Que Sierra Sierra

Sierra has raised $110 million to deploy technical developments

🔷 Subscribe to get breakdowns of the most important developments in AI in your inbox every morning.

We finally have some visibility on what former Salesforce CEO-in-waiting and OpenAI board member Bret Taylor has been up to: customer service bots.

A Sierra interaction

Sierra has raised $110 million from Benchmark and Sequoia, and from the sounds of it, it just plans to execute well on deploying technical developments made elsewhere. Notably:

“What that means in practice is that there’s not a single model producing a response from a Sierra agent.” In fact, Taylor says, it sometimes involves as many as seven models, including one they have dubbed “the supervisor” that monitors answer quality, and if it deems the answer questionable, it sends it back for reevaluation.Techcrunch

The “supervisor“ architecture is similar to the LLM-OS others have discussed:

Sierra is also pioneering outcome-based pricing, meaning they get paid when they resolve the issue. This is a big deal, as they get paid for value generated/cost saved and become directly visible in management accounting. It also allows them to charge substantially more and allows them to ask for and receive much deeper access into the customer’s service stack.

“We think outcome-based pricing is the future of software. I think with AI we finally have technology that isn’t just making us more productive but actually doing the job. It’s actually finishing the job,”Bret Taylor, CEO toTechcrunch

For now, it looks like a deep-integration enterprise software solution. They most closely resemble Palantir, with deployed engineers building pipelines into the customer’s data stack to extract, cleanup, and load data necessary for the product to run. Well-defined inputs, well-defined and limited actions that can be taken. If Palantir-Bot can order drones, why can’t your company bot cancel a customer’s shoe delivery?

Sierra:

• has access to retrieval augmented generation of company policies and documentation

• has customizable restrictions

• has allowed actions

• handoff or escalates to human agents if things get too tough

No doubt, over time, the “disengagement rate” will drop.

Ate’s View

Skeptical but hopeful. If it does work, the long hours of being on hold to change a reservation or ticket are finally over.