I have defined 2 new categories, 1 success 1 failed attempt, and am in the process of defining my 3rd. Machine Learning supports new business models and product categories. As I am getting into the late phases of my 3rd category, it is becoming obvious how few companies, even startups, know how to do this.
This post is a walkthrough of the top-level points to consider when building a new machine learning product that requires a new category definition to be successful.
What is a New Category?
What differentiates a Machine Learning Product from a Traditional Software Product?The built machine learning model is a product that could not exist without the model.Customers will pay for inference alone.Pricing is based on the model’s functionality.
What makes a Machine Learning Product part of a new category?The customer cannot get the inference from any other source.The training dataset is novel because it has been built by a process the business controls.The business’s research has created artifacts to support intellectual property, practical applications, and a sustainable (at least 2 years) competitive advantage.
The traditional definition does not apply to machine learning products. For traditional software products, new categories have no market awareness. Explain the concept and most listeners look at you like you are crazy or an idiot. Having several of these meetings is the earliest hint the business is wandering into new category territory.
The “You’re Crazy” reaction does not happen with machine learning products. People expect machine learning to enable new products. The buzzword makes people suspend their disbelief. Meetings end with the audience saying, “I get it. That is awesome. I want in.”
The product releases in alpha and the response is, “Not what I was expecting. Not seeing a fit for my needs. Not willing to pay a premium price point.” In my failed attempt at creating a new category, the business decided to pivot towards the UI after 3 months in alpha to rapidly expand the customer base.
Seemed like a great idea but looking back on it, the decision killed the product. The company moved further and further towards a traditional software product from that day forward. The value proposition was completely undermined because it was just another offering in an increasingly crowded field of AI enabled products. Worse, the business was staffed to build models then abruptly retasked with building a frontend heavy application and providing direct support for users.
It is up to the business to realize the product is in a new category. The business must understand the category early on to build a successful product strategy. The market will work to classify the product into an existing category unless the business establishes the product in a new one. The business must position the product to define and prove the new category. Value is a major aspect of proving the category, so any free pricing model is taken off the table. Customer 1 needs to be a paying customer.
How Do You Define a Category for a Machine Learning Product?
From experience I can say people will only invest in or pay for a Machine Learning Product if there have been early successes. It feels like a catch 22. How do you get paying customers if no one understands the value proposition?
Defining the category supports the value proposition and leads to paying customers. Start with a supportive target market. My experience is in B2B Machine Learning Products:Obvious, significant business problem with high costs associated with the in-place process.C Suite has metrics to quantify the problem and measure improvement. They have goals and timelines for improvement.No feasible solution available.A vocal user base within those businesses.
Angry users are motivated users. Businesses looking for revenue or margins are motivated customers. Performance goals are a request for proposed solutions. Customers with all 3 define the alpha release target market.
New categories are born when an angry user meets a creative Data Scientist. The initial customer base is part of the angry user’s professional circle.
The 1st alpha release feature set is based on the angry user’s domain expertise and validated by research into the target market. The frontend, if there even is one, is minimal.
Customers do not care if there is integration work or learning curve involved. The need is greater, and the new solution is the only option.
New categories are not easy. The perception that any new Machine Learning Product creates a new category is misguided. However, in those cases where a new category is being created, common product strategies fail.
How Do You Take a Machine Learning Product to Market in a New Category?
The product cannot go wide fast. Defining a new category focuses on perfecting core functionality. A large, diverse target market will pull the product in too many directions. Support, new feature development, and scale pull resources away from defining the category. The product will always be 1 feature away from going big and never get there.
The machine learning model needs to be improved to create distance between its capabilities and potential fast followers. As soon as a business has defined and proven a new category, competitors will rush in. Every resource spent on functionality outside of the machine learning model takes away from building a competitive advantage.
Product development must focus almost entirely on model development. Model improvements are dictated by pricing increases. Revenue, sometimes margin, is the key feedback loop for a Machine Learning Product in a new category. Defining the small target market keeps the close connection between the business and the product.
Key development activities are centered around applied research. Research artifacts are functional, either unique datasets or models. Only artifacts that support higher price points or new tiers should be getting resources. The product exits alpha when model improvements no longer support higher pricing.
Machine learning, especially research, can get siloed. The connection to pricing keeps the door between customers and Data Scientists open. In a new category, there is no specification unless the business actively works to build one. In a silo, the model starts leaning towards what the Data Scientists think is important. Pricing forces Data Scientists towards business centric decisions.
Beta Version Machine Learning Product Releases in a New Category
The first beta version should put the business at break even for the new product. It cannot be a perpetual cost center. Feature scope is dictated by a larger target market. Basic Product Strategy explains most of the key concepts so I will only highlight a few differences.
New features continue to be driven by price. The category has been defined and proven but not understood outside of the alpha target market. Beta is too early for new features to be customer dictated.
The business is still selling the category first, then the product. Beta releases go to customers once they have learned and bought into the new category. New features should reduce the learning curve and barriers to adopting the category. Value is already built into the alpha version. New functionality drives pricing by making the product less expensive to own, use, adopt, integrate, etc. Model improvements are split between competitive advantage and efficiency.
The business creates rituals around the product to build loyalty. New categories can be dropped as quickly as they are adopted during the beta versions. Selling a category leads to higher customer acquisition costs so retention is obviously important. Category first selling needs rituals to get customers attached to the product. Rituals follow the basics: commitment, cadence, community, content, and control.
Monetizing Machine Learning Products comes with new strategy planning considerations. The potential for a new category is one of them. I advise clients about Monetization Strategies for Machine Learning Products. Reach out to me by email for details: firstname.lastname@example.org.