Most build vs buy decisions are made on aesthetics, not math. Engineers like to build because building is fun. Executives like to buy because it transfers risk. Neither tendency produces the right answer.
The right answer is a function of six variables: cost, risk, time, control, customization, and vendor dependency. This post lays out the decision matrix, gives you the rule of thumb for each axis, and walks through four real examples.
The Six Variables
Cost
True cost of build = initial build + ongoing maintenance + opportunity cost. True cost of buy = annual fees + integration cost + switching cost down the line.
The mistake is comparing initial build cost (€80K of engineering time) against annual SaaS cost (€20K/year). The honest comparison is total cost over 5 years for each option, including maintenance.
Risk
Build risk: ships late, ships broken, key engineer leaves and knowledge dies. Buy risk: vendor goes out of business, raises prices 5×, gets acquired by a competitor, EOLs the product.
Both have failure modes. Build risk concentrates on execution; buy risk concentrates on dependency.
Time
How long until the system is in production and serving users? Build: 3-12 months typically. Buy: 1 day to 2 months for integration.
Time matters most when there is a market window. A payment system that ships in 6 months has lost the company €500K in revenue that competitors captured. A monitoring system that ships in 6 months has lost almost nothing — there is no urgency.
Control
How much do you control the product roadmap, pricing, terms, data ownership? Build: full control. Buy: vendor controls — they can change pricing, deprecate features, change terms of service.
Control matters when your business depends on specific behavior the vendor might not maintain.
Customization
How specific to your use case does the solution need to be? Build: 100% custom. Buy: limited to what the vendor allows.
Customization matters when "close enough" creates real user friction. For payments, "close enough" usually works (everyone needs basically the same flow). For core product features, "close enough" loses customers.
Vendor dependency
How easy is it to switch vendors later? Open standards (S3, Postgres) = easy. Proprietary platforms (Salesforce, Oracle) = hard.
Dependency matters more than founders think. The vendor you depend on heavily today might raise prices 3× in year 5, and you cannot leave.
The Decision Matrix
Score each option on the six variables, 1-5 each. Higher = better.
| Variable | Build score | Buy score | Winner |
|---|---|---|---|
| Cost (lower is better) | Often 2-3 (high) | Often 3-4 (medium) | Buy usually |
| Risk | Variable | Variable | Depends on vendor |
| Time to production | 2 (slow) | 5 (fast) | Buy wins |
| Control | 5 (full) | 2 (vendor) | Build wins |
| Customization | 5 (full) | 1-3 (limited) | Build wins |
| Vendor independence | 5 (none) | 1-3 (depends) | Build wins |
Total scores give you a directional read. But the weighting matters: for commodity functionality, weight Cost and Time heavily (buy wins). For core product, weight Control and Customization heavily (build wins).
Example 1: Payments — Buy Stripe
The classic "always buy" case.
- Cost: Stripe takes 2.9% + €0.30 per transaction. At €1M ARR, that is €29K/year. Building a payment system to global PCI-DSS standards costs €500K+ and takes 12+ months.
- Risk: Stripe is highly reliable. PCI compliance burden is on them. Their risk is your gain.
- Time: integrate Stripe in 1 week. Build payments in 12 months.
- Control: Stripe controls the feature set. Their roadmap is excellent.
- Customization: Stripe handles 95% of use cases out of box. Edge cases use Stripe Connect or custom flows.
- Vendor dependency: high. Switching from Stripe to Adyen is painful. Acceptable tradeoff for the cost savings.
Verdict: buy Stripe. The only companies that should build payments are payment companies.
Example 2: CRM for Niche Industry — Build
Real estate brokerage with 50 agents. Salesforce costs €150/user/month = €7,500/month = €90K/year. Salesforce does not understand real estate workflows out of the box.
- Cost: Salesforce €90K/year + €50K customization. Build cost: €120K once + €30K/year maintenance.
- Risk: build risk is real — but they can hire 1 engineer to maintain.
