Artificial intelligence (AI) enables major leaps in optimizing existing industrial processes towards Industry 4.0. Still, AI isn’t a one-size-fits-all magic money tree. To achieve effective return of investments (ROI) with this high-potential technology, with estimates by Accenture that gains of productivity through AI could reach 40% by 2035, you must invest wisely. Read on for details about successful business strategies to get the most of AI and to scale intelligent solutions for company-wide impact.
If possible, avoid projects with big scopes that would imply a long and arduous research phase, and focus on easy and quick wins that bring huge payback. Especially important in your first AI projects is to build internal trust in the technology, where AI serves all stakeholders towards a common goal. You don’t require a fully connected factory to start harvesting the power of AI: by connecting only the data points and sensors you need to solve your business cases, costs, bandwidth, and cloud consumption can be minimized.
Put effort on cases where solutions can be transposed to multiple assets: in the metallurgy sector, steel oxidation concerns every steel plant, for example. Optimizing a furnace to consume less energy with the same quality and yield, could also be adapted to similar furnaces worldwide.
A universal, clear-cut answer to this question is tricky. For companies just getting into AI and where confidence in the tech and its returns needs to be built from scratch, it can be suitable to start with proof-of-concepts exploring the practical usefulness of AI solutions to achieve the company’s goals, in order to gain trust, and prove the ROI before moving other AI projects forward.
If your company size, ability and AI experience is decent enough, several business cases can be explored in non-overlapping areas. In the manufacturing industry, you could for instance explore quality sustainability, performance and maintenance cases in parallel.
There’s also a side-benefit for morale in working with parallel projects: with several plans in development, an inconclusive individual research is less painful.
The more mature the company is in terms of data science and AI, the more it can treat AI projects as typical projects, applying traditional tracking and management. You would have to have built trust with your suppliers, the technology, and your teams. At the stage where you conducted multiple impactful AI projects, failure of some individual AI projects must be acceptable, in a characteristic startup mentality. Invest in multiple projects with a VC approach to AI R&D and bet that a couple of them work out and generate a large ROI.
Having clear, measurable goals shared by the whole company, is also essential in allowing you to assess whether current research could offer positive outcomes. Defining, quantifying, and tracking happiness changes is indeed more challenging to measure than work accidents reduction.
A metric allowing you to know precisely if you’re doing better is crucial in the research phase, where you’re looking to prove or refute hypotheses in the most efficient way. If you achieve 65% accuracy with a model where you invested a lot of time and team effort, it may make sense to switch development to another model that achieved 50% accuracy in two weeks.
At its core, AI is a science and a research topic, where you iterate and where it’s impossible to predict with certainty successful results. With 1 or 2 people per project, working for 2 or 3 months before assessing results, you can minimize the risk of a project to an acceptable level. With a set of deadlines, a predefined budget, and regular communication of tasks in the form of flash reports and other means, you’re also giving yourself the means to know when it’s advantageous to change the research scope and to refine the hypotheses as research progresses.
Some specific stages must remain sequential. Such is the case with the feasibility assessment. Instead of advancing on a project for 12 weeks and risk working for nothing in case of failure, make a feasibility assessment in 2 weeks to estimate how likely the project would be successful in 10 weeks, based on the pre-requisites and your readiness in terms of data amount and quality.
In practice, you define goals, set success criteria, estimate a ROI, and cross-pollinize the knowledge of people of different backgrounds with the collected data. You’ll generate additional hypothesis to test in parallel and will be able to build a plan - where to iterate, and where to investigate deeper into what you really need to solve your business issues.
Outsourcing data science may be useful as a first step if your company isn’t mature enough, especially if you negotiate for knowledge transfer by the vendor. In the long term, for maximum ROI and autonomy from vendors, you must build deep AI expertise internally. That’s why this knowledge transfer should be wide-ranging, and can include transparent models that can be inspected, insights gathered, descriptions on what was learned during the project, the methods they used and why, etc.
Apart from being able to conduct successful projects faster and safer, there’s a profound and lasting interest in harvesting internally the AI and data science knowledge accumulated during projects, through your own knowledge building or through knowledge transfer from vendors. In niche markets, it is advantageous to be the expert in your field, including in technological innovation. AI requires to be crafted to a specific set of problems, and the knowledge that you capture in building this AI is certainly a competitive advantage, and a competence that you’ll be able to offer through services.
On the Wizata platform, the Cases module allows you to assess the ROI of your AI and intelligent solutions initiatives, by combining priority management with a Loss & Savings estimation tool customized to quality, maintenance and performance problems.
Using a platform such as Wizata to orchestrate, centralize and manage all your AI, data science, knowledge and digitization initiatives also allows you to build on previous expertise and eases synergies within your team and across time, which can be an important factor in industries where a qualified workforce is lacking. Such a tool at the heart of your R&D also serves as a catalyst to foster collaboration, facilitates knowledge transfer, and the sharing of best practices sharing to all relevant stakeholders.