Build vs. Buy: What to Build Custom in a Healthtech MVP (and What to Buy)
Most build-vs-buy frameworks ignore what regulated software actually changes: once a feature touches clinical safety, PHI custody, or device classification, "buy" stops being a cost question. This is the decision rule engineering teams use to scope a healthtech MVP — which integrations to buy, which to build, and why the cheapest path sometimes becomes the riskiest one.
Why Generic Build-vs-Buy Advice Breaks Down for Healthtech
Most build-vs-buy frameworks treat the decision as a spreadsheet problem: compare the subscription cost of a vendor SDK against the fully-loaded cost of an engineering team, pick whichever number is smaller, ship. That framing works fine for a CRM integration or an internal admin tool. It falls apart the moment the component you're evaluating touches a patient.
In a regulated clinical product, the real question isn't what's cheaper this quarter. It's who owns the risk if this component fails, and whether you can actually prove that to an auditor, a payer, or a regulator. A vendor SDK that's perfectly fine for a consumer fitness app can be the wrong call for the same feature in a clinical workflow, not because it's worse software, but because its failure mode, its data handling, and its contractual guarantees weren't built for a HIPAA-covered entity. Teams that only scope on speed-to-MVP routinely rediscover this the hard way, mid-build, when a vendor can't produce a Business Associate Agreement or can't explain where PHI actually lives.
An outside MVP development partner is often brought in to solve exactly this before the first sprint starts. The goal isn't writing code faster. It's separating the components where buying is genuinely fine from the ones where it quietly transfers risk the startup can't actually offload.
The Real Decision Isn't Cost — It's Who Owns Clinical Risk
Every component in a healthtech MVP sits somewhere on a spectrum from "pure infrastructure" to "clinical judgment." The further toward clinical judgment, the more building in-house becomes non-negotiable, regardless of what the TCO spreadsheet says.
Clinical-Safety Triggers That Force "Build"
If a component's output directly influences a clinical decision, such as a dosing calculation, a triage score, or an alert threshold that tells a nurse to escalate, you need full control over its logic, its versioning, and its audit trail. A third-party vendor can't give you the traceability a reviewer or a plaintiff's attorney will eventually ask for, because the logic lives in their codebase, not yours. This is the line where "move fast" and "medical device software" stop being compatible goals.
Data-Custody and PHI Triggers That Force "Build"
Any component that stores, transforms, or routes protected health information needs a vendor who will sign a Business Associate Agreement and can tell you precisely where that data lives, how long it's retained, and who can access it. If a vendor can't answer those three questions in writing, the feature either doesn't ship with that vendor, or you build the data-handling layer yourself and let the vendor touch only de-identified inputs.
Where "Buy" Is Almost Always Correct (Non-Clinical Infra)
Authentication rails, payment processing, transactional email, error monitoring, and video transport make no clinical judgment, and none of them need to touch raw PHI if you design the integration boundary correctly. These are exactly the components where building in-house burns months of engineering time for zero clinical-safety benefit. Buy here, and spend the saved time on the components that actually need it.
A Practical Checklist — SDK or Vendor Evaluation Before You Buy
Before any vendor contract gets signed for a component that will touch clinical data or clinical workflow, run it through three question sets.
Interoperability and FHIR/HL7 Compliance Questions to Ask a Vendor
Does the vendor expose a FHIR-conformant API, or will your team need to build and maintain a translation layer against a proprietary format? A vendor that only speaks its own schema isn't disqualifying, but it means you're quietly building an integration layer anyway, just one you didn't budget engineering time for.
Data Residency and BAA (Business Associate Agreement) Questions
Will the vendor sign a BAA under HHS HIPAA requirements as a standard part of the contract, or only on an enterprise tier you haven't budgeted for? Where is data physically stored and processed, and does that match your organization's data-residency commitments to patients and partners? Get this in writing before integration work starts, not after.
Exit Cost — What Happens If You Have to Replace This Vendor in 18 Months
Can you export your data in a standard format, or does the vendor's proprietary schema make migration a multi-month project? A vendor that's cheap today but structurally impossible to leave later isn't a good buy decision. It's a deferred build decision with interest.
Component-by-Component: What Epica Teams Usually Buy vs. Build
Video/Telehealth (Usually Buy — e.g. Twilio, Vonage Video SDKs)
Real-time video infrastructure is a genuinely hard distributed-systems problem that commercial SDKs have already solved at a scale no MVP needs to re-solve. Buy the video transport layer; build the clinical workflow around it: session scheduling, consent capture, and the handoff into the patient's record stay yours.
