The pitch sounds almost too convenient: What if you could compress six months of McKinsey-style process consulting into a single workday? Interview every employee simultaneously, synthesize their collective knowledge into process maps and automation roadmaps, then present the whole package for about $50,000—a tenth the usual fee.
That's the proposition from Ontora, a San Francisco startup fresh out of Y Combinator that describes its approach as "100 conversations in parallel." Whether it works as advertised remains an open question. But Ontora isn't operating in a vacuum. A cluster of venture-backed companies—Linc AI, Lemma, Workstuff, Horizon, Patrium AI—are racing to prove the same basic thesis: conversational AI can conduct structured employee interviews at scale, capturing the undocumented tribal knowledge that traditional consultants spend months extracting through calendar Tetris and conference room sessions.
The timing isn't accidental. Industry analysts have framed 2026 as the year agentic AI either escapes what they call "pilot purgatory" or gets stuck there indefinitely. So far, the results are mixed.
The Trust Gap
At the RSA Conference in late April and early May, VentureBeat reported insights from RSAC 2026 painting a stark picture. Roughly 85% of enterprises are running AI agent pilots, yet only 5% trust those agents enough to deploy them in production environments. The gap isn't subtle, and it reflects a governance problem that most IT organizations haven't solved: agents need access to systems, data, and decision-making authority that few companies have properly secured or monitored at scale.
Conversational discovery tools occupy a different niche—perhaps a safer one. Instead of automating mission-critical transactions, they interview employees to map tacit knowledge. Ontora's founders, who bring backgrounds in process optimization at Porsche and AI automation within consultancies, describe the opportunity in blunt terms. A frequently cited figure suggests that about 80% of operational knowledge is tacit—undocumented, locked in employees' heads—though this lacks definitive contemporary peer-reviewed validation. Traditional consulting methods might sample 10% of an organization through interviews constrained by calendars and billable hours. AI-led interviews promise 100% coverage without the scheduling nightmare.
The broader market for process intelligence software has been growing fast, though from a modest base. Revenue reached $1.4 billion globally in 2024 and is forecast to hit $21.92 billion by 2030, according to Grand View Research—a compound annual growth rate approaching 60%. That's the projection for event-log mining vendors like Celonis, UiPath, and SAP Signavio, which have historically dominated the category. The forecast, based on a 2024 baseline, predates the current agent wave, and the shift to conversational discovery could accelerate adoption by lowering the technical barrier to process mapping. Or it could fragment the market further. No one seems entirely sure.
Why Now
Three factors have converged to make interview-first discovery viable, at least in theory.
The first is raw capability. Large language models can now conduct structured follow-ups, adapt questions to different personas or departments, and synthesize transcripts across dozens or hundreds of employees without human supervision. The agents don't get tired, don't double-book meetings, and don't unconsciously favor executives over frontline staff. Whether they ask the right questions is another matter, but the technical floor has risen.
Second, infrastructure. The Model Context Protocol—an open standard designed to connect agents to enterprise systems—has seen rapid adoption across vendors. It allows conversational agents to pull in meeting notes, support tickets, documentation, and event logs to cross-validate what employees say in interviews. Security researchers flagged remote code execution risks in some MCP implementations during April and May, prompting a scramble to harden registries, sign packages, and sandbox access. The standard is still evolving, in other words, but momentum is building.
The third driver is simple pressure. Gartner predicted in August 2025 that 40% of enterprise applications would feature task-specific AI agents by 2026, up from less than 5% the prior year—though this forecast may no longer reflect the most current expectations given developments since then. That timeline now looks optimistic for production deployments, but pilot activity is intense. Deloitte's State of AI survey, fielded in August and September 2025, found strong interest in agents but noted that roughly 80% of enterprises lack mature governance controls for agentic systems—a finding that, while somewhat dated, continues to resonate.
