Over the past months, ÉcoWatt has grown from a vague discomfort with how "energy efficiency" is measured into a much clearer vision of what we want to change, and for whom. We started with open data and a technical challenge, but quickly realised that numbers alone don't help people decide whether to put on another sweater, change their subscription, or renovate their building. The DPE label, Enedis data, public simulators and existing tools all provide pieces of the puzzle, yet our interviews showed that homeowners and renters mostly feel lost, under-informed, and driven above all by cost rather than climate alone. This tension between complex systems and simple, human decisions became the heart of our project.
As a team, our journey was anything but linear. We moved from military logistics to neighbourhood sharing, and only then to the DPE–Enedis challenge. Integrating a new team member, working in parallel during bootcamp, and repeatedly changing direction forced us to communicate more consciously and to accept not knowing the answer in advance. Each of us brought different strengths, data work, user research, facilitation, storytelling, and the project only made sense once we learned to let these complement, rather than compete with, each other. The early frustration of "too many ideas, no clear user" gradually transformed into a shared compass: understanding how the way we measure energy influences how people consume it.
Our research phase confronted us with the limits of open data as much as with its possibilities. We learned that we cannot track a single household across datasets, privacy protection is real, but we can reveal meaningful patterns at neighbourhood level. We also saw how strongly people's behaviour is shaped by constraints: the copropriété that must decide together, the renter who cannot renovate, the landlord who reacts to regulation, the agency caught between buyers' emotions and technical labels. Mapping these stakeholders and building our personas of Michelle and James helped us see that "the user" is not an abstract citizen but someone trying to keep their home warm, their bills manageable, and their decisions understandable.
ÉcoWatt emerges from all this as more than a dashboard idea. It is our attempt to translate messy data and opaque labels into something that speaks the language of everyday life: "Is my consumption normal?", "What would actually change if I renovated?", "What can I swap today to save money without freezing?" Our MVP is still an early sketch, and the data work ahead is substantial, but we now have a solid foundation: a clarified problem, concrete user groups, and a unique value proposition that combines real consumption, habits, and DPE structure in one place. Going forward, our challenge will be to keep that dual commitment, to analytical rigour and to human simplicity, so that ÉcoWatt remains not just a clever use of open data, but a tool that genuinely empowers people to see what they can swap, and why it matters.
Looking ahead, our next steps are to finish analysing the data in depth, refine how we merge and compare DPE and Enedis information, and identify a robust model that can generate reliable benchmarks and recommendations. This means moving from a conceptual MVP to a working engine that can:
Strengthening this analytical backbone will allow us to test our assumptions with real users, iterate on the interface, and move closer to an ÉcoWatt prototype that genuinely helps households see what they can swap, why it matters, and how to act on it.
The second deliverable marks a shift in what our project is. Our first deliverable established the problem and the people, our second one has been about building a working data model, a tested prototype, and a project that can grow.
The data analysis produced a machine learning model that predicts DPE labels from five inputs: size, energy indicators, building type, construction period, and heating energy which makes it reach 73% accuracy. That number is meaningful, but the more important finding came from the process of building it. Working with the ADEME datasets revealed that the real constraint is not computation but data quality. Too much information is missing, inconsistently recorded, or incomparable across years for any model to confidently assess whether DPE labels reflect reality.
This shifted the goal of the project. The model is purely a starting point and not yet the final product. The Supabase infrastructure built alongside it is designed to collect the information that is currently absent: each time a user completes the ÉcoWatt questionnaire, structured data about their home is added to a growing dataset. The more the platform is used, the more credible the analysis becomes. The product and the data work are not separate tracks, they rather feed each other.
We had twenty-one users test the two versions of the prototype across two rounds that encouraged us in how we would advance. It showed us that visual design was never the problem, and it still isn't as 92% of wave 2 users liked it, and 100% found it easy to use. The main problems we found were always about language and trust. Wave 1 showed that structural confusion was the dominant failure: users did not know what DPE was, could not interpret the output, and assumed the tool had broken during the loading state. All of those were fixed between the rounds.
