I sat down with Serge and Eric from Build Different Venture on April 9, 2026 to explore potential B2B health startup ideas. This write-up captures our research into each one: the investment thesis, why now, how it could work, and what the real risks are. A fourth idea grew out of my ongoing work with Otto, where I've been building AI-powered health data infrastructure and learning firsthand what practitioners and consumers actually need.
April 2026 · Kevin Ho
Build Different's superpower is in B2B. They found what I'm working on with Otto interesting, but they're not quite ready to put their arms around the idea yet. So we agreed to keep exploring everything on the table and compare notes. What follows is an honest look at each idea, with the goal of figuring out which (if any) is worth pursuing further.
The idea: give consumers a free health score (like FICO, but for health), build engagement around improving it, then monetize by connecting them to providers and wellness services. Credit Karma proved this model in finance and was acquired for $8.1B on ~$680M in annual affiliate revenue from 110M members. We wanted to understand whether the same playbook could work in health.
| Shift | What Changed |
|---|---|
| Regulatory unlock | The 21st Century Cures Act (enforced Sep 2025) now requires payers to expose claims and clinical data through FHIR APIs. For the first time, a third-party app can pull someone's medical history with their consent. This is the health equivalent of the Fair Credit Reporting Act access that made Credit Karma possible. |
| Sensors everywhere | 528M smartwatches globally, 92% using health tracking. Continuous heart rate, sleep, SpO2 all available through HealthKit and Health Connect. No clinic visit needed. |
| Cultural moment | GLP-1 changed the conversation. About 12% of US adults have used or are currently on a GLP-1, and 1 in 4 consumers plan to start one soon. Health metrics have entered mainstream culture. 45% of Gen Z and 48% of Millennials are already using AI for health information. |
| The science is there | AI biological age models now predict mortality more accurately than epigenetic clocks. PhenoAge and Framingham are clinically validated. The science exists. Consumer packaging doesn't. |
Score creation: We'd combine wearable data (activity, sleep, HRV), lab results (via FHIR APIs), and self-reported lifestyle factors into a composite "Health Score" with sub-scores for metabolic, cardiovascular, and recovery. No universal health score exists today, which is actually a feature: the company that creates the standard owns the category. We'd publish the methodology openly to build trust.
Engagement loops: Quarterly lab integrations for score updates, daily wearable sync for micro-feedback, personalized challenges, and GLP-1/medication tracking. The score gives people a reason to come back, which is the thing most health apps fail at.
Three different customers we could serve:
The US digital health market sits at $88-157B in 2025, growing 12-20% annually. At scale (50M users), even at Credit Karma's conservative ~$6/user/year, that's $300M+ ARR. Health product marketplace economics could push ARPU to $15-25. Longitudinal health data compounds into a real moat over time. Comparable exit: CK at $8.1B with 110M users works out to ~$74/user.
| Risk | Detail | How We'd Address It |
|---|---|---|
| No universal score | FICO is standardized and universally accepted. Nothing equivalent exists in health. Others have tried: Castlight IPO'd at $2B then cratered, Humanity's "H Score" stayed niche. | We'd have to create the standard. Publish the methodology openly, partner with academic med centers for clinical validation. |
| Retention cliff | Health app 30-day retention is 3-12%. Health is episodic, not a daily habit like checking your credit score. | Wearable sync for daily micro-feedback, quarterly lab score updates, GLP-1 tracking as engagement hooks. |
| Platform risk | Apple and Google control health data APIs and could build a competing score natively. | Apple has historically been a platform (HealthKit), not a marketplace. Building the monetization layer isn't in their DNA. |
| Regulatory | The Anti-Kickback Statute applies to referrals involving federal healthcare programs. | Structure as advertising fees (the CK model), not per-referral fees. Build health-specific legal architecture from day one. |
| Health literacy | Only 12% of US adults have proficient health literacy (NAAL). | The score has to be radically simple. A single number with color-coded sub-scores, not a clinical report. |
Buy Now Pay Later for elective medical procedures. The broad horizontal market is already dominated by Cherry ($2B+ valuation, 60K+ providers) and CareCredit (12M cardholders, 266K locations). The interesting angle here isn't competing head-on. It's a vertical-specific BNPL that goes deep into segments Cherry hasn't fully penetrated.
