I’m a younger brother, so I learnt to lose as soon as I learnt to crawl. I played in a youth futsal team, and there was one year when we lost ALL our games. This overexposure to defeat made me realise that I don’t need to win to be happy, but I do need a context that allows me to chase victory.
Don’t take yourself too seriously.
I have never taken myself too seriously. This way of thinking has made me very aware of my failures. I didn’t hide them, but I didn’t broadcast them either. Now I’m at a stage where I think I have to be proud of those failures. Maybe it’s because I’ve started listening to punk again; maybe it’s because I reject the false perfection on display on social networks. Punk is, after all, a cultural response to the social context, so maybe the two are very much linked.
LinkedIn, you are a very powerful professional tool, but social networking dynamics suck.
I wrote this post a while ago, but I recently discovered the concept of the failure resume, so I’ve changed the title from My Biggest Fuck-ups to My Failure Resume, and I’m going to start with my most resounding failures both as a data scientist and in my most recent management position. Here they are:
Top DS Failures:
- Inability to reproduce the code and get the same results in a research paper as part of my thesis. I learnt the importance of using version control, saving intermediate results and managing development environment dependencies well.
- Negative valuations in a production property valuation model. Yes, I could do that! That was a big relief for the people in my team. I learned the importance of model monitoring, how to analyse model behaviour when inputs take extreme values, and how to be wary of tree models.
- Putting the engineering team in a sweat. I had to cluster users’ geolocations and thought DBScan was a good fit. I did not think about the production release and availability of the library in Spark. As a consultant, I was asked to solve a problem and didn’t ask about the technical context, volumetrics or anything else.
- Business intrusion. In more than one project, I didn’t know how to deal with business constraints. We delivered data of dubious quality, which reduced our credibility, and also created a huge technical debt because we were not allowed to version the reports, so you had to carry one of them around with you historically.
Top management failures:
- Giving too much responsibility without enough support. I asked junior people in certain roles to take on responsibility without giving them sufficient support, setting them up for failure almost by design.
- Not making sure they got the message.
- One of the most important things you can do as a manager is to give feedback. On a number of occasions I have sinned by not making sure that the message I was sending was fully understood.
- Assuming that stakeholders had realistic expectations for AI projects.
- Under-communicating the engineering risks of AI projects to stakeholders.
- I make jokes when I shouldn’t. Sometimes I’ve been a loudmouth and made team jokes that didn’t go down well with certain people.
- Dedicating too many resources to projects that have not been thoroughly analysed.
This is my CV of mistakes so far, I’m sure I’ll soon be able to add more medals to wear on my lapel.
This post was last updated on Saturday the 25th of October, 2025.