Wave 2 showed that fixing structural confusion came down to fixing terminology. 64% of users hit walls on specific words in the questions, and the gap between ease of use (9.23/10) and perceived value (7.61/10) revealed something important: users could navigate the tool smoothly but at the same time still not be convinced enough to act on what it told them. So far we have mainly fixed the problem of terminology with explanation and informative boxes but ease and trust are not the same thing. That gap is what wave 3 needs to close and what we are working on.
The two tracks converge on the same conclusion: ÉcoWatt works as a concept and as an experience, but it has not yet earned full trust from our users or from the data. Users find it easy to use and mostly understand it, but are not yet certain they believe it and could act on the suggestions. The model is accurate enough to be useful, but the underlying dataset is not yet rich enough to be definitive. Both of these are solvable problems, and both point in the same direction of more data and more iteration.
The next phase is therefore not about building something new. It is about deepening what exists. We will work on refining the language of the prototype, expanding the dataset through real usage, and running a third round of testing. This round will be focused not on whether users can complete the flow but on whether they trust the output enough to change something about their home.
Our project was fundamentally shaped by the "Energy Performance Diagnoses" challenge proposed by Enedis. The objective was to bridge the gap between theoretical Energy Performance Certificates and real-world electricity consumption. This challenge served as our research compass, forcing us to move beyond abstract ideas and confront the reality of data matching and predictive modeling.
Participating in this challenge required us to balance high-level technical requirements with human-centered design. The open data challenge was a purely data-centred challenge, forcing us to go beyond what was required for the PBL course. Oppositely, the design-thinking focus from PBL allowed us to deliver a project that went deeper than "just" the data analysis, with thorough reflection, testing and critical thinking behind every decision. We were forced to reconcile the technical difficulty of cross-referencing DPE databases with real consumption data while simultaneously addressing the empathetic requirements of design thinking that we learnt from seminars. In our opinion, this refined our project into a robust one, where we learnt plenty of both sides.
This technical friction became a vital part of our meaning-making process. Instead of viewing data gaps as failures, we used them to refine our user personas, distinguishing between owners who can renovate and renters who must rely on habit changes. We learned that technical accuracy alone does not drive change. This realisation ensured our model wasn't just a statistical exercise, but a functional tool for social impact that translates complex conventional calculations into clear, actionable energy advice for real people.
While we were not ultimately selected as finalists for the Open Data University challenge and hackathon, we chose to view this not as a loss, but as an opportunity for reflection. The challenge provided us with a clear aim to work towards and made us motivated. ÉcoWatt remains listed on data.gouv.fr, the national open data platform, where it has already reached over 600 views. Hopefully, the project can be of value to others as a public tool. Additionally, as our process is transparent through this website, we hope future students may find it useful as well. Regardless, it is a clear indicator that our project is indeed of interest, validating our viability beyond academic evaluation.
Reflecting on this process, we recognised how our diverse backgrounds in politics, economics, and data science shaped our perspective. Some of us went into this with no experience interpreting this type of dataset, let alone creating something with it. The challenge demanded that we collaborate to gain a shared understanding and figure out what we were each individually good at. We stopped seeing data as an abstract asset and started treating it as a tool for social and environmental impact.
This journey has fundamentally transformed how we approach problem-solving. We have developed a future orientation where technical choices are inseparable from their social consequences, and where technical rigor and empathy for the end-user cannot be separated. We emerged not just as students who completed an assignment, but as better designers who understand that the value of a tool lies in its ability to bridge the gap between measured reality and the human experience.
ÉcoWatt is now on the way to becoming a real-world application ready to help users navigate the ecological transition. With a transformed business model, added inclusion of renovators, a marketing plan, and a proven viability, we are on the right path to deployment.