Comment: Mapping your route out of pilot purgatory

The UK’s artificial intelligence (AI) sector began 2022 on a positive note. In January, the government announced a new AI Standards Hub, as part of its National AI strategy.

The Alan Turing Institute, British Standards Institution (BSI) and National Physical Laboratory (NPL) are coming together to help shape global technical standards for AI. It’s a welcome step, as it means more support and guidance for UK businesses working with AI – which will number more than 1.3 million by 2040, according to government figures.

AI has enormous transformative potential for manufacturing businesses. It can revolutionise innovation, turbocharge their operational capabilities, foster agility and accelerate digitisation. But for its potential to be truly unleashed, firms must be able to move forward with AI at scale, and at pace.

That ability remains elusive for many of them, however. According to McKinsey, 74 per cent of manufacturers are stuck in ‘pilot purgatory’, unable to scale up their use of Industry 4.0 (i4.0) technologies such as AI.

Escape route

If they can’t deploy their AI applications at scale, then businesses won’t reap their powerful benefits. So how can they find their way out of pilot purgatory?

In my experience, getting AI adoption to critical mass means following these five steps: 

Use cases

Start by selecting a handful of very specific use cases. Identify the problem you’re trying to solve, and which technology will be most effective. And make sure your use cases will support the business’s key strategic aims, such as improving productivity or driving cost efficiencies. Most AI projects fail because they don’t define the problem they are after specifically enough, and instead apply generic AI tools to fairly generic problems; they substitute boiling the data lake for boiling the ocean, so to speak. The result is a lot of effort and time, with no actionable findings aligned to specific teams or goals.

Then once you’ve got a model that works, design a rapid, agile process for refining, rolling out, and aggressively scaling your chosen use cases up.

The machine health category, within predictive maintenance, is a good example of this. Powered by AI and IoT technology, machine-health solutions minimise breakdowns and plant closures, by forecasting equipment faults before they become critical.

These predictive maintenance capabilities help achieve objectives such as better productivity and cost efficiency, by keeping plants running and preventing downtime. And they can provide foundational insights for other use cases, such as asset lifecycle management, yield optimisation and predictive quality.  Businesses must lead with objectives when making AI decisions.

Value team   

The next task is to establish the teams that will be responsible for defining the tangible value your firm should expect from each use case, in quantitative and qualitative terms. And for figuring out how to derive that value – quickly, and at scale.

At the corporate level, a value team should include a champion, who brings a C-suite perspective and takes responsibility for enterprise-level results; and an enabler, who can provide vital functional expertise to the use case.

Then at local level, site leaders should champion and supervise the use case at the location(s) where it’s being implemented; while a problem-solver acts as technician and/or operator for the technology being trialled. Another common mistake in AI projects is that the problem statements and ideal solution space are defined by people far removed from the actual work. More successful AI-based solutions are human-centric, meaning the UI and the way the AI-driven insights are applied is designed from the needs of the actual users and the work backwards, not from the data science team forward.

Onboarding

The quicker your use case is up and running, the faster you’ll see results. This sounds obvious, but in the manufacturing world where delays lasting mere minutes can be costly, time is of the essence. And the world is littered with failed IT projects, so scepticism can be high until results are clear.

Businesses must work to a structured onboarding programme that covers installation, training, integrating the technology into workflows, setting targets, and achieving quick wins. This will allow sites to understand the technology, how to use it, and how to derive value from it, as quickly as possible.

Adoption

Quick wins gained during the onboarding phase will help the use case to build momentum, enabling you to harness additional value during the adoption phase.

Adoption begins by setting performance targets. For instance, a machine-health solution can allow maintenance teams to address 85 per cent of alerts, and make repairs within 10 days.

Each success delivered from the new system needs to be quantified – to help show the wider business benefit. From your sales teams to the boardroom, keep staff updated on the progression of the AI system, and make it clear how it can benefit their area of the business.

Scaling

Once adoption is high at the initial sites, the next phase is to expand the value that the technology provides. This can be done by capturing new types of value at current sites; and rolling the use case out to new locations.

That will require you to write a playbook. Package up the lessons learnt at the early sites, including any challenges and roadblocks, to accelerate time to value at your new sites. Leverage videos, presentations and ‘what it means to you’ approaches to show the impact from the point of view of the audience. The benefits of predictive maintenance from the point of view of the technician on the floor are very different from the VP of Operations, even though both gain great benefit from the technology.

You’ll also need to appoint ambassadors. Send champions from existing sites to visit the new locations. And use your corporate champions to reach other facilities and stakeholders. Over time, establish a community of champions to mentor others across the business on how to use the technology. 

The scaling stage is the time to take value to the next level. How to achieve this will vary depending on the use case. For a machine-health application, it could mean going beyond repairing machines early, to planning essential shutdowns more efficiently, or identifying systemic reliability problems.

A note of caution: as you build, adopt and scale use cases, your AI solutions will make a growing number of high-risk decisions, with less human oversight. Make sure you fully understand how the AI is trained and ensure you have the ability to get transparency on its decision making processes so you can gain more confidence in the accuracy of the solution and diagnose when something doesn’t make sense The best results come when AI insights are paired with human creativity and experience, so be sure to build an approach that leverages both.

Following a clear gameplan will help mitigate these risks. And it will ensure that you move out of pilot purgatory, and embrace the dramatic benefits that AI promises.

Saar Yoskovitz, CEO at Augury