Anticipated questions from seed and pre-seed investors — answered directly. No spin. If the answer is hard, we say so.
eHarmony uses a questionnaire-based regression model built in 2000 on data from 5,000 married couples. It has 32 compatibility dimensions — but they are all stated by the user. There is no LLM conversation, no visual calibration, no contradiction detection, and no adaptive updating of your model over time.
AIMM's key structural difference: we detect the gap between what you say you want and what you actually engage with, then resolve it. eHarmony takes stated preferences at face value. AIMM doesn't trust stated preferences — it calibrates them.
The second difference: eHarmony is a 25-year-old UX built around form-filling. AIMM is conversational, multi-modal, and fully autonomous — no forms, no swiping, matches delivered.
The psychographic interview is model-agnostic — we target structured output from any frontier LLM (GPT-4o, Claude, Gemini). The prompt chain and output schema are ours; the model is interchangeable.
Critically, the LLM is called once at onboard to build the psychographic profile. After that, all matching runs on vector similarity — no ongoing LLM calls per match. This means LLM cost is a one-time onboard expense (~$0.50–2 per user), not a recurring COGS item.
We maintain prompt compatibility across at least two providers at all times, with automatic fallback. Provider concentration risk is real but manageable — our IP is in the data and vectors, not the model.
The calibration flow presents the user with a curated set of images — generated or sourced from licensed stock — representing variation across the key visual feature dimensions we track (face geometry, style signals, social energy cues, physical presence).
User reactions (like/dislike/neutral) are mapped onto a feature-weighted vector. We run confidence scoring per feature: features with strong consistent signal become "fixed" in the match search; low-confidence features are released to avoid over-constraining the pool.
We do not use real people's photos in calibration — only generated images or consented stock, eliminating the ethical and legal exposure of using profile photos for training.
Year 1–2: pgvector on Postgres — sufficient for tens of thousands of users with proper indexing (HNSW). Zero additional infrastructure cost, familiar stack, easy to reason about.
Year 2–3 scale: migrate to Pinecone or Weaviate for sub-10ms approximate nearest-neighbour search at millions of vectors. This is a solved engineering problem — not a research problem.
The matching query itself is cheap: cosine similarity over a filtered subset (geo, relatability layer). At 100K users, worst-case pool is ~5K candidates per query — well within pgvector's capabilities without specialised infra.
Target onboard time: 18–25 minutes across two sessions. Session 1 (~12 min): psychographic LLM interview. Session 2 (~8–10 min, can be done later): visual calibration. Matches are not released until both sessions complete.
The 18-minute commitment is a feature, not a bug. It self-selects for users who are genuinely serious — the exact user we want. Casual users will drop off, reducing noise in the pool for everyone else.
We communicate this upfront in onboarding copy: "This takes 20 minutes. It's worth it. Most dating app users have spent 20+ hours swiping with nothing to show for it."
Every user onboard costs real money: LLM inference for the psychographic interview (~$0.50–2), image generation for visual calibration (~$1–3), and verification checks for Elite/Concierge. A free tier would burn cash on users with no intent to pay.
More importantly: a free tier pollutes the pool. AIMM's match quality depends on every user having a complete, calibrated profile. A low-commitment free user generates a low-quality vector that degrades matching for everyone. We want 800 serious paid users in Year 1, not 80,000 half-finished free profiles.
Acquisition without a free tier runs on: 7-day money-back guarantee, referral discounts, high-intent SEO ("why do I keep dating the wrong person"), and editorial press. These target users who are already convinced they need a better solution.
The comparison frame matters. Against Hinge ($35/mo), $499 is expensive. Against Kelleher International ($30,000–$45,000/year), $499/month is $5,988/year — a 5–7x discount for a service with deeper psychological profiling and no geographic constraint.
The Concierge user is not a Hinge user who wants a fancier app. They are someone who:
MillionaireMatch, Luxy, and Raya all prove this segment pays. We're targeting the same user with a fundamentally better product at a more accessible price.
B2B is Phase 2 — we do not pursue it until the consumer product has proven LTV:CAC and match quality. The first B2B target (M+18–20) is a niche dating app that has users but lacks deep matching technology. They license AIMM's psychographic engine to upgrade their match quality without building it themselves.
The commercial structure: API access fee ($1,000–5,000/month) + per-query pricing for match calls. A niche app with 10,000 active users running 2 match calls/day = ~$600K ARR to AIMM at $0.10/query.
