‘Failures’ can be progress, too

The Hydroinformatics knowledge exchange meetings are organised as part of Waterwijs, the joint research programme of Dutch and Flemish water utilities. One of the principal aims of these meetings is to exchange practical experience and knowledge during theme-based meetings.

Intelligent failure and sharing negative results

Geertje Pronk (KWR) kicked off the meeting with a review of different types of ‘failure’. Simple (you do something wrong), Complex (there is a combination of circumstances) and Intelligent (deliberate and meticulous experimentation). The Waterwijs ‘Exploratory Research’ programme provides opportunities to explore topics where the outcome is uncertain. In the programme, a decision not to go down a particular road on the basis of sound reasons is also a valuable result. Each exploratory study results in a report. Pronk emphasised that publishing negative results is very important so that multiple investments are not made in something that doesn’t work.

Image 1: Different types of failure.

AI: start small and simple… on one of your biggest problems

Alex van der Helm (Waternet) took us through a number of lessons learnt in data science projects that many participants recognised. What works well is to move ahead in steps with the ultimate goal in mind. It is a good idea to start off by tackling a small but essential problem with AI, rather than the complex whole. Another tip from Van der Helm was to get the basics right first before implementing a data science project. For example, if there is no automatic data flow for a data-intensive project, that will take a lot of time and become an obstacle to successful implementation. So start working first on another prioritised project where the data flow is already in place because the main value of data science projects is often to be found in automatic continuous use.

Reusable

Another tip from Van der Helm is to produce reusable data and code components with version management. It takes more time but, even if a project is not a success, you can use the components for subsequent projects. It is also a good idea to cast your net wide to learn about any plans inside the organisation so that a project will not be incompatible with future developments.

Image 2: When things don’t go as expected, there is often valuable information to share.

All disciplines needed for success

Finally, Van der Helm pointed out that colleagues all have their own expertise for the development of data science solutions. For example, a data scientist is not a machine learning (ML) engineer, and an ML engineer is not a product owner. You need all of these areas of expertise for successful implementation. Understanding this can prevent people involved from becoming demotivated: that is a shame and it is preventable.

Gap between model and practice

Peter van Thienen (KWR) took participants through a number of projects relating to the numerical optimisation of drinking water and sensor networks using KWR’s optimisation tool Gondwana. These projects went differently than expected, and we were able to learn from them. There was sometimes a gap between the model and practice. It was challenging to define the problem with end users. Other challenges were that the network models provided were not always complete, that there were additional circumstances that had not been taken into account, or that relevant data were difficult to retrieve or not formally recorded. A lot of data were needed from the drinking water utilities. Using the lessons learned from the initial projects, a more interactive working method was developed. The result was that Gondwana was eventually applied at and with nine drinking water utilities where the results of the optimisation calculations are now being applied in practice. Encouragingly, Gondwana has been nominated for the 2025 Waterwijs Implementation Award.

Image 3: Bridging the gap between theory and practice.

A game of failure and success

It is important for organisations to deal well with the fact that a project can fail. Colleagues should certainly not feel that they should stop trying: ultimately, that is not a sound basis moving forward. Tessa Pronk (KWR) made this clear to the participants with a strategy game based on game theory. In the game, the participants worked in a start-up and they could make annual decisions about developing innovations or continuing to do their day-to-day work. It was interesting to see that the actions of colleagues as individuals affected all their outcomes and the start-up.

‘Failures’ always have benefits

As always, the session ended with a discussion. One of the things to emerge was that ‘failure’ is, in part, also a question of perspective. Has something not worked out, or do you call it a failure? Or can you just change your goal, and does it work then? The participants felt it was particularly important to note that, in every failed project, you do move ahead: specific components are often very successful. There is a learning process that you would otherwise miss. However, it is important to shut down projects as early as possible when it turns out that something is really not possible. That can save a lot of expense. It was also suggested that a project can be successful in principle but that it can still fail if it does not ‘land’ with the right people.

The meeting gave the participants the opportunity to think about the fact that not everything always goes as planned but that this can also be valuable, especially when an ‘intelligent’ failure is involved.