| Shift | What Changed |
|---|---|
| GLP-1 financing gap | 48M Americans expect to start GLP-1 in 2026. Annual out-of-pocket costs sit at $3-4K+. Meanwhile, insurance coverage is actually shrinking: Blue Cross MA dropped weight loss coverage, CA Medi-Cal ended it in Jan 2026. The Medicare demo doesn't start until July 2026. The gap between demand and affordability keeps widening. |
| Cash-pay medicine growing | Longevity clinic market projected at $6B (2026) with 12.5% CAGR. Consumer healthcare financing is growing at 23% CAGR. Total US health spending surged to $5.3T. |
| Regulatory window | The CFPB withdrew its BNPL interpretive rule. The medical debt credit reporting rule was vacated (July 2025). Federal oversight is at a low point, which gives new entrants runway to establish themselves. |
| Cherry's breadth = weakness | 731 employees spread across dental, aesthetics, vet, and wellness. Trustpilot reviews show complaints about account friction and high financing costs. When you cover everything, you can't go deep on any single specialty's workflow. |
Vertical wedge strategy: Pick one underserved vertical and go deep with purpose-built provider tooling.
| Vertical | Opportunity | AOV |
|---|---|---|
| GLP-1 / Weight management | Largest TAM ($100B drug sales projected by 2030), recurring monthly payments, predictable default profile, massive unmet financing need as insurance pulls back | $3-4K/yr |
| Fertility | $15-30K per IVF cycle, emotionally driven purchase, existing players (Future Family, Bundl) are small. PatientFi only manages ~40% approval rates. | $15-30K |
| Longevity | $6B market, high-AOV cash-pay, affluent but price-sensitive consumers, subscription models create recurring financing opportunities | $5-25K/yr |
Capital-light model: Partner with embedded lending infrastructure (Unit, Bond, Galileo) to avoid carrying balance sheet risk. Merchant discount rates in medical BNPL run 2-8%. We'd build deep integrations with vertical-specific practice management software, something Cherry's generic API can't match.
GLP-1 financing alone: if 10M patients need financing at $4K average, that's a $40B addressable market. At 5% merchant discount, even 1% capture = $20M revenue. Multi-vertical expansion into fertility, longevity, and mental health creates a $50B+ combined financing TAM. Exit comparables: Cherry at $2B+ on ~$51M revenue (~39x multiple). CareCredit/Synchrony is a $20B+ franchise.
| Risk | Detail | How We'd Address It |
|---|---|---|
| Cherry expands | Cherry has the capital (Series C), 731 employees, and is already adding verticals. They could launch a GLP-1-specific product in months. | Lock in 500-1,000 providers in one vertical before Cherry takes notice. Speed is the only moat here. |
| Thin margins | Industry-wide BNPL unit margins are around ~1% of GMV and declining. Default rates of 1.2-2.5% plus cost of capital squeeze margins hard. | Target specialty verticals where providers have higher willingness to pay (fertility, longevity). Higher merchant fees = better unit economics. |
| GLP-1 price collapse | Manufacturer price cuts and compounded generics could push monthly costs below $50, which would eliminate the financing use case entirely. | Don't build a single-vertical company. Diversify early across verticals. |
| Capital intensity | Even with lending-as-a-service partners, we'd still need warehouse facilities or credit lines to scale. This structurally favors well-capitalized incumbents. | Pre-commit a bank sponsor or lending partner with medical-specific underwriting appetite before trying to scale. |
Not a head-on fight with Tempus. Instead, a narrowly scoped AI precision medicine platform targeting segments Tempus structurally underserves, specifically pharmacogenomics-led primary care and preventive health. The idea is to assemble a differentiated data asset from consumer and outpatient sources rather than competing for hospital oncology data.