Identity, Auth, and Patient Matching (Context-Dependent)
Generic authentication (login, MFA, session management) is a safe buy. Patient identity matching, resolving whether this is the same patient as the one in your EHR, is not: a false match or false non-match is a patient-safety incident, not a login bug. Buy the auth rail, build or tightly own the matching logic.
EHR/FHIR Connectivity (Usually Buy via Interoperability Platform, Build the Mapping Layer)
Connecting to dozens of EHR vendors from scratch is rarely a good use of MVP engineering time; interoperability platforms exist precisely to absorb that long tail. What you should still build and own is the mapping layer that decides how incoming data populates your clinical model. That logic encodes your product's clinical assumptions, and a vendor shouldn't own it.
Clinical Decision Logic, Scoring, Triage Rules (Almost Always Build)
This is the component category with zero exceptions in most Epica engagements. Any rule that tells a clinician or patient what to do next needs to live in code your team wrote, tested, versioned, and can defend, because it's the first artifact anyone asks for after an adverse event.
Payments/Billing (Buy, With a Compliance Wrapper)
Payment processing is a solved, heavily-regulated problem best left to PCI-compliant vendors. The only in-house work is a thin wrapper that keeps billing data separate from clinical data in your schema, so a billing vendor breach can never become a PHI breach.
The TCO Trap — Why "Buy" MVPs Get Expensive at Scale
Per-seat or per-encounter vendor pricing looks cheap at MVP volume and gets expensive fast once a clinic actually adopts the product. The crossover point, where cumulative vendor fees exceed what in-house engineering would have cost, arrives earlier than most teams expect, often within the first 12–18 months of real usage.
| Path | Cost curve | Where it breaks even |
|---|---|---|
| Buy (SDK/vendor) | Low upfront, scales linearly (or worse) with usage/seats | Cheaper until volume crosses the vendor's per-unit pricing threshold |
| Build (in-house) | Higher upfront engineering cost, flat marginal cost per additional user | Cheaper past the crossover point, and the only option where clinical-safety/compliance triggers apply regardless of cost |
Buying at MVP stage usually isn't the trap. The trap is failing to revisit the decision once usage crosses the line where the vendor invoice outgrows what the equivalent in-house component would have cost to build and maintain.
A 5-Step Scoping Exercise for Your Next Healthtech MVP
- List every component the MVP needs: video, auth, EHR connectivity, scoring logic, payments, notifications. Do this before assigning any of them to "buy" or "build."
- Flag every component that touches PHI or clinical judgment. Anything that doesn't is a safe buy candidate by default.
- Run the flagged components through the vendor checklist above (interoperability, BAA/data residency, exit cost) before assuming a vendor is viable.
- Model the TCO crossover for the components you're still unsure about, using your realistic 12–18 month adoption curve, not launch-day volume.
- Document the decision and the trigger that drove it: clinical safety, data custody, compliance, or cost. That way the choice gets revisited deliberately, not accidentally, when the product scales.
FAQ
Is it ever safe to buy clinical decision support logic?
Rarely, and only when the vendor's output is advisory-only, clearly labeled as such, and your team retains the final clinical judgment layer on top of it. If the vendor's logic alone drives patient-facing decisions, you need full visibility into and control over that logic, which usually means building it.
What's the fastest way to tell if a vendor is HIPAA-ready?
Ask for a signed Business Associate Agreement as a standard contract term, not an enterprise add-on, and ask them to describe in writing where PHI is stored and who can access it. A vendor that hesitates on either question isn't ready, no matter what their marketing page claims.
Should an MVP ever skip a BAA to move faster?
No. Skipping a BAA to hit a launch date converts a contractual gap into a compliance liability that doesn't disappear when the deadline passes. It just surfaces later, usually during an audit or a breach investigation, at a much higher cost than the time saved.
How do teams avoid vendor lock-in on the "buy" side?
Prefer vendors that export data in standard formats (FHIR resources rather than proprietary JSON), negotiate data-portability terms before signing, and keep your own mapping/integration layer thin enough that swapping the underlying vendor doesn't require rewriting your clinical data model.
When should a healthtech startup revisit a build-vs-buy call after launch?
Revisit it when usage crosses the TCO crossover point you modeled at scoping time, when a vendor's compliance posture changes (a BAA lapses, a data-residency commitment shifts), or when a feature that started as advisory starts directly driving clinical decisions. Any of those shifts the risk calculus that produced the original decision.
Need this done right? Epica builds medical software end to end. Talk to us →