AIoT systems are those which combine connected devices, operation data, software systems and AI to aid decision-making in environments where physical assets and processes are key
This combination is an opportunity for venture building, but not without additional constraints over and above those found in software products.
A good place to start is working backwards from the operational problem.
- Define The Physical-World Problem
A good AIoT system starts with a clearly defined operational challenge.
Aim to avoid statements that begin "We should use AI to..." and think first about what is already happening and what information is difficult to obtain:
Which assets or processes are involved?;
What information is currently challenging to get hold of?;
Which decisions are being made based on this information?;
Where are delays or gaps in visibility occurring?;
This defines an actual problem to solve, and not just a technical architecture.
- Map The Data Sources
With the problem defined, it's necessary to think about what data can help describe the relevant part of a physical process.
Depending on the environment, this is likely to involve connected equipment, sensors, existing operational systems or data sources - the key point is that it's not about volume, but relevance and reliability
A good AIoT architecture will define how the data from the environment can flow into systems for analysis, and will involve building an appropriate chain of custody:
Physical assets -> Data collection -> Connectivity -> Processing -> AI/analytics -> Operational decision
Each part of this chain must be relevant to the use-case.
- Think About Integration Early On
Industrial environments are rarely standalone technology silos
New capabilities are likely to need to work in tandem with existing processes and systems, and must be viewed through the lens of early integration and adoption.
A technically sophisticated solution is of little use if it cannot be made to work within an existing environment.
- Bind AI To A Decision
AI models should always have a clear purpose within a workflow, rather than being viewed as an end in themselves.
Ask what decision or action the model's insights are intended to inform, and build an appropriate chain between:
The data -> analysis -> insights -> decision -> action
The closer this is to the original operational challenge, the more likely it is that value will be delivered.
- Test If The Problem Is Repeatable
A single successful implementation is rarely enough for sustainable venture building - the most interesting opportunities will always be for organisations with similar underlying problems to solve
This raises the challenge of identifying which parts of the solution are portable, and which require reworking for each environment.
This distinction will define whether a single customer win is interesting, or whether there's potential for a more substantial product-led opportunity.
- Look Beyond The Tech Stack
A successful AIoT product always involves more than the model itself, and requires attention to the environment, integration requirements, user processes and the business problem being addressed.
This is why problem-first development can be so powerful - it binds together the technical solution and the opportunities for delivery in the real-world.
A venture-building approach to AIoT can be viewed as a progression through:
Industrial problem -> Data requirements -> System architecture -> AI/analytics -> Operational workflow -> Repeatable product
The actual architecture will vary according to the environment, but the principle of "problem-led tech" should always apply.
Conclusion
AIoT represents an opportunity to bring software intelligence to bear on physical-world operations, but there are additional challenges for building sustainable systems and products that must be navigated.
This makes for a fundamentally different challenge to pure software product development, and a problem-first approach will always be more valuable to venture builders.
Aperture Venture Studio focuses on venture building at the intersection of AI, IoT and industrial environments . Aperture Venture Studio provides a relevant example of this approach.
The most interesting opportunities in AIoT will always be where recurring physical-world problems can be addressed with new data and intelligent software.
Top comments (0)