Vercel's COO prices the AI agent that replaced nine SDRs at under $5,000 a year

Vercel's chief operating officer put a price on the AI agent that now does most of her company's inbound sales qualification: under $5,000 a year, for work that used to take a team of ten. Jeanne DeWitt Grosser laid out the economics in a talk at the SaaStr AI conference, recapped by SaaStr on November 6, and the same core figures appear in an earlier, independent interview she gave Tom Tunguz for his Theory Office Hours show: Vercel's inbound sales development team went from 10 people to roughly one, and a separate agent now closes 93% of the company's support cases.

What changed

The lead-qualification agent did not replace the team overnight. It launched running on about 20% of one engineer's time, then spent six weeks in shadow mode, producing output that a top-performing SDR reviewed before any of it reached a prospect. Only after that review period did Vercel cut headcount. DeWitt Grosser's own description of the result, as quoted by SaaStr:

"It performs like a 90th percentile rep now. 100% of the time."

Today the agent runs US lead qualification on roughly one person's time and Europe/APAC coverage on about 20% of another's. Vercel also automated 30 separate SDR workflows, and DeWitt Grosser said SDR quotas rose 30% in the quarter that followed. Her own math puts the return at 32 times the cost of the old ten-person team — a figure both interviews repeat.

The support side cost far more to build. The agent, which DeWitt Grosser calls "our internal Sierra," took three engineers roughly two months to bring to production. It now handles thousands of technical support cases a week at an all-in cost SaaStr puts near $150,000 a year: about $300 a month in infrastructure plus roughly $12,000 a month in model tokens. Separately, Vercel says a content agent completed 96% of its major marketing content updates last quarter, and a data-analyst agent called d0 now answers in under a minute questions that used to take a week through a data-science ticket queue.

Who this affects

This lands on whoever owns pipeline cost and support cost, not just engineering. A 90%-automated inbound SDR motion changes what a demand gen or marketing ops leader needs to staff for: lead scoring, routing and qualification increasingly happen before a human sees the lead, which shifts where marketing's job ends and sales' begins. It also resets the cost comparison marketing leaders will hear in planning meetings. If a named COO can point to a sub-five-figure agent doing SDR-grade qualification, every software marketing leader asking for budget to add reps should expect that comparison to come up, regardless of whether their own funnel looks anything like Vercel's.

The support numbers matter to a different audience: teams running a product-led or self-serve motion where support tickets function as an extension of onboarding. A 93%-automated queue frees headcount, but DeWitt Grosser's own figures show it isn't cheap to get there — three engineers and two months of dedicated build time, not a weekend integration job.

There's a marketing-specific wrinkle too. Vercel's content agent handling 96% of major content updates sits inside the same reporting line as the lead agent handling inbound qualification — meaning the line between "marketing content ops" and "sales development" is thinner at Vercel than it is at most software companies, where those teams rarely share a budget owner, let alone an agent stack. A marketing leader benchmarking against this case should ask whether their own content and SDR functions report up through enough of a shared pipeline-and-content view to make a single agent, or a shared agent budget, make sense — or whether the org chart would have to change first.

What to do differently

• Price out the review period, not just the agent: Vercel ran six weeks of human-reviewed shadow mode before cutting any headcount.
• Keep the lead-qualification cost comparison separate from the support-automation one when benchmarking your own stack — they are not the same order of magnitude.
• Track quota performance, not just headcount, when you automate SDR workflows. Vercel's reported 30% quota lift followed automating 30 specific workflows, not a single blanket rollout.

DeWitt Grosser's figures are self-reported, and the two interviews don't fully agree with each other: she gave the SDR headcount as roughly one person in the SaaStr recap and as 1.25 in the Tunguz interview, for what appears to be the same team. The gap is small enough not to change the substance, but a founder's or COO's own numbers, repeated across press-friendly interviews, deserve the same scrutiny a vendor case study would get before they shape next year's headcount plan.