The dating app built on relationship science — the bridge between the two rebellions reshaping how people meet, plugged into the real places where singles already gather.
Seed round · Consumer social · Confidential — includes live product demo
"Researchers fed 100+ profile measures — traits, values, dealbreakers — into machine learning to predict who would desire whom. It couldn't. Pre-match attraction was essentially unpredictable from anything a profile can hold."
Joel, Eastwick & Finkel · Psychological Science · 2017
Swiping is a bet on the profile. The profile is exactly where the signal isn't.
of Gen Z report dating-app burnout — highest of any age group
of women cite fatigue; 40% blame the lack of real connection
have been ghosted; hollow repetitive chat drives them off
surge in "matchmaker" searches — people pay humans to escape
They're not leaving dating. They're leaving the slot machine.
Forbes Health 2025 · Global Dating Insights 2026
Sitch — AI matchmaker, $5M seed, sells setups at $90–160/pack and sells out. 54% of dating apps now use AI, +333% in one year. Hinge's founder is building the next one.
Timeleft & Thursday selling out across dozens of cities. Run-club singles events +136%, game-night speed dating +400%, singles-event attendance +49% YoY. Even Tinder now throws IRL events.
The AI wave has the intelligence but no real-world moment. The IRL wave has the moment but no intelligence. Arc is the bridge.
TechCrunch · M13 · Eventbrite Meet-Cute Index 2026 · Morning Brew 2026
A partner's traits, looks and stated preferences — the entire content of a profile — are weak predictors of lasting relationship quality.
Commitment, appreciation, responsiveness — dynamics that only emerge through interaction — explain ~45% of the variance. (43 datasets, 11,000+ couples.)
"You can't rank chemistry before two people meet. You can only measure it in motion — and engineer the situations where it forms."
Joel et al. · PNAS · 2020 · full science base: Eastwick corpus, project doc 01
A finite First Glance deck learns your type; active learning picks each face and stops when the model is confident (typically 30–80). The only swiping in the product.
Members choose stylized avatars; match cards lead with behavior and a why-you-two note. Real photos unlock only when both accept — attraction is pre-verified, so decisions focus on what matters. 5 matches/week free, 10 on Arc+.
Talk → Do → Solve. Paired prompt decks, post-date reflections. Chemistry measured in motion, across real situations.
Strong signal → a table held at a partner café or a shared activity. Consented follow-ups close the outcome loop.
One metric no incumbent optimizes: connections that leave the app.
Arc integrates established platforms — Strava clubs, Meetup, Eventbrite, local run-club calendars — to surface running clubs, cycling groups, hiking buddies in every city. Match two people who both run and Arc says: "Saturday club meets 9am at the park." Shared effort, low pressure — the fastest-growing meeting channel (+136% / +400%). Instant IRL inventory, zero ops cost.
Partner cafés & restaurants take the match to a table: held reservations, date-night offers, "first round on us" sponsorships. Venues get guaranteed weeknight footfall; Arc earns a referral fee per seated date plus sponsored placements; daters get a frictionless plan.
Timeleft seats strangers at random. Arc fills the same rooms with people its engine already knows fit.
Launch cities chosen by activity density · platform ToS diligence on roadmap
the swipe leaderboard's happy-date rate — a Thompson-sampling bandit that learns from real outcomes, needing no chemistry prediction at all
Gini of attention received: the swipe "kleptocracy," reproduced in simulation — and dissolved by design
300×300 agents, 26 weeks, outcome model calibrated to Joel/Eastwick/Finkel 2017 · reproducible: analyses/arc_models.py
Buys capacity, not access: 10 matches/week (vs 5 free) and up to 3 simultaneous second-date pairs (vs 1) — always disclosed, with a universal "no multi-daters" consent toggle. $24.99 Gen Z tier at expansion.
Plus sponsored perks. Monetizes date activity, not time-on-app — the line that makes "we win when you leave" survive the P&L.
Ticketed small-group formats; member hosts keep supply cheap. Couple-mode & reports: unmodeled upside.
median member LTV (Monte Carlo, 20k members)
median LTV/CAC · P(>3×) = 62%
CAC payback at $49 monthly contribution
median lifetime — short by design: success-churn is the promise
couples formed in year one; ~1,055 cumulative by year two (liquidity model)
gender band enforced by acquisition throttling — at 40% female share the funnel runs at 80% capacity
Modeled scenarios with stated assumptions — swap in beta cohort data · analyses/results.json
global online dating
intent-driven daters, 25–45
5-yr science-first target
Willingness to pay is proven: Sitch sells $90–160 setup packs out; ~2M people pay Hinge $20–40/mo. Arc prices inside a validated band.
Statista 2026 · Business of Apps · TechCrunch · Match Group filings
AI matchmakers own intake interviews. Incumbents own swipes. Arc owns interaction → date → outcome chains, closed by consented real-world follow-ups.
More arcs → better bandit posteriors → better intros → more happy dates. Plus a venue & activity network that must be rebuilt city by city to copy.
Relationship-science advisors and ML talent recruited with the rarest asset in the field: this outcome dataset. Published Transparency Report keeps us honest — and unmatchable.
Tinder can copy a feature in a sprint. It cannot copy years of outcome-closed behavioral data.
18-month runway: build the interaction engine, sign the first venue partners, launch one activity-dense city past breakeven, and stand up the outcome dataset that becomes the moat. Path to Series A: city breakeven + published Transparency Report.
product & engineering
launch, growth & venue network
data science & research
operations & G&A
Allocations illustrative — tailor before sending
Chosen for how you two would connect — not who's nearby.
Sat 9am · Prospect Run Club · via Strava clubs
8 seats · Casa Nera · 3 people you've met before
That's everything. Nothing to scroll.
· Took a stranger's keys at a wedding — then drove him home
· "I build it wrong first, happily" — the IKEA test
· Organizes the Thursday trail-run picnic, every week
We can't predict chemistry — nobody can. But you're both "better in person" people who go deeper when an evening stalls, and you both run. That's worth three dates.
You chose each other for how you show up. Here's who you'll meet.
Thu 7pm · 4 min from you both · table held · Arc venue partner
Prospect Run Club · Sat 9am · via Strava clubs · activity graph
Escape room, cooking mission, or a build — you'll pick together.
"What's something you've made with your hands that you're quietly proud of?"
"A retaining wall that's secretly a bench. The city never noticed. Sit on it every lunch."
"A bookshelf that leans 4 degrees. It's load-bearing spite at this point."
"When she sketched the bench-wall on a napkin and gave it to the waiter as a tip suggestion."
Date 2 is already on the books: Prospect Run Club, Sat 9am. Shared effort, low pressure.
"We skipped the small talk and ran a 10k instead. Three months in, the leaning bookshelf holds her plants."
— Elena & Sam · matched on Arc
Tiers: 5 matches/week free · 10 + up to 3 concurrent pairs on Arc+, always disclosed, opt-out respected.
All profiles are illustrative and generated for this demo.
Venue & activity data via partner network + public platform integrations.
Let's build the bridge.
ak2170@scarletmail.rutgers.edu · Sources: Joel, Eastwick & Finkel 2017 · Joel et al., PNAS 2020 · Forbes Health 2025 · Eventbrite Meet-Cute Index 2026 · TechCrunch/M13 2025–26 · Match Group filings · Morning Brew 2026 · full corpus in project docs 01 & 08