Tuesday, October 8, 2024
No menu items!
HomeCloud ComputingCrescendo makes AI boring—and profitable

Crescendo makes AI boring—and profitable

There’s money in AI, but not where you think. Sure, Nvidia and the clouds are cleaning up, but they’re primarily selling to enterprises that are mostly kicking tires, not running mission-critical AI applications. Some, like the Financial Times, suggest that AI startups are cleaning up but, again, this could simply be a matter of reselling GPUs, as Chris Gaun points out.

So where is the money? I recently posited that the real money in AI will hit once its infrastructure is turned into applications. Perhaps I should have clarified further, though: The real money in sexy AI may be in the unsexiest of applications like…call centers?

Making AI boring

“Boring” is good when it comes to enterprise IT (no one wants drama from their Linux servers, for example), and it’s also good for AI. When a longtime friend, Zack Urlocker, pinged a group of friends about a small AI startup he’d joined called Crescendo, my interest was piqued. Urlocker is a bit of a billion-dollar man, with a knack for turning dull industries into cool cash. He’s the guy top Silicon Venture firms call when a hot portfolio company is about to make a major move. He served as COO of Zendesk ($10 billion exit in 2022), COO of Duo Security (acquired by Cisco for $2.35 billion), and executive VP of product at MySQL (acquired by Sun Microsystems for $1 billion).

Boring industries. Not boring businesses.

Crescendo has AI all over its website but, importantly, it doesn’t inflict AI on its customers, as Urlocker told me in an interview. Rather, Crescendo is making huge piles of revenue by making AI disappear into its customer service application. This isn’t yet another large language model (LLM) raising billions to make millions. This company has raised millions and looks set to make billions.

The company started when company president Anand Chandrasekaran saw an opportunity to create a new category in what we all think of as call centers. (The market sometimes uses the names “customer experience (CX) platform,” “business process outsourcing” (BPO), or simply “call centers.”) It’s a huge industry, teetering on $1 trillion in annual revenue, but it’s relatively low margin and, well, boring.

Chandrasekaran had three profound insights. The first was that rapid advances in AI could vastly improve the quality of customer service and the accuracy of their information. After all, call centers are fundamentally a commodity industry that sells answers, and business is better when you have more right answers. Margins are low (10% to 15% on average) so most large call centers are located overseas in areas with a sufficiently large talent pool of English speakers and where the cost of doing business is much lower than in North America. Innovation in the industry is, at best, incremental. The last major technology disruption was voice over IP (VOIP) about two decades ago, which gradually replaced plain old telephone service (POTS).

Chandrasekaran’s second insight was the criticality of keeping humans in the loop. Rather than offload everything to machines, Chandrasekaran saw the need to provide workers with more and better training that could be coupled with ever-improving AI to drive better accuracy and personalization. At the same time, he reasoned, the AI should be smart enough to know when it’s dumb so that it can more quickly hand off the hardest cases to a human domain expert.

His third insight was to change the business model. Instead of pricing hourly for services or by the number of headcount servicing a client, he’d charge only for outcomes. That last insight finally aligned the interests of all three parties involved: the client, the actual customer looking for answers, and the call center. The clients, usually fast-growing companies, get needed help managing customer queries; customers get faster, more accurate answers; and the call center, which now looks much more like a modern software company, has profit margins four times higher than traditional call centers.

Serendipitously, Chandrasekaran chanced to meet a pair of brilliant technologists with decades of call center experience, Tod Famous and Slava Zhakov, who shared the same vision. Even better, they had already written a working prototype. They didn’t think their own industry would wake up before it was too late. Crescendo was born.

AI that works with humans

Crescendo’s novel approach of using AI with humans still in the loop also addresses two elephants in the industry’s room. The first is deflection, which is a polite way of describing how hard companies make it for customers to even find out how to contact a company before getting shunted into the FAQ and phone tree wilderness. Then there is churn. Unsurprisingly, a lot of the industry’s jobs are pretty boring, leading to stratospheric employee churn rates of up to 50% a year.

Crescendo’s approach to AI seems smart. Instead of throwing hundreds of millions of massive clusters of GPUs and competing with OpenAI, Google, Meta, etc., for the latest PhD graduates in AI from MIT, they’ve put humans in charge. For Crescendo, AI doesn’t displace humans, it complements them. At the same time, Crescendo’s proprietary AI application can still benefit from tapping into the world’s largest LLMs and even the private knowledge bases of their customers to answer the vast majority of client questions. Best of all, customers claim they can go into production with Crescendo in just two to four weeks, and after the first month, AI is handling more than 90% of the queries automatically and accurately. To date they have yet to experience a single hallucination and there has been zero customer downtime.

You do the math. Assume all the players in this huge three-quarter-trillion-dollar industry are achieving their highest margins of 15%. That would be more than $112 billion in annual profit. Now try 60% margins. That would be $450 billion but with much happier customers and employees.

Urlocker just might be keeping his “billion dollar man” moniker yet again.

Explained: How Salesforce Agentforce’s Atlas reasoning engine works to power AI agents | InfoWorldRead More

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Most Popular

Recent Comments