| Metric | Value |
|---|---|
| Revenue (2025) | $1.27B (83% YoY growth) |
| Market cap | ~$7.6B |
| Pre-IPO funding | $1.3B+ |
| Profitability | Still GAAP unprofitable. -5.56% EBITDA margins in Q1 2026. Data segment gross margins slid from 76.8% to 69.7%. |
| Data moat | 40M+ records, 1.5M matched clinical + genomic, 7B clinical notes, 7M pathology slides, 7,000+ physician partners |
| Focus | Oncology (primary), cardiology, neurology. Underserves: primary care, preventive/wellness, community hospitals. |
| Shift | What Changed |
|---|---|
| AI costs collapsed | Foundation models for biology (AlphaFold 3, ESM, NVIDIA Evo 2) are open or cheap. Preclinical AI cost reductions of 30-70%. LLMs can synthesize clinical literature, lab results, and genomic variants into actionable reports at a fraction of what this used to cost. |
| Tempus is big but bleeding | Despite $1.27B in revenue, they're still unprofitable with operating expenses outpacing growth. The "do everything" approach carries real execution risk that a focused challenger could exploit. |
| Regulatory barrier dropped | The FDA's LDT rule was struck down in 2025 and they chose not to appeal. CLIA compliance without FDA device clearance is the lowest regulatory bar we've seen in years. |
| PGx hitting an inflection | The PGx market reached $8.14B in 2025, growing at 10.3% CAGR. VC funding in PGx startups surged 44%. Pre-emptive panel testing is moving from oncology into primary care and psychiatry, exactly where Tempus is thinnest. |
Wedge: PGx + AI interpretation for primary care. Tempus's neuropsychiatry dataset (~100K patients) is a side business to their $955M oncology engine. A focused entrant could own the PCP prescribing workflow: mouth swab, AI-powered drug-gene interaction report, EHR integration.
Build the data asset from outpatient + consumer sources. Instead of fighting for hospital genomic data, we'd aggregate consumer bloodwork (from labs like Quest via API), wearable data (the pharma wearables market hits $3.98B in 2026), and patient-reported outcomes. Cheaper to acquire, and uniquely valuable to pharma for real-world evidence studies.
Monetize through pharma data licensing. Tempus's Insights segment grew 38% in 2025 and 69.5% in Q4, which proves the model. A curated PGx outcomes dataset linked to consumer health data would be genuinely differentiated. Pharma pays premium prices for real-world evidence connecting genotype to treatment response.
Precision medicine market projected at $237B by 2031 (13.6% CAGR). AI in precision medicine alone reaches $60B by 2035. Precedent exits are strong: Roche acquired Flatiron Health for $1.9B and Foundation Medicine for $2.4B. A PGx-focused data platform reaching $100-200M ARR within 5 years would be a strategic acquisition target for any top-20 pharma company.
| Risk | Detail | How We'd Address It |
|---|---|---|
| Capital intensity | Even a "lean" approach needs CLIA lab partnerships, genomic sequencing, and data engineering. Expect $30-50M to reach meaningful data scale. | Partner with existing CLIA labs instead of building our own. Use the LDT pathway. Focus spend on data/AI, not wet lab infrastructure. |
| Tempus expands into our wedge | Tempus already has PGx in neuropsychiatry and is building out cardiology, radiology, and infectious disease. | Move fast in primary care PGx. Tempus structurally deprioritizes it because test pricing ($200-500) is 10x lower than oncology. |
| Talent scarcity | Genomics + ML + clinical is one of the hardest hiring trifectas in tech. We'd be competing with Tempus, Illumina, and big pharma AI labs. | Go remote-first. Recruit from academic labs where people have deep domain expertise but are undercompensated by industry standards. |
| Data network effects | Tempus's millions of clinical records create compounding AI advantages. A new entrant has to find data Tempus can't easily buy. | Consumer health + wearables + PGx outcomes is structurally different from Tempus's hospital-centric data. They can't easily replicate it. |
This idea grew directly out of what I've been building with Otto. Over the past year, I've built an AI health platform that extracts 85 biomarkers from lab report PDFs, calculates biological age (using PhenoAge and KDM algorithms I've published as open source npm packages), generates evidence-backed recommendations with PubMed citations, and runs 365-day health simulations. I already have 80% of an interpretation engine built. The question is whether repackaging it as a B2B SaaS for clinicians is the right move.
