CJ Desai on Enterprise AI and Finding the Buyer

At Sequoia’s AI event in 2024, ServiceNow president and COO CJ Desai discusses enterprise AI, product expansion and selling to CIOs. Sequoia was an investor in ServiceNow, and the company figures and product claims are those presented at the event.

The host highlights ServiceNow’s historical growth, margins and scale. These 2024 figures are a snapshot and should not be treated as current financial information or an investment recommendation.

Desai thanks Sequoia for backing the company and recounts a decision to pursue an independent path rather than accept an acquisition offer before its IPO. The story is his account of the company’s history.

He describes joining when the company had one large product and several smaller ones. Its platform strategy required deciding which new problems to solve and which buyers could actually be reached.

For a proposed product, his questions include who the buyer is, whether existing relationships provide access, who the competitors are and whether the opportunity is large enough. He emphasizes execution as well as the idea.

ServiceNow’s AI approach, he says, is to apply models to existing workflows where customers already have a concrete problem. The company acquired small AI teams, including Element AI, to build research and engineering capability.

Desai describes using open-source research and working with Hugging Face on models suited to ServiceNow use cases. He frames model selection around product needs, latency and the cost of serving customers.

He treats gross margin as a constraint: a large general-purpose model everywhere may be expensive, so his team explored smaller specialized models on existing hardware. This is an engineering and business tradeoff, not a claim that smaller models always perform better.

Asked how startups can sell to ServiceNow, he says a proposal should explain a specific improvement to the company’s customer workflows. He also sees potential partnerships where a startup needs access to the CIO buyer.

He uses employee hardware provisioning as an example of an enterprise workflow: approvals, inventory, configuration, security and delivery may involve several departments. Automation aims to make the overall process easier, not just one screen faster.

On a second product line, Desai advises founders to understand the core market and buyer before expanding. ServiceNow’s own timing and scale were specific to its circumstances, not a universal revenue threshold for every startup.

He says the company expanded into security, HR and customer service after establishing its core IT business, using CIO relationships to reach adjacent buyers. Product expansion still requires separate validation of each buyer’s needs.

When prioritizing AI features, he looks for a customer pain point and a plausible improvement in outcomes. He describes large public-sector investments and certification requirements as part of the cost of entering that market.

Desai argues that enterprise buyers ask both what an AI feature costs and what return it can deliver. He recommends explaining a concrete outcome and recognizing that realized savings may differ from a vendor’s projection.

He disputes the idea that the CIO is irrelevant to enterprise purchasing. In his experience, CIOs remain important for decisions affecting infrastructure, cost and revenue-related workflows; buying structures vary across organizations.

Asked what has and has not worked, he says extracting and matching information from varied invoices and documents remained difficult in his experience. He contrasts this with narrower prediction tasks based on established workflow data.

One example predicts an employee’s upcoming computer refresh and asks for confirmation before proceeding. Desai emphasizes a human decision point, reflecting enterprise customers’ preferences in that use case.

The host closes by summarizing buyer identification, model cost and customer outcomes. Desai adds that the company made mistakes while scaling and continues to learn which opportunities to prioritize.