|
Hi, I wrote a bit about this a bit in a twitter thread[0]. Here's another reply you might find useful for development and digging a bit deeper instead of implementing every feature request[1]. We're a boutique consultancy specialized in machine learning. Given the projects, our clients are exclusively large organizations. Repeat business is frequent and we amortize on the relationship we built with them. The clients are from the CEO leveraging his network. During the past two years, we systematically refined our process to qualify prospects, properly onboard the client by making how we do business clear [what's ML, what it can do, what it can't do, what are the pre-requisites, necessity to have support from top management, necessity to have those we will build things for present at the table from day one, necessity to have a cadence and regular meetings to check progress and eliminate divergences as soon as possible not to build the wrong thing, getting requirements right, starting small with prototypes, getting to the job to be done and solving problems]. The people involved in a project are in the meeting on our side, and we nudge and insist that the people involved in the project, especially their domain experts and those who will use what we'll build be present. We have scars working only with executives insisting they knew best. We never want to build something that's not used, even if we're paid to do it. That's a waste of life. We have ruthlessly and consistently refined this and we've generated more revenue with 6 people (and only three working on the projects) than we did with 17. The others are working on our product and we don't want to disturb them. The product is our MLOps platform[2] that we use, precisely, to deliver value to our clients consistently and eliminate toil. Setting up environments (fresh collaborative notebook environments), scheduling long-running notebooks, experiment tracking, model deployment and monitoring, etc. This is the "abstract" section of the twitter thread: leveraging the experience curve of all the projects we have done in the past. Post-mortems on projects that have failed, why, how, when, etc. - [0]: https://twitter.com/jugurthahadjar/status/131066829330549965... - [1]: https://news.ycombinator.com/item?id=26075799 - [2]: https://iko.ai |
Your process makes a lot of sense. Our problem is defining what you already have - "we're a butique consultancy specialized in machine learning."
I see you put a lot of work into it, it's amazing, congrats! Curious to hear how long you take the contracts for?
I believe that we are aiming for something similar to establish. I totally believe that studio like this can delivery much better work with less people that are well played together and are really good at what they do.
Again, thanks for sharing the story!