Build vs Buy: When Custom Beats SaaS (and Vice Versa)

I get pulled into this fight roughly twice a month. A founder is convinced they need to build their own auth, or their own analytics, or their own billing system, because 'we will save money long-term' or 'no SaaS does what we need.' Almost always, they are wrong — and the wrong call burns 3 to 6 months of engineering capacity that should have gone into the actual product.

I have also seen the inverse: founders gluing together 14 SaaS tools to do something their core product needed to own, and watching the integration tax slowly kill them. Both failure modes share a root cause: no clear rule for what deserves a build.

Here is the rule. It has saved my clients millions of dollars over the last few years and it fits on a Post-it.

The one rule: build your differentiator, buy everything else

Build only the things your customers specifically pay you for. Buy the supporting layer. Most startups have a single core differentiator — an AI workflow, a marketplace algorithm, a domain-specific UX, a proprietary dataset. That is the 5 percent that deserves your engineering hours. The other 95 percent (auth, email, payments, CRM, analytics, support, error tracking, file storage) has been solved better and cheaper by SaaS than you can solve in-house.

What to buy by default in 2026

These categories are commodities in 2026. The cost to build any of them in-house is 10 to 50x the cost to buy, and the SaaS roadmaps move faster than your team will. Default to buying.

• Auth — Clerk ($25/mo + $0.02/MAU), Supabase Auth (free tier generous), Auth0, WorkOS for SSO. Building auth in 2026 is a six-figure security mistake waiting to happen.
• Email transactional + marketing — Resend ($20/mo), Loops ($49/mo), Customer.io ($100/mo), Postmark. Deliverability alone takes 6+ months to do badly.
• Payments and billing — Stripe (2.9% + 30¢), Paddle (~5%, merchant of record), LemonSqueezy. See stripe vs paddle vs lemonsqueezy for the deep comparison.
• CRM — HubSpot (free tier real), Pipedrive ($24/seat), Attio ($29/seat). The schema flexibility of modern CRMs has made building your own pointless for 99 percent of B2B startups.
• Error tracking & monitoring — Sentry ($26/mo), Datadog (expensive but powerful), Better Stack. Building your own observability is a 6-engineer-quarter project nobody asked you to do.
• Customer support — Intercom ($39/seat), Plain ($59/seat for the modern alternative), Crisp ($25/mo for budget). Plus AI deflection (Fin, Chatbase) bolted on top.
• Analytics — Posthog (free up to 1M events), Mixpanel, Amplitude. Self-hosted Posthog at scale costs less than the engineering required to maintain a homegrown event pipeline.
• File storage — S3, Cloudflare R2 (zero egress), Backblaze. Never roll your own object storage. Ever.
• Search — Algolia, Typesense, Meilisearch (self-hostable). Postgres full-text search works until it does not.
• AI inference — OpenAI, Anthropic, Replicate, Together AI. Use the API. Do not host Llama yourself unless that is the company.

What to build, even when SaaS exists

These are the categories where I have repeatedly seen custom builds out-earn the SaaS alternative — but only when the criteria above are met.

1. Your core product workflow

If your differentiator is 'we organize X better than anyone,' the data model and UX of X belong in your codebase. Trying to build a Notion-on-top-of-Airtable for your core product creates a brittle Frankenstein that breaks the moment you need to ship a feature your no-code platform does not support.

2. AI orchestration if AI is the product

If your value prop is 'AI agent that does X,' the orchestration layer (prompt management, tool calling, state, evals, retries, fallbacks) is your moat. Use OpenAI or Anthropic for inference. Build the layer above. LangChain and LlamaIndex are starting points, not ends.

3. Pricing and billing logic when it is non-standard

If your billing is per-seat with proration and an annual toggle, Stripe Billing handles it. If your billing is 'usage-based with three metered components, a credit system, custom invoicing for enterprise, and a partner revenue share,' you will end up writing significant logic on top of Stripe regardless. Build the abstraction layer.

