Companies selling batteries of any sort to businesses grapple with the fact that the devices are expensive, take up valuable space and tend to require a painful and lengthy sales cycle.

Those downsides can get in the way of the appeal, which is to save companies loads of money on their electricity bills by dodging peaks and demand charges. As such, several commercial battery vendors have shifted their business models, but none have done so quite like Axiom Exergy.

The Bay Area startup launched in 2014 with the thesis that lithium-ion batteries made little sense for businesses with high thermal loads, like grocery stores, and that thermal storage could optimize cooling and slash bills for considerably less upfront investment. 

The company got a few initial "refrigeration batteries" into stores, including Whole Foods and Walmart. Along the way, it built a software platform that tapped into existing refrigeration systems' data collection in order to predict upcoming peaks and front-load cooling sessions accordingly.

Then something unexpected happened, said co-founder and CEO Amrit Robbins in an interview Monday. "We were really blown away by the results," he said. "Customers asked to deploy the software without the hardware."

Instead of convincing building energy managers and their bosses to hand over square footage to a tank of salt water, the software makes a much simpler pitch. Refrigeration controllers already collect hundreds of data streams, but they typically only get used if something goes wrong and triggers a manual inspection, Robbins said. The logs get wiped every couple of weeks to make room for more under-utilized data.

Axiom simply inserts a small computer box into the existing refrigeration controller, then feeds the data stream into an AI algorithm alongside relevant information, such as weather and utility rates, and uses its analysis to ramp cooling up or down to minimize the customer's bill. It calls this Axiom Cloud, and the company has contracted more than $1 million in bookings over the last few months, Robbins said; he hopes to have 100 sites in operation by the end of this year.

The shift from hardware-centric to software-centric sales dramatically lowered Axiom's cost of goods sold, sped up the sales cycle and generated recurring subscription revenues. Those are all things that venture investors like to see, which should please Axiom backers including Shell Ventures, GXP Investments (now Evergy Ventures) and former Tesla CTO JB Straubel.

But it also fundamentally alters the addressable market, Robbins said. Axiom used to chase the geographically and numerically limited subset of grocery stores that would have considered a lithium-ion installation and then sell them on a lower-cost thermal battery.

"Our total available market now has expanded to almost any grocery store out there," he said.

Robbins hasn't given up on the old hardware concept; careful management works up to a point, but it can't replace a physical vessel of pent-up cooling. Now, though, he can save customers money immediately, while building up real-world datasets to help grocery companies pinpoint which stores could benefit most from the extra heft of an onsite thermal battery.

Maybe 20 to 25 percent of a fleet of grocery stores would want to add the hardware after running Axiom Cloud for several months, Robbins estimated.

Besides selling directly to grocery chains such as Whole Foods, Axiom has found interest from third-party maintenance contractors, including The Arcticom Group and Clima-Tech Refrigeration and HVAC. Keeping chillers up and running is a matter of great financial significance to grocery store owners; in addition to cloud-based controls, Axiom also offers AI-based predictive maintenance.

This service materialized after the virtual battery at one store started throwing out "nonsensical" results, Robbins said. Investigation revealed that the algorithm assumed the equipment was in good working order; the gobbledygook emerged because the refrigeration system was facing an imminent catastrophic failure.

Axiom called the customer to share the findings. Two weeks later, Robbins said, the equipment failed and knocked itself out for two days. Now that type of predictive maintenance has been formalized as an official service offering.

This application of AI has made inroads in other industries. GE, for instance, touts a "digital twin" program that computes when and how power plants are likely to break down, so that plant operators can act before that happens. But it hasn't yet become commonplace for grocery stores.

"We’re bringing AI to the commercial refrigeration space," Robbins said. "There's a lot of pent-up demand for it."