Build custom software when your SaaS spend on a single workflow passes €60,000 per year, no off-the-shelf tool fits your operating model after two evaluations, or the data itself is your product. AI-assisted teams now ship the kind of platform that used to take a quarter in six weeks. That timeline collapse moved the build versus buy crossover earlier in every company's life.
Every founder hears the same advice: don't reinvent the wheel. In 2026, that advice is two years stale.
You got here by buying. €15,000 a year of subscriptions runs payroll, billing, CRM, and ops with two clicks, and that was the right call. This article shows you the math and the operational signals that flip the decision in 2026. You get five buy signals, five build signals, a cost crossover table, and a 30-minute decision tree.
- The crossover moved. Custom build cost dropped 3 to 5 times in 24 months because AI-assisted teams ship in weeks, not quarters.
- €60K is the new €150K. A SaaS workflow over €60,000 per year on per-seat pricing now sits in the build zone, down from €150,000 in 2023.
- Build doesn't mean build everything. Keep auth, payments, and email on commodity SaaS. Build only what makes you different.
- Per-seat punishes scale. SaaS is cheap until your headcount compounds. Custom is expensive once, then flat.
- The hidden buy cost is glue. Zapier sprawl, two-way sync hell, and integration tax dominate total cost after year two.
The crossover moved earlier in 2024 and 2025
In 2023 a fundable MVP took 12 to 20 weeks of an experienced team. By the end of 2025 the same scope shipped in 4 to 8 weeks of an AI-assisted team. I documented the cost tiers in the 2026 MVP cost guide: €5,000 to €25,000 covers the lean tier that replaces most internal SaaS workflows.
That timeline collapse changed the math. A custom platform that used to cost €60,000 and take a quarter now costs €15,000 to €30,000 and ships in six weeks. SaaS pricing went the other way: per-seat plans rose 8 to 15 percent year over year through 2024 and 2025, and the cheapest tiers got feature-gated harder.
In a recent project, I shipped a financing-certificate platform for a real-estate workflow in a short delivery window. It included a multi-step form, automated PDF generation, an admin dashboard, partner-bank comparison, and certificate turnaround inside the promised service window. Stitching the same flow from form builders, document tools, CRMs, and a credit-bureau integration would have created an expensive, fragile stack. Owning the workflow makes the promise credible.
If your SaaS line items on a single workflow already top €40,000 per year, the full-stack applications service scopes a custom system around that workflow and can ship focused first releases in 4 to 10 weeks.
Five signals you should buy, not build
- The workflow is generic. Authentication, payments, scheduling, transactional email, and basic CRM are commodities. Building these in 2026 is a choice to spend money on undifferentiated infrastructure.
- Under 50 internal users with no clear scale path. Per-seat SaaS is cheapest in the bottom range. Build only after you can model the seat count three years out.
- No product or engineering DNA in-house. Custom software needs someone to brief, accept, and steward it. If you cannot describe the workflow in writing, hire a consultant before hiring a builder.
- The vendor charges per feature, not per seat. Per-feature pricing punishes the vendor when you grow, not you. Ride that curve until it stops.
- Customers expect the standard tool. Stripe Checkout, HubSpot CRM, and Slack carry trust signals. Replacing them with custom UI introduces friction your buyers do not want.
Buy is the default, not the failure case. Most agencies sell builds, so I say this against interest: every founder under 50 internal users and €40,000 of total subscription spend should keep buying. Custom is an asset, not a status symbol.
Five signals you should build, not buy
- The data is the product. Your unique aggregation, scoring, matching, or workflow logic is the moat. SaaS will never expose it the way you need to.
- No vendor fits after two evaluations. Two failed pilots is the signal. The third pilot is sunk cost. Stop, write the spec, and price the build.
- Per-seat pricing on 50+ internal users with no relief. A €40 per-seat plan at 80 users runs €38,400 per year before integrations. Custom auth on a Postgres database costs €0 per seat after the build.
- Integration tax dominates total cost. Zapier sprawl, two-way sync brittleness, and glue code maintained by an ops engineer cost more than the SaaS line items they connect. Collapse the workflow into one app and the connectors disappear.
- Compliance or data residency that no vendor will sign. German invoicing rules, GoBD archival, EU-only data residency, sector audits. If you have asked three vendors and got three no's, build.
