Share
BLOG

Increase the Effectiveness of Your Machine Learning Systems by Keeping Humans in the Analysis Loop

Summary:
Increasing the accuracy of analytics technologies drives modern business, and getting more realistic results applicable to on-the-ground business operational conditions means orgs can make better decisions faster.

Keep Your Experts on the Front Lines

Increasing the accuracy of analytics technologies drives modern business. Whether it’s an advertising targeting system, a consumer demand forecasting platform, or a supply chain planning algorithm, getting more realistic results applicable to on-the-ground business operational conditions means orgs can make better decisions faster. The key to “better and faster” is that the technology continues to deliver applicable results. The best way to do this is to allow domain experts to assert their expertise by influencing the outputs of the technology. Enter human in the loop machine learning.



The image above shows what we’ve seen across industries and use cases. Incorporating machine learning into analytics technologies can be very expensive, but if a 5-7% gain in accuracy makes up for the investment, you’ve got your ROI. Even more importantly, developing the technology in a way that enables domain experts to exert control over these technologies leads to another incremental gain early on, and also continues to refine the technology over time as business conditions change on the ground. 


Let me demonstrate by way of example.


Supply Chain Planning

Here’s an example from one of our customers in the seed production space. The global supply chain planning team continually replans their strategy to cultivate seeds and distribute product to each of their regional bases to make sure their customers have the inventory they need on time. They make these plans up to three years out. As you can imagine, the situation evolves all of the time due to unforeseen conditions in the marketplace, the demand from customers, their facilities, and even severe weather! To build their forward-looking plan, they use a tool that looks like this:



As you can see, the tool recommends a plan based on current conditions, delivery milestones, and knowledge of the last plan enacted. A typical machine learning based system would do the same, and likely generate some excellent results (we know because that is industry-standard technology). The difference here, and the thing that makes this system so powerful, is that the recommendations also integrate the knowledge of the supply chain planners who have decades of industry experience.



This works by monitoring the interactions of the supply chain planner with the tool (using standard web app click stream monitoring technology), and transforming those interactions into “features” that power the machine learning supply chain plan recommendation engine. When a user makes changes to a recommended plan (based on their experience and accumulated knowledge), the information is then incorporated into the system so that the next time a plan is generated by the recommendation engine, it uses the domain expertise learned in the last go-round. This happens continuously and automatically. Every time a planner uses the app, the system gets smarter and generates more realistic plans. This leads to a fully adaptive system that evolves as your business operations change in real time.



Integrating human in the loop machine learning into your analytics technologies is the next evolution of business intelligence. There is no substitute for the institutional knowledge aggregated by your organization’s domain experts. Leverage those assets all the way across the business by investing in human in the loop machine learning.



Subscribe to our quarterly newsletter

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.