Home  >   Blog  >   What I Think About When I Talk About ML Product

2020-08-29

What I Think About When I Talk About ML Product

  Support by donation   Gift a cup of coffee


This article is part of the series: How to Talk About AI

Everyone wants to leverage machine learning (ML) in their daily work, and I'm sure they will be excited about your awesome ML product whatever the detail is. But, what do we, as a product developer, really have to think about when we create ML products?

Based on my experiences and what I learned from the following insightful articles, let me summarize five key questions we should ask ourselves.

1. Do we have a clear mapping between business and technical problems?

Regardless of whether it's ML-based or not, our product should translate high-level business requirements into specific technical problems.

Since “AI/ML” sounds exceptionally attractive but cannot solve all the problems in the world, making sure the connection is particularly important.

For example, if our business requirement is to improve product margin, a technical problem our product handles is solving an optimization problem to find an optional product price.

On the other hand, the following scenarios easily lead us in the wrong direction.

“We want to use AI to grow our business” — A business requirement isn’t clear enough. What metrics are you tracking? What is your measurable goal?

“We want to produce 2x more units in a single factory by using ML” — Inappropriate choice of technology. In fact, ML might bring an innovative solution (e.g., the brand-new architecture of factory machines designed by ML algorithms), but it’s still unclear if that’s the case for this type of hardware/machine-oriented business requirements.

2. Is using ML for solving the problems cost-effective?

Similar to the other product features, our first step is to understand customers and their problems.

Once we identified a set of problems, one important question has naturally arisen: Which problem(s) could ML solve?

Finally, we must justify cost vs. performance gain when we apply ML to these problems: Is solving this problem by ML cost-effective? What alternatives we can think of?

It is important to note that ML is generally expensive when we undergo an end-to-end development lifecycle from data collection and preprocessing to training and evaluation; more human resources, as well as computing power, are surely needed over a long period of time. If you come up with a much simpler solution, try it first of all.

3. How quick can we have “MVPs”?

Compared to the other engineering problems, the development of ML-based products tends to be longer (and even be a never-ending story) since continuous evaluation & optimization is the core of a successful ML solution.

Consequently, it’s easy to spend an infinite amount of time on building “MVP”, which won’t be “minimum” anymore.

We must pay special attention to building an end-to-end solution as quickly as possible in the shortest path. (Of course, we can say the same thing for all the product features, but ML, in particular.)

Here are some examples of what to do at a minimum:

  • Collecting a reasonably small set of data before building a complete data pipelines.
  • Running Exploratory Data Analysis (EDA) for understanding a big picture of the data.
  • Starting from the simplest model (e.g., linear model) with offline evaluation.

Notice that answering Questions #1 and #2 helps you to find an appropriate approach you could take.

4. How do we ensure humans in the loop?

It depends on a target persona of your product, but, since running a feedback loop is a must for ML, the product should have a way of providing human feedback in the end-to-end pipeline, even if the feature is designed as an out-of-the-box packaged solution.

As we covered in Question #1, the ML solution must be strongly tied to business objectives, and hence customer needs to be able to measure the quality of ML-based outcomes and be confident on it. Otherwise, ML models guide their business in the wrong direction and cause a serious problem in some cases.

That is, in order to make your product reliable and usable for end-users, a fully automated solution won’t work, and leaving a space for human interactions in an appropriate way is highly important. It doesn't matter if the model uses state-of-the-art techniques and shows the best accuracy in a lab setup.

5. Why should customers use your ML solution?

Nowadays, there are so many options to leverage ML in the customer’s business process:

  • Human
    • The in-house data science team
    • Non-experts who learned the basics from a book/online course
    • Consultancy
  • Tools
    • Jupyter
    • Python
    • Numpy
    • Spark
  • Infrastructure
    • AWS
    • GCP
    • Azure
  • ...

Hence, unless there is a strong competitive advantage in our solution, our customer doesn’t necessarily have to choose it.

Examples of competitive advantages include:

  • Strong professional service team supporting ML model deployment.
  • A domain-specific model built by unique private dataset.
  • Seamless integration with user's daily business process and external tools their business relies on.

Bottom line

While offering no ML features definitely has a negative impact on the modern tech businesses, making the functionalities worthwhile to customers is rather difficult.

Therefore, I want to carefully ask the questions to myself when I think about the ML product:

  1. Do we have a clear mapping between business and technical problems?
  2. Is using ML for solving the problems cost-effective?
  3. How quick can we have “MVPs”?
  4. How do we ensure humans in the loop?
  5. Why should customers use your ML solution?

This article is part of the series: How to Talk About AI

  Share


  Categories

Machine Learning

  See also

2022-10-20
Why We "Productize"
2022-01-01
Ethical Product Developer
2021-05-26
Hi Product Managers, Are You Creating Products That *You* Love?

  More

Last updated: 2022-05-07

  Author: Takuya Kitazawa

Takuya Kitazawa is a freelance software developer, previously working at a Big Tech and Silicon Valley-based start-up company where he wore multiple hats as a full-stack software developer, machine learning engineer, data scientist, and product manager. At the intersection of technological and social aspects of data-driven applications, he is passionate about promoting the ethical use of information technologies through his mentoring, business consultation, and public engagement activities. See CV for more information, or contact at [email protected].

  Support by donation   Gift a cup of coffee

  Disclaimer

  • Opinions are my own and do not represent the views of organizations I am/was belonging to.
  • I am doing my best to ensure the accuracy and fair use of the information. However, there might be some errors, outdated information, or biased subjective statements because the main purpose of this blog is to jot down my personal thoughts as soon as possible before conducting an extensive investigation. Visitors understand the limitations and rely on any information at their own risk.
  • That said, if there is any issue with the content, please contact me so I can take the necessary action.