reasons people underestimate, especially the device layer and long-term firmware reality.

Then the discussion pivots. If AI has reached the mainstream while IoT remains invisible, could an AI-approach to IoT finally reset the adoption curve?

Why hasn’t IoT scaled the way the industry predicted?

Afzal’s view is that many pilots succeed technically but organisations are not ready for what success implies.

He describes early low-power network optimism across NB-IoT, LoRaWAN, and Sigfox. The assumption was that cheaper connectivity would unlock “mass scale” deployments. What actually happened was slower. Not because the radios did not work but because people, processes, and incentives did not change at the same pace.

Afzal also points to an awareness gap that still hasn’t closed. Even use cases discussed for years, such as smart waste management and fill-level sensors, remain unknown to many target buyers inside municipalities.

The result is a pattern the industry knows too well: pilots happen, the business case stays vague, and scaling stalls.

Is the device really the biggest failure point in IoT?

Afzal argues yes, and Nick agrees from hard operational experience.

The key claim is simple: in a full IoT stack, most components can be swapped. A SIM can be changed. A dashboard can be replaced. Data can be routed elsewhere. But if the device is wrong, you have a costly problem that is difficult to recover.

Afzal highlights why this hits hardest in hardware:

• Hardware experimentation is expensive
• There is no perfect “debugger” for antenna placement and physical design
• Device failure wastes time and components at once
• The device becomes the least interchangeable layer in the stack

Nick reinforces this with a specific operating principle from Eseye: they will not take on a project without the ability to audit the device firmware, because firmware behaviour determines whether global connectivity and switching can work reliably over time.

Why firmware matters more than connectivity for long-term success

A recurring theme in the episode is that many teams treat connectivity like a commodity and underestimate device-side engineering.

Nick describes “the dance” between the SIM, modem, and processor. When projects move beyond a single-network model into more flexible global approaches, device firmware can become the limiting factor.

This is why firmware shows up as a root cause late. IoT projects often look fine early on, then fail after time in the field, when edge cases appear and device behaviour becomes harder to control.

The takeaway is not that connectivity is unimportant. It’s that device firmware is often the single biggest determinant of whether the system stays healthy at scale.

What is the “awareness gap” that keeps IoT invisible?

Afzal offers a useful contrast between IoT and AI adoption.

He describes how mainstream users can name AI tools and explain why they use them. His example is his mother, who understands AI through tools like ChatGPT for writing letters or creating invitations. She expects AI-driven automation to change work.

At the same time, she uses connected devices at home without framing them as IoT. The experience is “smart”, but the concept of IoT stays invisible.

Afzal’s argument is that invisibility slows demand. If people cannot name a capability, they do not go looking for new use cases. That is true in consumer life and inside enterprises.

Can AI make IoT easier to adopt?

Afzal is optimistic about the potential of AI and IoT but raises a practical constraint. Many IoT solution providers are already stretched, surviving on a small number of customers. Adding AI features can improve outcomes and marketing credibility but it also requires time, money, and skills.

He gives an example using IoT sound sensors. Without AI, a device might alert when noise exceeds a threshold. With AI, the alert could become more meaningful by classifying context, such as laughter versus distress. That type of context makes the alert more actionable.

The promise is clear: AI can increase the value of IoT signals by turning raw readings into decision-ready insight.

The risk is also clear: the companies that need the boost most may not have the resources to deliver it.

What does an AI-first model look like?

Nick introduces an “upside down” way of thinking: start with AI and business processes, then use IoT as the data layer that feeds those outcomes.

He shares a smart hospital example from a recent podcast with Borda Technology where the value is not framed as “we deployed sensors.” It is framed as process performance and bottlenecks, with IoT tags and sensors providing the evidence.

In that example, the system maps the real flow of people, equipment, and patients. The output is a management view of where time is lost and where throughput can improve. The “IoT project” becomes a business operating improvement project, supported by data.

That framing matters for adoption because it speaks to leaders in outcomes, not infrastructure.

Could AI also solve IoT integration complexity?

Nick raises a second possibility: AI could reduce the integration burden that slows IoT projects. He describes a familiar reality. IoT systems often become a tangle of platforms, APIs, and dashboards. Teams spend months moving data from one place to another. Even after connectivity works, integration remains fragile.

The question is whether AI can help assemble signals and insights without hand-building every connection and rule. Afzal believes it’s possible but emphasises that humans still need to decide what data to bring in. AI can help interpret and organise but cannot replace the human decision to instrument the world and connect the right points. He adds a telling observation. When he asked mainstream LLMs for digitisation ideas in sectors like healthcare, construction, farming, and city development, the answers focused heavily on automation. They did not surface IoT use cases.

That suggests a real limitation today: if AI models are not exposed to IoT thinking, they will not naturally propose it.

What leaders should do next

This episode lands on a practical conclusion. IoT and AI can reinforce each other, but only if leaders treat the foundations seriously.

1) Treat device and firmware as first-order risks

If you cannot trust the device layer, you cannot trust the data. If you cannot trust the data, you cannot scale AI outcomes on top of it.

2) Sell outcomes, not infrastructure

Adoption accelerates when projects are framed as business improvement, not a sensor rollout.

3) Bridge the awareness gap internally

If operational teams and decision-makers do not understand what is possible, demand will stay weak even when solutions exist.

4) Collaborate with AI stakeholders, not only IoT stakeholders

Afzal’s closing insight is strategic: IoT providers may need to align with AI consultants and AI-led transformation teams, because those teams already have executive attention.

Catch the full episode

Listen now to hear why rethinking IoT through an AI lens may be the reset the industry needs.