Reval's first product attacked the sort of corporate chore that survives mostly because everyone assumes someone else will do it: the reference call. A recruiter wanted three references, then spent an hour or more emailing, scheduling, dialing, transcribing and turning friendly praise into something a hiring manager could use. In early 2024, Reval offered a shareable link instead. A reference could talk asynchronously with an AI agent, from any device, while the software asked follow-ups and assembled the result.
It was a tidy wedge into a messy market. It was also downstream. By the time references enter the process, a company has already written a job, found applicants, judged résumés, run screens and decided whom to advance. Reval could save time without controlling the outcome its customers actually bought: a good person, hired quickly.
The bot leaves the reference call
The first thing to fail was not the reference checker. Publicly, Reval kept selling and improving it. What broke was the assumption that one efficient task was enough territory. The company moved outward until its agents could take a job description, pull out the requirements, A/B test anonymized listings, match candidates from an internal pool, run outbound sourcing, conduct tailored screens, check references, score the results and schedule the people a client wanted to meet.
The new pitch was blunt: paste in an open role and receive high-quality, high-intent candidates ready to interview within 48 hours. Reval called itself an "agentic recruiting firm," a phrase that mattered because the company was not merely licensing another dashboard. It was taking responsibility for the top of the funnel and charging against the result.
“The product is the outcome.”Seth Tilliss, co-founder and CEO
That distinction placed Reval between categories. It competed with traditional agencies, internal talent teams and a stack of point products for sourcing, outreach, screening, scheduling and references. Its expertise was less a proprietary HR ritual than the orchestration of those steps: keep the agents running, preserve candidate context and turn a crowded applicant pool into a smaller set of conversations worth having.
The closest alternative depended on who held the budget. A talent leader could hire another recruiter, pay a contingency agency, assemble tools around an applicant-tracking system or ask the existing team to keep doing the work by hand. Reference specialists could automate the last verification step. Newer AI recruiting platforms could attack sourcing and screening. Reval's differentiation was to bundle the sequence and accept placement risk. A client did not have to decide which sourcing database or interview bot deserved another seat; it had to decide whether the submitted person deserved an interview.
The price of an interview-ready person
Reval's public direct-hiring price began with a one-time job kickoff fee from $50. If the client hired someone Reval introduced, the success fee was 10 percent of first-year base salary. The company compared that with the 20 to 30 percent commonly charged by traditional agencies. It advertised no retainer and a 90-day guarantee if a hire left or was terminated for cause, subject to the terms of the arrangement.
Public success-fee comparison
The economics had changed before. One earlier founder post promoted a model where clients paid $250 to interview a candidate and nothing when they hired, with an average hire said to cost less than $2,000. The later contingency model is evidence of learning, not inconsistency to hide. Reval was testing where customers felt value: access to a promising conversation, or the completed hire. It ultimately planted the meter at the outcome.
Recruiter-facing features could also carry recurring subscription fees under Reval's terms, though a fixed public subscription schedule was not available. That hybrid makes sense. A founder filling two roles wants a shared-risk search. A recruiting agency wants operating leverage it can use repeatedly across clients.
They built software by running the service
The smartest move in the Reval story came before its software reached outside recruiters. The team used the agents to operate Reval itself. According to Tilliss, that internal operation onboarded 50 direct clients, handled 150 active requisitions, screened 30,000 candidates, booked hundreds of interviews and made a few dozen hires. Each awkward intake, weak match and unexplained rejection became product material.
Then the team put the same back office to work for recruiters. In a private beta, it completed 10 contingent hires in March 2026. A later update said 25 firms were running on the system after two weeks. Reval described the product as another teammate working around the clock - one that delivered booked interviews rather than a spreadsheet of scraped profiles.
This is what changed their mind about the size of the product. Direct recruiting exposed a larger repeated workflow than reference checking ever could. It also revealed a second customer: not only employers who needed hires, but recruiters who wanted to handle more clients and requisitions without adding the same amount of headcount.
“Recruiting capacity shouldn't scale with recruiting headcount.”The shared Reval and Metaview thesis
Why the candidate pool mattered
Reval arrived as generative AI made applying to jobs cheap. A candidate can tailor a résumé and apply hundreds of times. Employers receive more material, but each document carries less reliable signal. Tilliss argued that strong candidates get buried with fake profiles, mass applications and merely plausible fits. Reval's answer was to screen once, capture preferences, skills, work samples, compensation, location and references, then reuse that richer record across fitting opportunities.
Its public FAQ described a pool of nearly 30,000 high-intent candidates. Natural-language talent search let a recruiter describe whom they needed rather than assemble filters. Feedback on rejected candidates flowed back into later matching. For candidates, the promise was fewer duplicate screens and a faster route to a human. For hiring teams, it was a claim that the pile had already been cleaned.
FabFitFun, LangChain and Enver Studio were among the customers publicly named. Metaview said Reval had worked with more than 100 enterprises and recruiting agencies. That is meaningful evidence of use, though not proof that every role or customer achieved the promised speed or savings.
There was a cultural clue in the way the team talked about those numbers. The public updates counted requisitions, screens, booked interviews and hires, not model benchmarks. Reval appeared to organize around operating throughput: use the agents internally, watch where humans intervened, and revise the workflow. The rhythm is closer to a staffing desk with software engineers beside it than to a conventional SaaS team shipping a feature and waiting for usage charts. That operator posture also made the company easier for Metaview to understand. Its proof lived in completed work.
The acquisition was a shortcut in both directions
On August 11, 2026, Metaview announced that it had acquired Reval for an undisclosed price. Reval's founders, Tilliss and CTO Aditya Gupta, joined a company already serving more than 5,000 organizations. Metaview wanted to accelerate fillmore, an AI coworker designed to own the operational top of the hiring funnel from sourcing through booked screening calls.
Metaview brought distribution and an existing recruiting platform. Reval brought three years of operating scars, customer relationships and a compact thesis about selling capacity rather than seats. The fit was unusually legible: one company needed to move faster on an agentic product; the other wanted to take its approach to thousands of organizations. Reval's website now points visitors toward the fillmore waitlist.
What a builder can steal
Start with the hated chore. Reference checks were narrow, measurable and full of scheduling friction. A wedge does not need to be the final company.
Become the operator. Running a direct service forced Reval's software through real requisitions, messy preferences and consequential decisions before it was packaged for recruiters.
Charge where value becomes visible. Reval experimented with per-interview pricing, then settled publicly on a lower contingency fee tied to a hire.
Keep the human handoff explicit. Agents did intake through scheduling. Employers still interviewed, evaluated and made the hiring decision.
Where this playbook does not work
- The role is poorly defined, and the client cannot turn taste into must-haves, preferences and useful pass feedback.
- The candidate market is so small or relationship-driven that a reusable pool and automated outreach add little.
- The employer reviews slowly. A fast top of funnel cannot repair a stalled human process.
- Biased historical preferences are treated as truth. A learning loop can repeat bad judgment as efficiently as good judgment.
- Candidates do not consent to or trust AI-led screening and reference handling, especially where local employment or privacy rules demand additional care.
Reval did not prove that hiring can be fully automated. Its own design stopped at the human interview, and its terms made clients responsible for employment decisions. What it demonstrated was more specific: a small team could use agents to compress the repetitive work around that decision, learn by delivering the service itself and sell an outcome at agency-shaped prices.
The amusing part is that the original check mark never disappeared. Reference checks survived inside the broader product. Reval did not abandon its first idea so much as demote it - from the whole pitch to one verified line in a candidate package. That is often what a useful pivot looks like: not a bonfire, just a feature finding its proper size.