Every application gets a full read. Every rejection gets a reason.
EvalLens gives your screening committee a first pass it can trust: every application read in full and scored against your rubric with page-level evidence. Your team reads reports instead of raw applications — and humans make every call.
Rerun your last intake.
Don’t take a vendor’s word for how it would perform on your cohort. Test it against decisions you already made.
Send 20–50 applications from your previous cohort. Free up to 20.
Within 48 hours you get the evidence-linked reports, the ranked board — and a side-by-side against your committee’s actual decisions, including any rejected teams the panel would have flagged for interview. Larger batches are scoped with the managed pilot.
Your last cohort is the ground truth you already own. The retro-test measures agreement and misses on it — before you commit to anything.
Hundreds apply. A handful get a real read
- 150–2,000applications per cohort — from a regional program's team of two to a global intake's team of five, always in a two-to-three-week windowthe pilot measures yours
- 2–3 minthe read a founder's six months of work actually gets under deadline pressuredeadline math
- #447 ≠ #1late applications get a different read than early ones — recency and fatigue, the reviewer biases every program director recognizesreviewer-bias literature
- 0substantive feedback most rejected founders receive. They remember, and they post about itfounder forums, passim
“Why did this team score 7.8?”
Every score unpacks into a chain your committee can check — and challenge — before it enters the result.
The next time a partner asks “why did we pass on them?” — you’ll have the page number. Independent AI reviewer roles, no score peeking; disagreements escalate, never average away; incomplete evidence moves scores down, never up.
Five steps. Your committee stays in charge
EvalLens prepares the structured first read. Your committee makes the cohort decision.
- 01
Keep your intake
Founders keep applying through F6S, Typeform, YouNoodle or your own form — we ingest the batch: structured answers, decks, video links. Batch export (CSV + files) works today, platform connectors are scoped during setup, and committee scores live in EvalLens or sync back to your sheets. Or use our branded application page. Nothing changes for founders. You get: your stack, untouched.
- 02
The full read — every application, every page
Form answers and decks are read in full, coverage logged so you can verify what was read. Application #1 and #1,400 get exactly the same rules, on the same day quality. You get: a coverage log per application.
- 03
Independent AI reviewer roles — not people — score on your rubric
Six adversarial AI reads — team, market, feasibility, traction — that can't anchor on each other: each scores independently, without seeing the others' numbers. Evidence first, then the score: incomplete evidence moves scores down, never up. Where reviewers split by more than a band, the case escalates to your team as an open question. You get: committee-grade discipline on all 1,400.
- 04
Your committee decides — we feed it, not replace it
EvalLens is the first-pass layer under your existing process: your 15 mentors, partners and screeners keep scoring and arguing exactly as they do — starting from evidence-linked reports instead of raw applications. Confirmations and overrides are logged; the shortlist is built from human decisions. You get: Selection Day time spent on the borderline 30.
- 05
Feedback for every founder. A record for every board.
Feedback is drafted from the evidence and approved by your team before anything is sent — every rejected founder gets a real answer under your name, safely. The program keeps a selection memo and full decision log: the audit trail LP-funded and publicly funded programs are asked for. You get: reputation compounding + an audit-ready record.
Day 0 vs Selection Day.
Your process, as it is
Your criteria and weights — and if the rubric lives in your committee's heads, the setup session turns how you already decide into a written rubric. It's yours to keep either way. Plus your existing intake form and this cohort's batch.
Everything read, everything defensible
Every application fully read with a coverage log · an evidence-linked report and interview questions per team · a ranked board with the borderline flagged · a shortlist built from your committee's scores · approved feedback for every founder · a selection memo and decision log for the board.
Founder data is a liability we take seriously.
Never trained on
Your applicants' documents are processed only for your evaluation — never used to train models. That's the sentence your founders and counsel are looking for.
Closed perimeter
Server-side processing, project-level isolation, reports only in an authorised workspace — no public links. GDPR-aligned processing terms (DPA) with every program.
Your rules at scale
Enterprise track: dedicated processing, your approved models, retention periods you define — scoped in a security review before the pilot, not after.
Priced per intake, not per seat — cheaper than the skim. Failed runs are never billed; education discount for university-affiliated programs. See pricing or book a call.
Asked by every program director
- Does the AI pick our cohort?
- No — by design. AI scores are advisory; the shortlist and ranking are built exclusively from your committee's decisions, with every confirmation and override logged. There is no mode where the machine decides.
- Where does our screening committee fit?
- Exactly where it is now. EvalLens is the first-pass layer underneath: your mentors, partners and screeners keep scoring, arguing and running Selection Day — starting from evidence-linked reports instead of raw applications. We feed your committee; we don't replace its workflow.
- Our applications aren't decks — they're forms, Looms and links.
- That's the normal case for accelerators, and it's supported: structured form answers and decks are read in full today; video responses and external links are handled on the program-scale track, configured during setup. Applications are read in the language they arrive in.
- What about bias — non-native English, unconventional founders?
- The panel applies the same written rules to every application, never fatigues, and logs the evidence behind every score — so bias becomes inspectable instead of invisible. Blind first-pass configurations (identity fields masked) are available. And the retro-test lets you audit outcomes against your own past decisions before you rely on it.
- Do you train on our data? Which models touch it?
- Never. Applicant documents are processed solely for your evaluation under a GDPR-aligned DPA, with project-level isolation. Model routing is server-side; on the enterprise track evaluation runs on your approved models — including inside your environment.
- How do we start without risking a live cohort?
- Two zero-risk paths: the 48-hour retro-test on a closed intake, and a parallel pilot on your next one — your team works as usual while EvalLens reads the same applications, and you compare time, coverage and shortlist agreement at the end. The managed pilot is a fixed-fee engagement, scoped on the call.
Run the retro-test on your last cohort.
30 minutes with you: we map your rubric, pick the batch, and you see agreement-and-misses on decisions you already made. 48 hours later, you'll know. Free through August 31, up to 20 applications from a past cohort.