4. Internal admin tooling for high-volume ops

Retool ($10/seat) and Forest Admin handle most internal tooling. But if your ops team is in your admin panel for 6+ hours a day, building a custom admin pays back inside a quarter. The friction reduction compounds.

The three triggers to switch from buy to build

I tell founders to keep buying until two of these three triggers fire. Hit two, and start scoping a migration. Hit three, and you should have started 6 months ago.

1. Trigger 1 — The bill exceeds $5,000/month and is growing. SaaS pricing is linear; engineering costs are step-functions. At $5K/month and rising, a one-time $30K to $80K build amortizes inside 12 to 18 months.
2. Trigger 2 — The capability has become part of your core product, not a supporting tool. If your customers now pay you partly because of how you do X, X belongs in your code, not in someone else's API.
3. Trigger 3 — The vendor has a roadmap conflict. They are launching a feature that competes with you, deprecating something you depend on, or signaling acquisition by a hostile party. Lock-in becomes a real risk.

Build vs buy comparison: 12 capabilities scored

Real costs across twelve common startup capabilities. 'Build cost' is upfront engineering effort to ship a v1. 'Maintenance' is ongoing engineering cost annualized. 'Recommended' is what I tell most pre-Series A founders.

The hidden costs of building you forgot to budget for

Founders who quote build costs almost always quote the v1. The actual cost is 3 to 5x higher across the lifetime of the feature. Add these line items before you commit.

• Security and compliance — every custom auth or payment flow needs penetration testing, SOC 2 evidence, and ongoing patching. $20K to $80K/year you did not budget.
• Edge cases — international addresses, weird emails, time zones, currency rounding, retry logic. Each one is a 2 to 5 day bug you will not see until production.
• On-call burden — every custom system someone must wake up for. Engineers leave teams that have too many homegrown systems on-call.
• Vendor parity drift — when Stripe ships a feature, you do not. Customers ask for features you do not have, and you spend the rest of your life catching up.
• Knowledge concentration — the engineer who built it leaves and takes the knowledge. Now you have a critical system nobody understands.

The hidden costs of buying you forgot to budget for

Buying is not free either. The honest version of the trade-off includes these.

• Integration tax — every SaaS needs to talk to every other SaaS. By tool 8, you have a real Make.com / Zapier bill and a fragile pipeline.
• Per-seat creep — what costs $25/seat at 10 seats costs $25 × 80 = $2,000/month at 80 seats. Plan for it.
• Roadmap dependency — you ship when they ship. If a vendor stalls, your customers blame you.
• Data egress and lock-in — exporting your data is rarely as clean as the marketing implies. CRMs especially.
• Compliance scope — every vendor that touches PII expands your SOC 2 scope. Vendor management becomes a job.

A 5-step decision rubric for any specific call

Run any specific build vs buy question through these five steps. It takes 15 minutes and prevents most bad calls.

1. Step 1 — Is this a thing customers pay you specifically for? If yes, build. If no, continue.
2. Step 2 — Is there a credible SaaS that does 70 percent of what you need at under $1,500/month? If yes, buy and customize the gaps. If no, continue.
3. Step 3 — Is the SaaS bill projected to exceed $5K/month within 12 months at your growth rate? If no, buy. If yes, continue.
4. Step 4 — Is the capability moving toward becoming core to your product? If no, keep buying despite the cost. If yes, continue.
5. Step 5 — Run a 30-day spike: build the simplest version, measure the actual engineering cost. If the spike runs over budget, buy. If it lands clean, build the v1.

Where I help

An Architecture Audit (from $1,499) maps every build vs buy call in your stack and outputs a 90-day plan with named tools, migration paths, and ROI estimates. For ongoing strategic input, I run a fractional CTO practice from $2,999/month. If you have a specific build that has cleared the 5-step rubric, the MVP Build Sprint starts at $3,500.

If you are still deciding what tasks to even own, see automate vs hire 2026 first. If your build call is specifically about hosting, the deep dive on AWS vs Vercel vs Railway will help.