Employing a rigorous approach to the words and phrases at the heart of our products, ABIe’s avatar-driven interface quickly provides accurate answers to policy questions while streamlining the preparation of sales quotes.
ABIe’s volume has soared from 20,000 questions during the first six months online to 1.2 million a year, in large part because we’ve expanded her role in the business. And though her workload has exploded, ABIe is the opposite of a cost center—our expenses to develop her were more than offset by call-center savings even before we celebrated her first birthday. Having recently turned three, she continues to contribute to the bottom line.
Here are 5 lessons than can help mid-sized companies get their own ABIe.
1. Don’t think you have to spend tons of money. We didn’t. At the high end of artificial intelligence is cognitive computing, involving systems with the capacity to learn. At the less-expensive end is a knowledge-based approach that organizes highly accessible and interactive systems of information. That’s our playing field right now, and we can build on it in the future since all of our moves there are necessary components for more ambitious AI efforts. In fact, we think of ABIe as our precursor to cognitive computing on a shoestring.
2. Find the right AI partner. During the selection process we learned that while many vendors say they have AI expertise or claim the off-the-shelf applications can achieve business objectives, that was not the case. We worked with Earley Information Science because they could accurately scope and scale the project within our budgetary and time parameters. ABIe was up and running in less than a year, which is very fast given all of the moving pieces in a knowledge-based system.
3. Give YOUR ABIe the right information. ABIe pulls her answers from a data warehouse, where all the relevant knowledge relating to our products and processes has been organized. In AI speak, that means our numbers, charts, words and phrases have been chopped, chunked, tagged and HR-optimized to give ABIe the ingredients for those answers. The accuracy and value of the answers depends not only on all the necessary ingredients being present, but also on their freshness and preparation.
4. Be ready for mistakes and course corrections. Inevitably, we got some things wrong in the way we prepared our ingredients, despite assigning three managers to the project along with nine subject-matter experts. Our biggest mistake was providing ABIe with too many ingredients, making some of her answers too thorough. It was as if we were asked for the time and then told the questioner how to build a clock. It took trial and error—in part, by debriefing agents after their sessions—to arrive at very specific, and far more useful, answers along with the governance, metrics and change management processes to make controlled, methodical updates.
5. Look for new ways to use ABIe. ABIe is tireless and never asks for a raise. And she’s almost entirely a sunk cost. So any new applications can only swell the bottom line. One of our unexpected dividends was further savings at the call centers, where attrition rates can be as high as 33% a year. ABIe cut the training time in half for replacements. In the larger picture, ABIe has morphed from a how-to role in policy administration into something of an oracle on marketing, underwriting and other elements of the business. The sum total of our knowledge is now in the warehouse and ABIe has become the go-to answer place for all employees. Next up: we’ll be turning ABIe’s face outward, to help customers.
One last tip: Change is constant in the virtual world, as in ours. As valuable as our ABIe has been, some younger siblings are bound to come along in the AI family, and we’ll be happy to hire them, too.