- Time: Salesforce live in 3 months. Build live in 6 months.
- Control: full control over their workflows.
- Customization: custom flows for listings, showings, commission splits work natively.
- Vendor dependency: none.
Verdict: build. The customization advantage outweighs the time-to-production penalty when the workflow is industry-specific.
Example 3: Monitoring — Buy Datadog
SaaS company with 30 services. Considers building monitoring vs buying Datadog.
- Cost: Datadog at scale is €30K-€100K/year. Build with Prometheus + Grafana is €20K initial + 0.5 FTE maintenance = €80K/year.
- Risk: monitoring failure during incident = no visibility. Datadog is more reliable than self-hosted.
- Time: Datadog live in 1 week. Self-hosted Prometheus live in 1-2 months.
- Control: Datadog roadmap is excellent.
- Customization: Datadog handles 99% of use cases.
- Vendor dependency: medium. Datadog cost scales with usage.
Verdict: buy Datadog until you are above €5M ARR. Then consider migration to OpenTelemetry + self-hosted to control cost.
Example 4: Core ML Algorithm — Build
AI startup whose core product is a proprietary recommendation algorithm. Considers using OpenAI vs building.
- Cost: OpenAI calls at scale = €50K/month. Building proprietary model = €500K + €100K/year inference cost.
- Risk: dependency on OpenAI's product decisions is existential.
- Time: ship with OpenAI in 1 month. Build proprietary in 9 months.
- Control: proprietary algorithm is the moat. OpenAI dependency is the opposite of a moat.
- Customization: proprietary model fits exact domain.
- Vendor dependency: with OpenAI, total. With proprietary, none.
Verdict: ship MVP with OpenAI, build proprietary as soon as PMF is confirmed. The IP is the business.
Default Recommendations by Category
| Category | Default | When to flip |
|---|---|---|
| Payments | Buy (Stripe, Adyen) | Only if you ARE a payment company |
| Authentication | Buy (Auth0, Clerk, Supabase) | Specific compliance / on-premise needs |
| Email delivery | Buy (Resend, Postmark, SendGrid) | Never — always buy |
| Monitoring / observability | Buy under €5M ARR, hybrid above | When Datadog bill exceeds 1 FTE |
| Analytics | Buy (PostHog, Mixpanel) | Privacy-critical apps |
| Error tracking | Buy (Sentry) | Never — always buy |
| CRM | Buy (HubSpot, Salesforce) | Industry-specific workflows |
| Customer support | Buy (Intercom, Help Scout) | Never — always buy |
| Core product features | Build | If it is truly commodity |
| Differentiated algorithms | Build | If you are early and validating |
The 5-Year Total Cost Math
Always model 5 years, not 1.
Build: initial build cost + (maintenance rate × initial cost × 5).
Buy: annual vendor cost × 5 + integration cost + estimated switching cost.
If you only model 1 year, building always looks expensive vs buying. Over 5 years, the math often flips for core differentiated systems.
Three Mistakes I See Most Often
- Building commodity: "We built our own Stripe replacement to avoid fees." 18 months later, the system is a maintenance burden and the company is paying more in engineering time than they would in Stripe fees. The classic.
- Buying differentiation: "We use Generic SaaS CRM for our differentiated workflow." The vendor adds friction, slows ship velocity, and reduces customization. The CRM becomes the bottleneck.
- Forgetting maintenance: "Build cost is €80K." After 3 years of maintenance at 30%/year, true cost is €152K. Almost no one models this correctly.
The Build Diary Question
Before deciding to build, write down: "In 3 years, will we wish we had bought this?" If yes — buy. The future-you knows things present-you does not.
And the inverse: "In 3 years, will we wish we had built this?" Often this is yes for systems that touch the core differentiation. Buy a placeholder now, plan to build before year 3.
Build vs buy decision review
If you have a build vs buy decision on the table and want a second opinion, I do 60-minute reviews. We score the six variables, model the 5-year cost, and make the call.
Book a discovery call