The real cost math, with 2026 numbers
Numbers below assume a single workflow replaced. Multiply by the number of workflows you would consolidate.
| Approach | Year 1 cost | Year 3 total | Owns the code |
|---|---|---|---|
| SaaS suite (form + PDF + CRM + auth) | €18,000 | €72,000 | No |
| SaaS plus Zapier glue | €24,000 | €96,000 to €120,000 | No |
| Custom build by agency | €15,000 to €30,000 | €21,000 to €42,000 | Yes |
| Custom build by in-house team | €80,000+ | €140,000+ | Yes |
Custom-by-agency wins year one and widens the gap every year after. SaaS spend compounds with seats and feature gates while custom maintenance stays flat. The agency line uses a scoped custom build through the full-stack applications service plus integration scope. The in-house line assumes one mid-level developer at €70,000 loaded cost, which is the realistic floor in Berlin or Munich.
If your existing site runs on WordPress and you are debating the same crossover, the WordPress versus custom development comparison walks the platform-specific math.
What 'build' actually means in 2026
Build does not mean build everything. The 2026 stance is composable: own the differentiated logic, rent the commodity. In that platform, the owned parts were the financing form, bank-comparison engine, certificate generator, and admin dashboard. The rented parts were Better Auth for sessions, Resend for email, Stripe for any payment, Vercel for hosting, Sentry for errors, and PostHog for analytics.
The codebase is an asset. The subscriptions are rent. Year three of a SaaS-only stack leaves you with a renewed contract and zero negotiating room. Year three of a composable custom stack leaves you with a working app, ten upgrades shipped, and the option to swap any rented piece in a week.
My default stack: Next.js, React, TypeScript, tRPC, PostgreSQL, Drizzle ORM, Better Auth, and Vercel. Type-safe end to end, one deployment pipeline, and the same language from form to database. Less glue, fewer outages, and a junior engineer can read it on day one. I picked this stack to make handover cheap, not to lock you in.
Real example: replacing a stitched SaaS stack
A real-estate workflow that issues personal financing certificates has a clear build-or-buy boundary. Buyers submit structured financing data, receive a binding PDF certificate inside the promised service window, and the system compares partner-bank options behind the scenes. Before custom, the same flow would have looked like Typeform plus DocSpring plus a CRM plus Zapier plus an Excel rate sheet, with an admin user paying for every seat.
The buy total: about €60,000 to €100,000 per year. That covers the form-builder tier needed for logic, the PDF tool's per-document price, CRM seats, Zapier task volume, and the engineer-hours to keep two-way sync alive. Plus a brittle product the moment any vendor changed their API.
The build total: one focused delivery cycle on the full-stack applications service, a single Next.js codebase that is type-safe from form to database, strong Lighthouse performance, fast page loads, and the option to ship small changes without waiting on a vendor roadmap. The practical breakdown belongs in a scoping call, not a named client story.
A 30-minute decision tree
Walk these five questions with your CFO and your operations lead. Twenty minutes if you already know your subscription totals.
- Is the workflow differentiated? If the answer is no, buy. Differentiated means the way you do this thing is part of why customers pick you.
- Is your annual SaaS spend on this workflow over €40,000 and growing? If the answer is no, revisit in six months. Below €40,000 the math rarely flips.
- Have two vendors failed to fit your model? If the answer is no, run one more pilot first. The third failed pilot is the build trigger, not the second.
- Do you have a delivery partner who ships in 6 to 8 weeks? If the answer is no, fix that before you commission a build. Long timelines kill custom projects more often than budgets do.
- Will the build pay back inside 18 months? Compare year-one custom cost to two years of SaaS at current trajectory. If payback runs past 18 months, rent and revisit.
Five yes answers means build. One no means rent and run the audit again next quarter. The wrong build is more expensive than three more years of SaaS, so the discipline is worth the half-hour.
webvise builds custom software for the buy-to-build crossover zone: shipped in 4 to 10 weeks, with code your future engineers can read. The same composable stack runs across every project, so handover is cheap from day one. If you are sitting on €40,000 to €120,000 of annual SaaS spend on a single workflow, describe the workflow and you will get a straight answer on whether to build or keep buying.