Functional medicine (FM) is a branch of medicine focused on root-cause diagnosis through extensive lab testing. These aren't your typical annual physicals. A single new FM patient might get a GI MAP (84 markers), an organic acids test (78 markers), a DUTCH hormone panel (20+ markers), and a comprehensive blood panel (40+ markers), all of which need to be cross-referenced against each other. One practitioner training resource claims a first patient analysis can take up to 10 hours. An IFM survey of 7,000 practitioners found 71% of first visits run 60-120 minutes, and over half see 5 or fewer patients per day.
The core complexity: FM practitioners use "optimal" ranges rather than standard lab reference ranges. Standard ranges reflect population averages (including sick people). Optimal ranges are narrower and designed to catch dysfunction earlier. This requires pattern recognition across body systems (thyroid-adrenal-gut interplay, for example) that isn't taught in medical school. It's the kind of thing that takes years of clinical experience to do well.
Honesty check: the "10 hours per patient" number comes from practitioner training content, not peer-reviewed research. There's no rigorous published study quantifying exact time spent on lab interpretation alone. The pain is consistently described across practitioner blogs, training materials, and tool marketing, but it's practitioner-reported, not formally measured. That's a gap we'd need to validate ourselves.
This space isn't empty. There are tools addressing parts of this workflow, and understanding what they do (and don't do) matters:
| Tool | What It Does | Pricing | Gap |
|---|---|---|---|
| BloodGPT | Upload any blood test PDF, get AI interpretation in <2 min. Both B2C (500+ subscribers) and B2B (pending medical device cert). Founded 2024, Cyprus. Raising $1M at $10M valuation. | $0.30-0.96/report (volume-tiered). B2C annual subscription also available. | General blood tests, not FM-specific. No optimal ranges, no multi-test cross-analysis (GI MAP + DUTCH + blood panel together). |
| Kantesti | Claims 50K+ healthcare professionals, 2M+ analyses, 127 countries. White-label and EMR API available. | Free tier, single reports from $4.49, enterprise custom pricing. | General purpose. No FM-specific optimal ranges or cross-panel reasoning. |
| OptimalDX | Blood chemistry analysis with functional/optimal ranges. Established, training-focused. | $197/mo unlimited. | Rule-based, not AI-powered. No multi-test cross-analysis. |
| LabDx | Automates multi-test analysis (GI MAP, DUTCH, OATs) into single PDF reports with clinical guidance. | Unclear (Skool community access). | Niche, rule-based, small operation. Not AI-powered. |
| SelfDecode Labs | 1,500+ lab markers with genetic context, AI assistant (DecodyGPT). Free for practitioners, revenue from client subscriptions. | Free for practitioners. | Consumer-first product. B2B is secondary. |
The pattern is clear: general AI blood test tools exist (BloodGPT, Kantesti) but they don't understand FM-specific optimal ranges or cross-panel reasoning. FM-specific tools exist (OptimalDX, LabDx) but they're rule-based, not AI-powered. Nobody has combined AI with FM-specific clinical depth yet.
| Shift | What Changed |
|---|---|
| Longevity went mainstream | Bryan Johnson's Blueprint raised $60M from celebrity investors. Peter Attia's Outlive sold millions of copies. The longevity market hit ~$30B in 2025 growing at 8%+ CAGR. Consumer demand is pulling more and more practitioners into this space. |
| Practitioners professionalizing | IFM launched its first formal certifying board (IBFMC) in Sep 2025, which signals the field is maturing. DPC clinics, a natural adjacent buyer, grew to a $70B market with 58% employer-sponsored memberships. |
| LLMs crossed the clinical threshold | Stanford Medicine built a Claude-powered lab interpretation tool for clinical use. BloodGPT and Kantesti claim 98%+ clinical accuracy. The technology actually works now. |
| Regulatory tailwind | The FDA's Jan 2026 CDS guidance clarified that clinical decision support software matching patient data to treatment guidelines can be exempt from device regulation. That's a big deal for a product like this. |
Buyer: Solo and small-group functional medicine practitioners. There are roughly 20-35K of them in the US. Secondary buyers: longevity clinic chains, DPC practices adding advanced labs, telehealth platforms.
