Posting on Substack seems to require three elements to be completed before submitting the post: 1) enter title, 2) enter subtitle, and 3) some type of content. They have not offered any type of content length guidance. I guess you can write as much or as little as you want. Right now I’m working on my first Substack post and have drafted the three required elements for that post. I’m going to try to keep these posts at a conversational level with a bit of depth, but that is going to require some editing. My focus is on reworking the content for the new forum. I’m going to publish the content here on the weblog and on Substack. The inside look of how the content was created and any changes or modifications will live here. It feels like Substack is a very fleeting medium that has maximum engagement at the moment of publication. We will see what happens throughout the next 12 weeks of content creation. That will either be true or some type of momentum will build up along the way.

Here are the details of my first Substack publication build out…

Enter title… Machine Learning Return On Investment (MLROI)
Enter subtitle… A brief look at how understanding ROI helps unlock ML use case success

Content: 292 words starting and ~750 words as a finished product

Be strategic with your ML (machine learning) efforts. Seriously, those 6 words should guide your next steps along the ML journey. Take a moment and let that direction (strong guidance) sink in and reflect on what it really means for your organization. You have to take a moment and work backward from building strategic value for your organization to the actual ML effort. Inside that effort you will quickly discover that operationalizing ML efforts to generate strategic value will end up relying on a solid return on investment plan. Taking actions within an organization of any kind at the scale ML is capable of engaging without understanding the potential return on investment or potential loss is highly questionable. That is why you have to be strategic with your ML efforts from start to finish.

That means you have to set up and run a machine learning strategy from the top down. Executive leaders have to understand and be invested in guiding things toward the right path (a strategic path) from the start. Make an effort to just start out with a solid strategy in the machine learning space. It might sound a lot harder than it is in practice. You don’t need a complicated center of excellence or massive investment to develop a strategy. Your strategy just needs to be linked to the budget and hopefully to a budget KPI. Every budget results in the process of spending precious funds and keeping a solid KPI around machine learning return on investment levels will help ensure your strategy ends on a strong financially based footing for years to come. All spending should translate to some key performance indicator of some type. That is how your result will let you confirm that the funding is being spent well and that solid decision making is occurring. You have to really focus and ensure that all spending is tied to that framework when you operationalize the organization’s strategic vision to be aligned financially to the budget.

That means that the machine learning strategy you are investing in has to be driven to achieve a certain return on investment tied directly to solid budget level key performance indicators. You might feel like that line has been repeated. If you noticed that repetition, then you are paying attention and well on your way to future success. That key performance indicator related tieback is only going to happen with a solid machine learning strategy in place. It has to be based on prioritizing and planning for return on investment. Your machine learning pipelines and frameworks have to be aligned toward that goal. That is ultimately the cornerstone of a solid strategic plan when it comes to implementing machine learning as part of a long term strategy.

Be ready to do things in a definable and repeatable way. Part of executing a strategy with quality is doing things in a definable and repeatable way. That is the essence of where quality comes from. You have to know what plan is being executed and focus in and support the plan in ways that make it successful at your desired run rate. In terms of deploying machine learning efforts within an enterprise you have to figure out how the technology is going to be set up and invested in and how that investment is going to translate to use cases with the right return on investment.

Know the use case instead of letting solutions chase problems. Building up technology for machine learning and then chasing use cases is a terrible way to accidentally stumble on a return on investment model that works. The better way forward is to know the use cases and have a solid strategy to apply your technology. That means finding the right ML frameworks and pipelines to support your use cases in powerful ways across the entire organization.

This is a time to be planful. Technology for machine learning is becoming more and more available and plentiful. Teams from all over the organization are probably wanting to try proof of concepts and vendors are bringing in a variety of options. Both internal and external options are really plentiful. It is an amazing time for applied machine learning. You can get into the game in a variety of ways rapidly and without a ton of effort. Getting your implementation right and having the data, pipeline, and frameworks aligned to your maximum possible results involves planning and solid execution.

Your ML strategy cannot be a back of the desk project. You have to be strategic. It has to be part of a broader strategy. You cannot let all of the proof of concepts and vendor plays drive the adoption of machine learning technology in your organization. That will mean that the overall strategic vision is not defined. It happened generally because it might have a solid return on investment and the right use case might have been selected by chance from the bottom up in the organization. That is not a planful strategy.

Know the workflow you want to augment with ML and drive beyond the buzzwords to see technology in action. You really have to know where in the workflow and what pipelines are going to enable your use cases to provide that solid return on investment.

At some point along the machine learning journey you are going to need to make some decisions…

Q: Where are you going to serve the machine learning model from?
Q: Is this your first model build and deployment?
Q: What actual deployments of model serving are being managed?
Q: Are you working on-premise for training or calling an API and model serving in your workflow?
Q: Have you elected to use a pretrained model via an external API call?
Q: Did you buy a model from a marketplace or are you buying access to a commercial API?
Q: How long before the model efficiency drops off and adjustment is required?
Q: Have you calculated where the point of no return is for model efficiency where ROI falls below break even?

What’s next for The Lindahl Letter?

Week 2: Machine Learning Frameworks & Pipelines
Week 3: Machine learning Teams
Week 4: Have an ML strategy… revisited
Week 5: Let your ROI drive a fact-based decision-making process
Week 6: Understand the ongoing cost and success criteria as part of your ML strategy
Week 7: Plan to grow based on successful ROI
Week 8: Is the ML we need everywhere now?
Week 9: What is ML scale? The where and the when of ML usage
Week 10: Valuing ML use cases based on scale
Week 11: Model extensibility for few shot GPT-2
Week 12: Confounding within multiple ML model deployments

I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review.