That friction—between ambition and readiness—has created demand for lower-risk use cases. Internal process mapping doesn't expose customer-facing workflows to agent errors. It generates insight without immediately automating anything. For executives eager to show AI progress without risking a high-profile failure, it's an appealing entry point.
Proof Points (Sort Of)
Ontora's website includes a testimonial from a strategy manager at Vertiv, though the company hasn't published independent audits of its ROI claims. The core pitch centers on speed and cost: "McKinsey-style $500K+ vs Ontora ~ $50K" and "3–6 months vs 4 hours." The product runs structured interviews in parallel, synthesizes responses into themed insights and process maps, then outputs a prioritized automation roadmap. It also exposes a knowledge graph via the Model Context Protocol and REST APIs, allowing other agents—Claude, ChatGPT, and the like—to query the findings.
Linc AI, another Y Combinator alum, combines AI-led interviews with desktop activity capture and meeting bots to build what it calls "AI-Native Process Intelligence." The company's site claims results like 50% of support tickets auto-routed within two weeks and 70% less manual accounts payable entry at 98% accuracy. These are vendor-published figures, not third-party audits, but they illustrate the ROI framing: compress discovery time, quantify opportunities, then deploy automation against the highest-value targets.
Fractional AI published a case study roughly two weeks ago describing an enterprise discovery project for a Fortune 100 client using Superintelligent's AI Interview Agent. The engagement covered 150 employees in two weeks at approximately $500 in compute costs. The case study is single-source and lacks precise dating beyond the publication window, but it signals that large organizations are at least testing interview-first discovery at scale.
Meanwhile, established process intelligence vendors are grafting conversational layers onto event-log mining platforms. On May 7, Celonis and Microsoft announced an alliance linking Celonis Process Intelligence with Microsoft's Agent 365 control plane. The partnership introduces "Agent Mining," which analyzes agent reasoning and decision logic to identify bottlenecks in pilot deployments. The narrative is explicit: too many GenAI pilots fail because they lack operational context. Conversational discovery supplies that context; process mining quantifies it from system data.
The Governance Wall
The shift from survey forms to asynchronous AI-led interviews is reshaping how enterprises approach operational discovery—assuming they can navigate the governance thicket.
The EU AI Act's transparency obligations take effect in early August, requiring enterprises to disclose AI use in certain employee-facing scenarios. California's employee data provisions under the CPRA, in force since January 2023, impose notice and access rights for recorded interviews. New York City's Local Law 144, effective since July 2023, mandates bias audits for automated employment decision tools—a framework that could apply if interview outputs inform performance evaluations or promotions.
Security posture matters, too. Cisco President Jeetu Patel warned in February that agents require "background checks" and major security investment before enterprises should treat them as digital coworkers. SailPoint, Microsoft, and other identity vendors are building access models and audit trails for agent workforces. As Gartner noted in late April, Fortune 500 companies could be running more than 150,000 agents by 2028. Managing that sprawl will require centralized governance that most organizations don't yet have.
The category is also fragmenting fast. Some vendors focus on internal transformation (Ontora, Linc, Lemma). Others target consultants themselves (Workstuff). Still others emphasize due diligence or exit interviews (Sensay, KS-Agents). The common thread is parallel execution and automated synthesis, but the use cases are splintering.
What Comes Next
The integration trend points toward hybrid discovery: conversational agents surface undocumented workarounds and edge cases, while process mining platforms quantify those patterns from system logs. The Celonis-Microsoft partnership is an early signal. Expect similar moves as vendors blend narrative insight with KPI-grade baselines.
IDC forecasts that by 2027, half of enterprises will use AI agents in some capacity, and agent systems will approach half of AI spending by 2029. Whether interview-first discovery scales into that future depends on factors the vendors don't control—trust, governance, and proof points beyond their own claims.
But the calendar compression alone has captured attention in executive suites racing to operationalize AI before competitors do. Four hours instead of four months is a compelling hook, even if the reality turns out messier than the pitch. Perhaps especially then.
