Take it up to 11 with a human-in-the-loop

This pandemic has revealed the power of AI to anticipate our wildly changing shopping habits. Whenever we take an online shopping excursion now, we are treated to a barrage of hand sanitizers, face masks, and stay-at-home loungewear, where last summer it was swimsuits, sunblock, and last-minute vacations.

According to MIT Technology Review, once COVID-19 hit, it took only days for Amazon search terms to serve up pandemic-related products. But soon, some retail systems began to behave oddly, disrupting automated back-office processes like inventory, fraud detection, marketing, etc.

The machine-learning models that enabled these processes were trained on “normal,” seasonal shopping behavior and designed to make decisions without any humans-in-the-loop. This was a reminder that the most effective way to avoid disruptions is seamlessly involving humans to provide feedback adjusted for new (and ever-changing) circumstances–proving that nothing understands humans, quite like humans.

As our lives get increasingly intertwined with AI, we accumulate more and more evidence that design decisions behind these powerful systems are critical (see this story by our friends at ArthurAI), and it’s becoming clear that outcomes are much better when humans can jump in to provide real-world context. Just like in our personal lives, this is equally true in the enterprise.

70 is the new 💯

When AI involves humans, it can also tackle a new class of cognitive tasks that have been out of reach for most machine learning (ML) systems. In the enterprise, a big hurdle for deploying ML-enabled automation has been that it had to complete tasks 100% to add value to a workflow.

Many ML initiatives languished at 70% - 80% task completion, always just short of what was required for production deployment. The new class of human-in-the-loop (HITL) ML systems overcome this challenge by making it easy to escalate harder cognitive tasks for employees to resolve, marrying the best of both worlds.

This HITL design is at the heart of Alkymi’s approach to reliably let enterprises automate repetitive tasks, transforming how work gets done. Approached intelligently, human input becomes a valuable feedback loop that enables an increasing amount of automation as the system learns – a process often referred to as active learning.

Even if AI is embedded to automate only 70% of a workflow initially, it’s a massive productivity gain. Some might even throw a 💯 on that.

A human-in-the-loop is peak 2020s

Businesses see enormous potential for AI and automation to make work more efficient and massively scale output. According to Deloitte, "Over the next three years, executives expect automation to increase their workforce capacity by 27 percent: equivalent to 2.4 million extra full-time employees."

One of the most serious impediments to operational efficiency, data extraction from emails and documents, will soon be a thing of the past. Think of service requests stuck in email inboxes, insurance claims, customer onboarding materials—all of this unstructured data can be extracted and assembled in real-time. Employees will spend less time processing data and more time with customers.

‍By working in tandem with AI, any business can reach the peak capacity of its offering. AI is a simple and efficient way of upskilling workers and ensuring that a company can handle any issue or query that comes along. AI is redefining how work gets done, but in the end, it’s a modern-day version of teamwork. AI plus humans equal amplified results.

See how Alkymi can help take your business to a whole other level. Schedule a demo.