AI at the Edge

The learning curve

At AnotherPeak we have witnessed a rapid increase in requests for Edge Computing consultation and Proof-of-concept labs, driven part by economics and part compliance. We have peeled back the hype and exposed real business drivers across an expanding range of edge functionality. One area of increasing interest is AI at the Edge.

Here are some initial observations which may help you on the AI journey.


We utilise Edge Computing to address cloud based issues such as compliance, latency, frequency and cost.

It is a hybrid model, typically with local triggers and actions augmented by cloud based configuration, policy, analytics, reporting and storage.

Therefore, Edge variants include Mobile Edge(4/5G), IoT/IIoT and Remote Working, encased in a visibility jacket to show board level value.  It is too easy to get embroiled in the technology and forget about the business, so board level buy-in of a strategic Edge Compute plan is key.

That strategy must avoid a silo’ed approach, especially in the IoT space.  Too many hardware vendors have reached outside their comfort zone, grappled and mostly failed to implement meaningful cloud based services to the benefit of no one.

End-users/MSPs need to pick a suitable Aggregation Platform to deliver estate wide visibility via  Private or Public cloud deployments.

Fortunately there is an excellent and expanding choice.

Know your edge

Many consider the Public Cloud as an infinite compute and storage resource. The Edge is finite, even with emerging federated compute and storage techniques.

The emergence of containerised edge functionality offers huge benefits to the operator but comes with tighter integration, deployment and operational requirements.

Once again there are frameworks/blueprints to successfully deliver this.

AI basics

There are business, employee and technical considerations in the Great AI journey.

Depending on the complexity of your AI engine you are creating a continuous improvement process. This is not a software licence with annual renewal, this is a living thing requiring nursing, feeding, burping and the occasional nappy change. Start with a MVP, and review, tweak and retest. Ensure the process scales and new functionality can be added if required.

People rarely embrace the threat of redundancy. Too often AI is sold as pure automation, yet humans and AI engines have complementary strengths and weakness which in turn drives both an education process and engagement when low confidence results need human intuition.

To be successful ensure everyone is onboard and part of the feedback loop.

Start with something simple that illustrates the value of AI within your organisation.

Set realistic expectations and success criteria against a costed life-cycle model. Too many IoT deployments are ‘science projects’ with no real ROI.  AI must not follow that same path.

Reference Material 

One area pivotal to a successful AI deployment is the right reference material. To determine how many carrots are left on my supermarket shelf I must train the engine to recognise a carrot in all its forms, in a range of applicable lighting and potentially via different calibre camera technology.

Such activities take time and resource. Reference data might be cloaked in compliance and security restrictions, so qualify before hand, but the more the merrier.

If you cannot get video, get still images and build from there.

The goal is to release early and learn(continuous improvement) but remember I need reference data to train the engine(80%) and some to test(20%).

We have noticed an emerging dividing line related to the complexity of the AI engine.

The more complex, the more compute and the drift away from local processing to cloud based services, at which point the compliance, latency, frequency and cost issues raise their heads again. 

Moving forward Edge Compute will become more federated extending(with micro services) the capabilities to process locally, but I don’t think we are there yet.


Like 5G there is a degree of hype around AI.

Selling a vision is one thing, delivering on inflated promises is another. 

But AI is much more than a point product, it is a corporate culture, a mindset that if implemented correctly can rapidly improve business efficiency without alienating the work force.

In the ever expanding Edge Computing paradigm, AI has found a welcoming partner, a platform on which to deliver its true potential across numerous market sectors.

Welcome to AI at the Edge.