So, by means of small design experiments, sub-systems of your use case can be tested in order to assure validity of the outcome based on high quality data. It is nearly impossible to do the systems engineering part first time right; testing and validating shows that, the majority of the time, things in practice don’t go the way you expected. In our example we found that at one line, basil leaves were added to the pizza, causing an error in the cheese detecting algorithm that was only focusing on yellow pixels in images. Our solution: have an agile approach, respond quickly, and test again.
The final step of solving your use case is to embed it in your daily processes. When you got all systems in place, your data is ingested and meets all requirements, your data scientists modeled the solution, and the output of the model is embedded into an app: then the result needs to be put into practice. Looking at the pizza-case, the final setup (camera, cloud, dashboard, PLC adjustment) had to be implemented at the pizza belt and the factory workers needed to be trained in a different, less manual way of working. Embedding the solution in your daily processes means tying everything together, making your solution part of the factory instead of being a tool on top of the factory. This step is critical for the success of your use case. After all, a solution has to be used in order to be useful – make things productional!
To become data-driven is, as mentioned before, not something that can be achieved overnight. Change management requires a shift in culture and ways of working; a process that takes time. There are many use cases to execute, requiring a multitude of competences. It is a long-term investment that can only be fully redeemed if your multidisciplinary team works well together and maintains the balanced focus of business, engineering, and operationalization.
Turning data into value is not a one-off project, but a journey that requires hard work, persistency, and a data factory. Smart use of data offers numerous opportunities, such as: predicting maintenance, increasing machine efficiency, or improving production quality. Utilizing these opportunities to their full potential requires a step-by-step approach, boiling all data opportunities down to small sub-projects: data use cases.
To successfully embed such a data use case into day-to-day business, it needs to be executable in a short period of time (preferably 4-8 weeks), have C-level support, and add direct value to the business. Take a three-step approach for this: understanding and narrowing down the business case, engineering the system, and operationalizing your outcome in production.
1. Narrow down the use case
In many companies, tons of data is generated, providing the input to automate, predict, or optimize business processes. With a large collection of data opportunities to choose from, the trick is to make it as small as possible; and to select the most viable use case that directly adds value to your business, and can at the same time be implemented in a short period of time (4-8 weeks). Do not aim at far-reaching goals (‘solving world hunger’), but narrow down to a small, explicit and concrete goal. Below is an example of how we narrowed down a top-level data opportunity into a use case.
Pizza case
Top-level data opportunity:
Reduce costs in the production process
Zooming in:
Prevent waste of having too much topping on pizzas
Refine:
Cheese is the most expensive ingredient, so let’s start with this. It is distributed by means of a cheese waterfall. The pizzas lay on a belt, transporting them underneath this waterfall. If the belt runs faster, the cheese topping decreases. If the belt runs slower, the cheese topping increases.
Use case:
Optimize the belt speed in order to get the perfect cheese distribution
So, with many potential use cases at your disposal, where do you start? The question of how to prioritize is often difficult. One way to go is to just start and see where you end. Our preferred way is a more structured approach: create a backlog with narrowed down use cases, score them on a number of features, and select the use case with the highest ranking. Interested to read more about this approach? Check out our article on selecting use cases.
However, the steps described above will not work without a driver; a committed C-level sponsor who believes in the value and steers the project. This sponsor needs to have the vision of changing the company by smartly using data and systems and transforming ways of working, not of steering a one-time project.
2. Engineer the system & data
By choosing your use case, you defined an end goal. In the example above, this end goal is to set the optimal belt speed for cheese distribution. The next step is then to make an overview of what is needed to achieve that end goal. We can provide insight into the required data and systems by means of systems engineering (see video below). This implies designing what systems and techniques are needed for each step in the process. It gives you an end-to-end overview of everything required for your use case and allows for making a detailed planning.
Joost Meijer on systems thinking - a different approach to get results
It is then a matter of acquiring all data and making it available. Some data will be neatly digitalized and pre-processed already, some will be only available offline, and some might not exist yet. In case of the cheese distribution use case, the machine data was already available, but a camera setup to accurately determine cheese-density still had to be built. Step by step, we built this setup and iteratively improved it to reach the accuracy we aimed for.
Back to overview