Pricing: Per-report ($15-30 per interpretation) or monthly SaaS ($199-499/month). At the midpoint, a solo practitioner processing 40 patients per month pays about $350/month, which falls well within the $25-450/month these clinics already spend on software.
Expansion path: Start with bloodwork interpretation as the wedge, then layer on genomics/methylation analysis, then wearable data integration, then build toward a full "longevity clinic operating system" covering protocols, supplements, and patient engagement. Each layer increases switching costs and ARPU.
Initial SAM: 25K practitioners at $4K/year = $100M. Realistic penetration of 10-15% in 3 years = $10-15M ARR. Expanded TAM: Add DPC clinics (2,500+ practices), employer wellness, and telehealth longevity platforms and the addressable market grows to $300-500M. Data flywheel: More interpretations make the AI better, which attracts more clinics, which creates a richer aggregated dataset valuable for pharma research. Exit comparables: Fullscript acquired Rupa Health (Oct 2024). Fullscript serves 100K+ providers and is approaching $1B revenue. Healthcare vertical SaaS trades at 8-15x ARR.
| Risk | Detail | How We'd Address It |
|---|---|---|
| Small initial market | The 20-35K practitioner estimate comes from combining IFM-certified practitioners (~1,200 certified, ~15-20K trained), A4M/ABAARM-certified physicians (~8-10K US), and licensed naturopathic doctors (~6K), with significant overlap. That's a thin starting point for a venture-scale business. | FM is the beachhead, not the ceiling. DPC clinics, growing fast with 58% now employer-sponsored, add 10-20K potential buyers. To establish relationships, we'd start in the Cerbo and Rupa Health communities (tight-knit, word-of-mouth driven) and build presence at IFM and A4M annual conferences. |
| Platform risk | Cerbo already highlights abnormal lab values. Fullscript/Rupa owns lab ordering and could add AI interpretation as a feature. | Move fast. Don't just flag abnormals, provide contextual clinical reasoning with FM-specific optimal ranges. Build before Rupa ships. |
| Liability | Physicians retain ultimate liability for AI-assisted decisions. Malpractice implications are still unclear. | Position as "decision support" rather than "diagnosis." Transparent sourcing, ability to override, clear disclaimers. |
| Practitioner trust | FM practitioners pride themselves on clinical judgment. Some will resist AI on principle. | Augment, don't replace. Show the reasoning. Let practitioners see and edit the AI's logic. |
| Commoditization | Several AI blood test analyzers already exist (BloodGPT, Kantesti). Most are consumer-facing. | The B2B moat comes from EHR integration, workflow embedding, and FM-specific depth (optimal ranges, not just reference ranges). |
| Dimension | Health Score | Medical BNPL | Tempus Comp. | Clinic SaaS |
|---|---|---|---|---|
| Capital to PMF | $2-4M | $8-15M | $30-50M | $500K-2M |
| Time to revenue | 6-9 mo | 9-15 mo | 18-36 mo | 3-6 mo |
| Incumbent threat | Low (no direct analog) | Critical (Cherry, CareCredit) | High (Tempus, Roche) | Low (Rupa/Fullscript is the risk) |
| Regulatory risk | Medium (Anti-Kickback) | High (CFPB, state laws) | High (CLIA, FDA) | Low (CDS exemption) |
| TAM at scale | $15-25B | $40-50B | $237B | $300-500M |
| Exit ceiling | $1-8B (CK comp) | $500M-2B (Cherry comp) | $1-2.4B (Flatiron/Foundation) | $120-500M (Rupa comp) |
| Stage fit | Seed / Series A | Series A+ (needs capital) | Series A+ (needs team + capital) | Pre-seed / Seed |
Supporting market research for Idea 4. Practitioner count, market size, current software stack, competitive landscape, and potential beta candidates.
