Every founder we talk to draws the same picture on a whiteboard. Application, screening, interview, offer. Four boxes, an arrow between each one. The whiteboard never lies about what people think the work is. It also never shows where the work actually breaks. In nine out of ten startup hiring funnels, the bottleneck is screening, and it is not even close.
This essay is about why screening is the only stage that controls outcomes, why the standard applicant tracking system optimizes everything except screening, and what to change so a five-person team can run a hiring funnel that beats a Series C company with a head of talent.
The hiring funnel is not four equal boxes
The hiring funnel looks symmetrical on the whiteboard. It is not symmetrical in practice. One stage decides almost every downstream metric: screening. Whoever you let through screening is the pool you interview. Whoever you interview is the pool you offer. Whoever you offer is who joins. If screening is bad, you are running expensive interview loops on candidates who were never going to be hires. According to the SHRM 2025 Recruiting Benchmarking Report, the average US time-to-fill is around 44 days, and the bulk of that is the screening-and-shortlist window, not the offer or onboarding window. The single highest-leverage thing a startup can do for hiring speed is fix the screening layer, not buy a faster scheduler or a glossier careers page.
The reason most teams miss this is that screening feels boring. Calendar tools and offer letter automation feel like product. Screening feels like reading. So tools focus on the parts that are easy to demo, and the part that decides outcomes is left as manual work for whoever has the lowest standing in the room that week.
Why most ATS optimize the wrong layer
Most applicant tracking systems were built between 2008 and 2016 for enterprise HR teams whose problem was not screening quality. It was workflow visibility. A 5,000-person company with twenty recruiters needs an audit trail, structured stages, EEO reporting, and a single source of truth for every applicant. So legacy ATS optimized for those things, and screening got handled with a Boolean keyword filter bolted on top. That filter is the layer where almost every startup hiring funnel actually leaks.
Greenhouse, Lever, Workable, and Ashby are good products at what they were designed for. They are pipeline managers. The pricing reflects it. Workable starts at $149/mo for under-20-employee teams and scales with headcount, not hire count. Greenhouse, per buyer-reported data on PriceLevel, runs roughly $50–$150 per seat per month plus four-figure implementation fees. You are paying for a workflow product whose differentiator is not screening quality. Then, on top of that price, you still need to read every resume yourself.
The Harvard Business School project on Managing the Future of Work has a name for what this produces: hidden workers. Joseph Fuller's 2021 study estimated 27 million qualified Americans get filtered out by automated keyword screening for reasons unrelated to whether they could do the job. That is the cost of optimizing the wrong layer.
What we learned at Amazon about ranking systems
Before CurriculoATS, our founder Dev spent years at Amazon working on search and recommendations. The lesson from those systems applies almost directly to hiring. When the input is messy and the user is busy, the only thing that matters is the ranking function. If the top ten results are good, the user is happy and conversion is high. If the top ten are wrong, no amount of UI polish saves the experience. You can have the cleanest faceted filters in retail e-commerce and still lose because the ranker is bad.
Hiring is the same shape. The recruiter or founder will look at the top ten or fifteen resumes for a role. The other 200 might as well not exist. So the only question that matters is whether the right candidates are in those top ten. Every other feature, the email templates, the Slack integrations, the offer letter PDFs, sits downstream of that one decision. If the ranker is keyword-based, the top ten will be whoever wrote the resume most strategically. If the ranker reads for outcomes (revenue, teams, systems shipped, problems solved), the top ten will be whoever did the most relevant work.
This is why we built CurriculoATS around a multi-signal evaluation that produces a 0–100 composite score with a written paragraph explaining the reasoning for every candidate. The paragraph is the part founders care about. It lets you trust the ranker without trusting it blindly. Read more about how the scoring works on our Impact Scoring page.
The metrics every founder should track on the funnel
Most startup founders track time-to-fill and stop there. Time-to-fill is a lagging indicator that tells you the funnel was bad three weeks ago. The leading indicators that actually predict funnel health are easier to instrument and rarely measured. Track four: applicants-to-first-round ratio (under 5% means screening is throwing away signal, over 12% means screening is too loose), first-round-to-second-round conversion (under 30% means the shortlist has the wrong people on it), reasoning-paragraph review time per candidate (over 90 seconds means the model is not adding leverage), and shortlist overlap between AI and human (under 60% means either the AI or the human is wrong, and the founder needs to debug which). The McKinsey research summarized in The shape of talent in 2023 and 2024 shows skills-based hiring adoption climbing from 40% in 2020 to roughly 60% in 2024, and the operational signal that separates teams doing it well from teams doing it nominally is the second metric: first-round-to-second-round conversion. Teams running outcome-based screening tend to land at 40-55%; teams running keyword-based screening tend to land at 18-28%. The gap is the model, not the recruiters. Reviewing these four numbers weekly catches funnel leaks fast enough to fix them in the same hiring cycle, rather than waiting for time-to-fill to ring an alarm a month after the damage is done.
How a startup founder fixes screening this week
If you are a founder who is hiring without a recruiter and your funnel is leaking, the changes that move the needle are smaller than you think.
- Stop ranking by keyword density. If your current ATS is sorting by token match, every candidate who has applied to more than five jobs has gamed it. Switch to a system that ranks by outcome (revenue generated, teams led, systems shipped) or read every resume manually. There is no middle ground that works.
- Write the JD as outcomes, not duties. Replace "responsible for X" with "in the first six months, this person will ship Y." That document is what your screen should match against. Outcomes-to-outcomes matching is roughly an order of magnitude more useful than duties-to-keywords matching.
- Cap the shortlist at twelve. Twelve is small enough that you can read every reasoning paragraph in one sitting. Past twelve, attention falls off and decisions get random.
- Keep one rejection bar visible. Write down the two specific things that disqualify a candidate. "No production ML experience" or "no founder/0–1 history." If you can't write two things, you do not know the role yet.
- Review the funnel weekly, not quarterly. Time-to-fill at startup pace is days, not weeks. A monthly review is too slow to spot a leak.
None of this is novel. What is novel is that almost no team is actually doing all five at once.