3 min read

A Framework for Crushing The Competition With Machine Learning

When every one of your competitors is using machine learning, how do you find the edge to beat them? (especially with scarce resources).

Over hundreds of discussions we have had with entrepreneurs, we have seen that the most savvy founders are finding competitive advantage setting up their companies machine learning strategy to OUTLEARN the competition. And we’re seeing them even against agile incumbents like Facebook or Google who have more formidable data assets.

The pattern we’re seeing emerge is to: focus “narrowly” on “the right tasks” and drive the only thing that matters, PERFORMANCE. Breaking it down, we think the equation can be split into three parts:

When every one of your competitors is using machine learning in their product, how do you find an edge to beat them?
  1. EXPERIENCE: Focus On “Small” Data — if you can tightly determine the “right” data for training and new observations you can outlearn the competition. While they are passing in noise, you’re passing in signal! This is hard to do (given the entire premise of machine learning is NOT knowing which data is right!), but even efforts to pass in cleaner data can yield great results.
  2. TASK: Define Tasks Precisely — Take a business approach and tightly define the tasks that really matter to customers better than your competition, and model them appropriately. This requires you to understand your customers’ needs and your industry better than the competition.
  3. PERFORMANCE: Own the Performance Linchpin — if you own improvement in performance when the competition doesn’t — or isn’t as singularly focused — you will be able to consistently improve your algorithm’s learning component. Without this, you cannot win in end-to-end machine learning. Occasionally, it can be as simple as having a Human-in-the-Loop (such as having sales people mark the quality of the understanding in real time), or after the fact sending any feedback to the model for improvement. The key is owning both the output, and the ability to feed back performance output as training input back into the model for improvement.

Creating a better feedback loop is how startups today can gain a competitive edge, with even less data. It’s about owning the performance metrics of the system and using those to help improve the overall machine. So as an entrepreneur, don’t cut yourself short due to a lack of training data or an imperfect algorithm. Focus on giving your algorithms tighter tasks and driving their performance with creative feedback loops for continual improvement.

To that end, we also see lots of new opportunities for entrepreneurs to capitalize by building a machine learning edge (and many more we’re excited to hear about!). In the “request for startups” vein — there are a few areas we’re looking to find great entrepreneurs:

  • Crowd Sourced (or Manufactured) Data Labeling — developing creative ways to outsource data labeling or featurization or alternative approaches to “manufacturing pre-labeled data”.
  • Narrow “ROI-Centric” Applications — that deliver measurable performance improvements that grow topline are much easier to implement into existing business practices because they create value you can capture. This makes the complexity of ML tangible by linking it with results business leaders can recognize.
  • Machine Learning Testing & Optimization — in a world where it is hard to “see under the hood” or even best measure performance relative to other models — we’re interested in seeing a solution for benchmarking performance, (A/B) testing, & multiple-model deployment.

We are always excited to meet bold entrepreneurs at Underscore, so tell us, how are you thinking about building an advantage with machine learning? What edge or opportunity do you see?