Transforming your organization with the Machine Learning Canvas
If you’ve had a look at the Machine Learning Canvas (MLC), chances are that you are also interested in the ML transformation of your organization, and that you’re wondering how to best facilitate it and mobilize the right resources…
You might be wondering how the MLC fits in this transformation? Here’s my answer in 4 steps, based on the AI transformation playbook published by Andrew Ng a couple of months ago, and followed by a short survey!
1.EXECUTING PILOT PROJECTS. Your very first objective should be to create momentum. Some ML systems can be particularly challenging to build, so Ng recommends picking your first projects wisely: they should be technically feasible and show traction within 6 months — not years! You can (and maybe should) involve external partners for that, to help assess feasibility, design and implement your ML system. The key point at this stage is not to build an internal ML capability, but to make one team in your organization more successful thanks to ML, and thus to create momentum. The MLC is a great help to better specify your initial ideas, formalize the first ML task(s) to tackle, align with a good value proposition, and get a more precise idea of the feasibility and potential impact of your ML system.
2. BUILDING TEAM. This should only be of concern once you’ve had success with pilot projects. Ng focuses his recommendations on larger enterprises with a market cap from $500M to $500B, with many potential applications of ML across divisions and business units; for these, it can be more efficient to build an in-house ML team that helps the whole organization execute new projects, by hiring the right people: ML Engineers, Data Engineers, Data Scientists, ML Product Managers. They should champion usage of the MLC across the organization. One of the tasks of this team will be to build an internal platform that centralizes ML assets (datasets, pipelines, feature stores, configs, models, APIs) and makes them accessible across the company. Of course, this is costly, and it requires buy-in from the C-suite — which is one more reason to demonstrate value from pilot projects as a first step.
3. PROVIDING BROAD TRAINING. To me, there are two points here:
- ML talent is hard to find, so you want your existing team to gain new skills in practical ML. Besides, building a whole new ML team might not be the best strategy for smaller organizations: training current employees is the alternative.
- You can’t leave it up to engineers alone to do all the work. The next step in the ML transformation will be to develop and execute an ML strategy, so you also need a certain understanding of ML at the executive level. Execs will need to make appropriate resource allocation decisions and to collaborate smoothly with those leading and supporting the organization’s ML projects. In larger enterprises, division leaders will also need to be able to set direction for ML projects, monitor and track progress, and make corrections as needed to ensure successful project delivery. You should share the MLC book with everyone, so they can start learning more!
4. DEVELOPING STRATEGY. This one requires the most experience with ML. Hopefully you will have gained some from steps 1–3, but you can also leverage the experience of outsourced partners. You want to turn your ML systems into key assets for your organization. This requires alignment with long term business strategy, Objectives and Key Results — which the MLC helps with. Your organization’s data strategy should be at the heart of its ML strategy. Data collection is paramount to keeping your models relevant, high performing, and valuable. Using the MLC to specify strategic ML systems will make it much clearer which data you need to acquire, and where to invest for that. Ng says that he has “tragically seen CEOs over-invest in collecting low-value data”…
I’d like to point you now to the super short survey I’m running:
How are you participating in your organization’s ML transformation?
Please let me know 🙏 I’m looking forward to the results!
Cheers,
Louis
