Analytics are a significant enabler to get high value in the manufacturing industry. But how can you prove that you deliver value?
According to Gartner, more than half of all analytics projects fail because they are not completed within budget or on schedule, or because they fail to deliver the features and benefits that are optimistically agreed at the outset.
Based on my International experience, the approach and methodology works the same way everywhere.
Here are eight crucial tips to help you avoid misleading the project with poor problem forecasting and solving.
1. Start with the right business question and get started
Many analytic projects fail because the right business questions have not been discovered. For instance, analytics might discover that a small change within a process leads to much more efficiency, and save the company millions. This justifies the implementation of analytics, machine learning or predictive maintenance. Therefore, the very first step in applying data analytics successfully is a deep dive into business questions. If you have the right question – even if it seems small – it will be much easier to get your project started on the right footing.
2. Define the ‘why’ to help understand the ‘how’
After collecting the business questions to be answered by analytics, ask yourself: Is the right data available? Am I allowed to make use of the available data? Is this viable as well as feasible?
Once the business pain points and potential business improvements are determined and validated, the project must get kicked off. Together with a business consultant, goals, budget and a clear roadmap must be lined up. Determine clearly how you are going to measure success, in cooperation with your project sponsors, as well as the operations, IT and finance teams.
3. Dig below the surface and address the root cause
I have recently been working in the plastic molding injection industry, and our customer wanted to predict failures. Accurate predictions would enable the redeployment of workloads to other presses in the unlikely event of a loss of production. An unexpected failure to a heavy mechanical or hydraulic part could impact machine availability for days, as faulty parts must be sourced and replaced.
Nobody could understand the cause of one particular faulty part. It seemed the fault itself was a consequence, rather than the root cause of the machine failure. That faulty part was changed, but the signs of a new failure were still unknown. However, over time a drop in hydraulic pression, increased axis vibration and overheating components were all found to be indicators of an imminent failure. New IoT instruments could be added to monitor these new factors, and avoid intrusion in the press. The increased data was then used to better understand equipment behaviour and boost prediction accuracy.
4. Make sure you have the right methodology
There are no good or bad methodologies, as every one of them has its own strengths and weaknesses. Ultimately, no single methodology works across every business model and industry. One size does not fit all.
Consider your resources, the industry, and the kind of projects you will be managing before deciding on a methodology. I am in favour of brainstorming ideas from all team members, but also with business consultants in the same industry from other countries. Checking the feasibility of any methodology considering time to operationalization versus cost is very important. A simple pilot is relevant to validate a concept based only on data files. Start small and grow bigger.
5. Put the right team in place
Roles have to be clearly determined and contracted, with both organizations ensuring any proposed methodologies will capture success. KPIs must be contracted to make sure we reach or exceed the success criteria.
Ideally, a system architect, project manager, data scientist and a business consultant are paramount for success.
On the other side of the table we need to have subject matter experts, an IT engineer and a sponsor, whose job it is to carry any roadblocks that might occur along the course of the engagement up to the leadership.
6. Design and implement
Compliment it with a powerful analytics and machine learning software, math functions and stats.
With everyone aware of their role, the roadmap created, ROI defined, and tools selected, now it’s time for the difficult part. How do you ensure your hypotheses are valid, and that small animals in the room aren’t hiding the elephant?
Try other angles, challenge the status quo, use graphic interfaces to confirm findings. Cross-compare different methodologies to see if you achieve similar results.
7. Manage and get results
Once the pilot works fine, prove that you have delivered. Competitive advantage comes from capitalizing on uniqueness. Every client is different, but not all organizations have the potential to exploit that uniqueness in a way that no one else can match.
8. From buzzwords to business benefits
How did it help me achieve my objectives? Is the ROI there?
According to Forbes, “Enabling condition monitoring processes that provide manufacturers with the scale to manage overall equipment effectiveness (OEE) at the plant level can increase OEE performance from 65% to 85%”.
Again, there is no ‘one size fits all’ across industries. For some, even a 1% saving could make a significant difference.
Analytics, machine learning and predictive maintenance are key to successful transformation in the industrial sector. Success is on hand if we manage to use a clear methodology, with great teamwork and a powerful engine.
Want to learn more about how to successfully apply analytics in your business? Join our Teradata Universe Event.