| Segment | Count | Confidence |
|---|---|---|
| IFM Certified (IFMCP) | ~1,200 | Confirmed (IFM) |
| IFM-trained (not certified) | ~15,000-20,000 | Estimated (IFM directory) |
| A4M/ABAARM certified | ~8,000-10,000 (US) | Confirmed (A4M) |
| Licensed Naturopathic Doctors | ~6,000 | Confirmed (AANP) |
| Total addressable (US) | ~20,000-35,000 | Estimated (significant overlap between categories) |
| Segment | Size (2025) | Projected | CAGR |
|---|---|---|---|
| Longevity clinics | $5.35B | $9.55B (2030) | 12.2% |
| Broader longevity market | $29.8B | $46.9B (2031) | 8.2% |
| FM lab testing | $8.1B | $15.3B (2033) | 5.9% |
Sources: EIN Presswire, Mordor Intelligence, VMR. Note: FM lab testing estimates vary 2x across sources ($8B vs $12.5B). We used the conservative number.
Functional medicine clinics use a fragmented set of tools with a critical gap in the middle:
| Category | Tools | Gap? |
|---|---|---|
| EHR/EMR | Cerbo (market leader for FM), Practice Better, OptiMantra, Elation, Jane App | No |
| Lab ordering | Rupa Health ($43M raised, acquired by Fullscript Oct 2024) | No |
| Lab interpretation | Manual. Practitioners spend 30-60 min per patient. Some use LabDx (rule-based, niche) or SelfDecode Pro (consumer-first). | YES |
| Supplements | Fullscript (dispensing + Rupa integration) | No |
| Patient engagement | Heads Up Health (data aggregation), various portals | Partial |
No dominant B2B SaaS product provides AI-powered lab interpretation embedded in the clinician workflow. The space is fragmented and early.
| Company | Model | Threat Level |
|---|---|---|
| FunctionalMind (John Snow Labs) | B2B AI for FM clinicians, NLP-powered, evidence-based | Medium (research partnership, not polished SaaS) |
| LabDx | Rule-based FM lab analysis (GI MAP, DUTCH, OATs) | Low (niche, not AI-powered, small operation) |
| SelfDecode Pro | DNA + labs platform, 83M genetic variants | Low (consumer-first, B2B is secondary) |
| Fountain Life / Zori AI | Proprietary AI health assistant | Low (internal use only, not sold as SaaS) |
| Deep Longevity | Biological age from bloodwork, B2B API | Low (narrow scope, aging clocks only) |
| InsideTracker | Consumer biomarker platform | Low (not built for clinician workflow) |
Prioritized by: high lab volume, tech-forward culture, and not building proprietary AI internally.
| # | Clinic | Why | Risk |
|---|---|---|---|
| 1 | Wild Health | Genetics-first, telehealth, tech-native team, data-driven protocols | May have internal AI plans |
| 2 | Parsley Health | Telehealth FM pioneer, VC-backed, high patient volume, standardized protocols | Larger org = slower decisions |
| 3 | Function Health | 160+ lab tests at scale, needs interpretation at volume, Austin HQ | Consumer model, MD review is secondary |
| 4 | Lifeforce | Bloodwork-centric longevity model, data-driven, telehealth | Emerging, smaller scale |
| 5 | Mid-size independent FM practices (Cerbo + Rupa stack) | Fastest to onboard, highest pain (solo practitioners spending hours on interpretation), most responsive to time-saving tools | Fragmented, harder to find |
Excluded: Fountain Life (building Zori AI internally), Cleveland Clinic FM (academic institution, slow procurement), Next Health (retail model, less bloodwork-intensive).
Research compiled April 2026. Sources include IFM, A4M, AANP, Mordor Intelligence, Grand View Research, Verified Market Research, Crunchbase, KFF, Gallup, Tempus SEC filings, and direct product analysis. All market projections are estimates and should be validated against primary sources.