The Sales Call That Happened Before the Sales Call

The first objection in an enterprise software deal often arrives long before the first call. A buyer wants to know whether the product handles a particular integration, where the data lives, or what happens when a contract changes. The company website offers a form. The answer is somewhere in a sales engineer's head, a Slack thread, or a slide last updated in March. By the time somebody finds it, the buyer has opened another tab.

Docket's business is built around that awkward interval. Its AI agent sits on a B2B website and starts a conversation, answering from company-approved knowledge. It asks questions, determines whether a visitor fits the seller's criteria, books a meeting when appropriate, and writes the exchange into a CRM. The sales rep gets a record of what the person actually wanted, not a solitary email address and a hopeful lead score.

A park meeting and a very fast check

Arjun Pillai knew the problem from the inside. At ZoomInfo, where he served as chief data officer after selling his earlier company Insent, his team supplied the technical and competitive answers that helped sellers move deals forward. He had previously built Profoundis with Anoop Thomas Mathew, who became Docket's co-founder and chief technology officer. Their first Docket idea was an AI-powered sales engineer: a system that could make those scattered answers available when an account executive needed them.

Foundation Capital's account of the founding has the pace of a short story. Pillai had just left ZoomInfo. He and Mathew had decided to work together again. A meeting with investor Ashu Garg took place in a park near Bangalore and lasted four hours. Foundation's seed commitment followed in 20 days, by Pillai's telling. The disclosed seed round was $5.3 million; in July 2024, Mayfield and Foundation Capital led a $15 million Series A, bringing disclosed funding to $20.3 million.

The early product helped sellers search a company's accumulated expertise and produce sales material. That made sense for large, technical deals. It also put Docket in a crowded neighborhood with knowledge search and sales enablement software. Then the company looked at a different point in the journey: what if the person needing the answer was the buyer, still on the website, before a seller knew the buyer existed?

“What if we didn't just help sellers answer product questions?”Arjun Pillai, describing Docket's 2025 shift

The question moved to the website

Pillai's year-end account says Docket extended its agent in 2025 from answering seller questions to handling the opening exchange with a visitor. The agent could discover a use case, retrieve approved information, show a slide or video, decide whether to route the conversation, and book a meeting. Docket calls the resulting record an Agent Qualified Lead, or AQL: a prospect with conversation history and qualification context attached. Its website now describes an inbound demand platform, with voice, text, and optional avatars.

That change matters because a form knows almost nothing. A name and company may tell a rep whom to call; they seldom explain why. Docket's agent tries to capture the buyer's question while it is still live. A visitor asking about Adobe AEM integration needs a different next step from someone seeking a jobs page. If the agent can distinguish the two and pass that distinction into Salesforce or HubSpot, the first human call begins several minutes further along.

The interesting word in Docket's technology description is “approved.” Its Sales Knowledge Lake gathers product documentation, pricing rules, security material, competitive positioning, and the lessons hidden in calls and work chats. Docket says the material is cleaned, versioned, and governed before its agents use it. The point is less glamorous than a fluent voice: a polished answer about an obsolete price is still a bad answer. When the approved material does not cover a question, the company says the agent should acknowledge the gap and hand off.

A test in public, with rules attached

Zenity offers a useful look at what deployment actually entails. It sells security software for agentic AI, so an inaccurate public answer would be especially embarrassing. According to Docket's case study, Zenity required accurate high-level responses before going live. The team fed the agent website material and a sanitized sales deck, instructed it to avoid unsupported pricing and support claims, and patched competitor positioning. It set discovery questions before the email request, sent conversation data into HubSpot, and reviewed results with Docket every month.

Those are vendor-reported case-study numbers, not a controlled experiment. They do, however, describe a real sequence: decide what the agent may say, connect the handoff, measure a specific outcome, and keep tuning. Docket also reports that a multi-region engineering firm had 70 qualified sales-intent conversations in 13 days after launch. Its publicly named customers and deployments span Demandbase, Whatfix, ZoomInfo, Sybill, ScreenMeet, Cart.com, and Zenity. The common thread is a sale complex enough that the buyer has questions worth answering before a meeting.

What the agent costs

Docket sells subscriptions based on monthly website traffic. The Growth plan starts at $3,000 a month, billed annually, for sites with up to 20,000 monthly visitors. Scale begins at $4,000 a month for 20,000 to 100,000 visitors; enterprise pricing is custom. Docket says plans include implementation, the knowledge foundation, CRM integration, meeting booking, support, and unlimited conversations. It advertises a typical rollout of seven to fourteen days and says it does not currently offer a free trial.

The purchase case is strongest where inbound traffic is already valuable and buyers regularly ask detailed product, security, or pricing questions. A company with little traffic, a simple self-serve product, or poorly maintained documentation has a tougher calculation. The software can conduct a conversation; it cannot make an uncertain product promise safe merely by saying it politely. That is why implementation is partly an editorial job: decide which answer has authority, retire the old PDF, and give an uncertain question a human destination.

Docket's own 2025 recap supplies the clearest reminder that this is operational work. One pilot walked away mid-engagement. Pillai says the team responded with an emergency debrief and focused on ingestion reliability, human handoff, security controls, and faster implementation. The company also reported 27 closed-won deals by year-end and more than 250,000 questions answered for customers that year. Those figures tell two stories at once: demand for this kind of tool, and the cost of getting an agent wrong in public.

“One pilot walked away mid-engagement.”Docket's 2025 company recap

The copyable part is the discipline

There is a modest experiment here for any revenue team, even one that never buys Docket. Take the ten questions a buyer asks just before agreeing to a meeting. Find the current, approved answer to each. Decide which question needs a specialist, which answer must never be improvised, and what context a rep would want before calling back. Then inspect where those questions land on the website. If the route is a generic form, the company has found the gap Docket sells into.

Pillai's original insight was that the answer probably exists somewhere. Docket's newer insight is about timing: an answer found after the buyer leaves is often no answer at all. The agent is the visible part of the product. The deeper product is the promise that, when a buyer asks a difficult question, the company knows what it is willing to say.