When Machine Learning Just Isn't Enough

April 18, 2017


At SaaStr earlier this year, I spoke about the huge potential of machine learning in SaaS. In that talk, I broke down some of the advances in ML that might be useful for software companies. In the discussion that ensued, I stressed the importance of not letting the technology obfuscate the value proposition of the software. Yes, ML is a huge step forward, but it’s not enough by itself. In fact, it likely isn’t the most challenging part of building a disruptive product.

During a recent interview, Salesforce Chief Scientist Richard Socher spoke about the importance of workflows.

There are three key phases of enterprise A.I. rollout: data, algorithms, and workflows…you have to be very careful to think about how to empower users and customers with your A.I. features. This is very complex but very specific. Workflow integration for sales processes is very different from those for self-driving cars.

Of the three, workflows are the hardest. Only by nailing the workflow will a user grant you the time and permission to wow them with machine learning.

Designing software that fits a user workflow is difficult. Better workflow is the hallmark of great software. Achieving it requires studying, researching, and distilling the way users work today. Some founders write software to solve their own problems, and understand the existing workflow from experience. Other founders interview hundreds of potential users and customers before writing code. A smaller group have great instincts. There are many ways.

ML can improve the workflow by reducing data entry time (RelateIQ), improving efficiency (Infer), providing novel insight (Chorus), among many other use cases.

But the challenge is always the same. How can software improve a current workflow to such an extent that a user is willing to stop their current workflow and learn a new one? The product’s value proposition must overwhelm the user’s inertia to keep working the same way.

Machine learning enables startups to inject a new type of magic to their product. In every great product, there’s a bit of magic. But it’s the workflow that keeps the users coming back.

The post When Machine Learning Just Isn't Enough appeared first on Tomasz Tunguz's blog.

Previous Article
Is There a No Man's Land in SaaS ACVs?
Is There a No Man's Land in SaaS ACVs?

A founder asked me recently if a dead zone in ACVs (average contract value) exist around the $10k price point.

Next Article
The Four Dimensions of a Demand Generation Portfolio
The Four Dimensions of a Demand Generation Portfolio

After a startup establishes product market fit, scaling demand generation becomes the the next major chal...