Secondary targets: HR tech startups doing co-founder or team compatibility matching, and relationship coaching platforms that want to personalise coach-to-client pairing.
Standard referral: give $20 off first month, get $20 off next month. Both parties pay — the discount just reduces friction on the first month.
The deeper mechanic: inviting someone you know directly improves your own match quality. Because AIMM matches on psychographic compatibility, someone whose values and life context you already understand well will produce a richer vector — and if they match with you, it validates the model. This creates genuine word-of-mouth beyond the discount incentive.
They can copy features. They cannot copy the architecture — because the swipe mechanic is their business model. Hinge generates revenue when users stay engaged on the app swiping. A system that delivers 3 matches and then stops needing to be opened is structurally incompatible with their engagement-driven revenue model.
Match Group cannot replace the swipe mechanic without cannibalising their DAU/engagement metrics — the numbers their stock price is valued on. This is the classic innovator's dilemma: the thing that would make their product better would make their business worse.
AIMM's incentive is aligned with users: we win when people find a relationship and leave. No incumbent can say this.
Known uses voice-based AI interviews — single-modal, no visual calibration, no contradiction detection. Sitch uses LLM + human hybrid — which limits scale and keeps costs high.
AIMM's differentiation is the combination of four layers no one else has simultaneously: psychographic LLM profiling + visual attraction calibration + contradiction detection + adaptive behavioral updating. Each layer alone is replicable. Together, they form a compounding model that gets harder to match over time.
We're also the only player targeting the Concierge/HNW segment with ID verification — which Known and Sitch don't touch.
Correct — and we say as much in the market research. The relevant number is the SAM: $2.8B representing the premium dating + AI-driven segment in English-speaking markets where users pay $20+/month. The Concierge segment alone (users who would consider a $5K–$30K traditional agency) is estimated at $400M+ of that.
Our Year 3 SOM of $18.5M ARR represents 0.66% of the SAM — a very conservative target. Hinge grew from launch to $400M ARR in under 5 years in a saturated market. We're entering a category that effectively doesn't exist at consumer scale yet.
This is the hardest operational challenge and we don't pretend otherwise. Our mitigation strategy:
This is the question we want investors to ask, because it proves the incentive alignment is real.
The honest answer: yes, successful users churn. But this is true of every matchmaking service — the business model survives because the pool of single, relationship-seeking adults continuously regenerates. Divorces, breakups, people coming out of long relationships — the TAM replenishes itself.
The deeper answer: success stories are the most powerful marketing tool we have. Every user who finds a relationship through AIMM becomes a word-of-mouth case study. The referral mechanic compounds. A platform with a reputation for actually working will attract more new users than it loses to success.
Compare to traditional matchmaking agencies: Kelleher International has operated for 30+ years on exactly this model.
For the mass market — yes, probably. That's why it's opt-in, Concierge-tier only. Users who choose this tier are explicitly opting in to verification as a trust signal. They want to know that the people they're being matched with are who they say they are.
The comparison frame matters. Traditional matchmaking agencies require in-person background checks, income verification, and ID as standard practice. Luxy requires tax returns and a driver's licence. We're not introducing a foreign concept — we're digitising what the premium market already expects and accepts.
Implementation: Stripe Identity handles the document scan — AIMM never stores the document, only the verification result. The profile shows a "Certified" badge. Balance is never shown, only income tier. Users control their data and can request deletion at any time.
This is a legitimate concern and one we take seriously. The risk is that the attraction vector, trained on individual reaction patterns, could narrow matches to a single archetype — or worse, reflect and amplify existing societal biases in physical attraction preferences.
Our mitigations:
We present all outputs as approximations, never verdicts. The model is a tool that surfaces possibilities — the user makes every decision.
Strategic acquirers fall into three categories:
IPO path: less likely at seed stage to project, but a $50M+ ARR business with high gross margins and a B2B API component is a credible standalone public company.
Comparable exits: Hinge acquired by Match Group for $51M (2018) → $400M ARR by 2023. Thriving Abroad (niche) → smaller multiples but fast. We're building toward a $500M–$1.5B exit at scale based on 8–12x ARR multiples in premium SaaS/consumer.
Fair. Here's the honest case:
What we're asking for is $3M to build the product, prove match quality NPS ≥ 70, and demonstrate LTV:CAC ≥ 5x. Those are the Series A gates. If we hit them, the next round is de-risked. If we don't, the downside is bounded.