Airtable as scaffolding: build what you plan to take down

Our thesis is falsifiable: for an internal business tool, the rational decision is often to build scaffolding first — a no-code base assembled in a few days, explicitly temporary — to validate the real process, data and usage before any heavy investment. If usage takes off, the validated structure travels. If not, you spent days, not months.

The software industry, meanwhile, sells the definitive: requirement documents, eighteen-month plans, tools "for the next ten years." The numbers tell another story. In 2025, ~31% of IT projects succeed, half drift, nearly one in five is abandoned — and the success rate has been flat since 2017.

Scaffolding is not an admission of weakness. On a construction site, it is certified equipment, designed from day one to be taken down. That is exactly the status of a well-run internal MVP.

The largest empirical study available covers 5,392 IT projects (Flyvbjerg et al., 2022). Its central result: mean cost overrun of 1.8x, median of 1.0x. Most projects barely drift — a minority explodes, up to a documented 280x: a project budgeted at $1,500 ended up costing $425,000.

The danger of a long project is therefore not "+20% of budget": it is the fat tail of the distribution, which standard models underestimate. A business-tool prototype built in days does not remove that risk; it caps it at a few days of work and a monthly subscription.

What doing nothing costs — Launching a long build directly means accepting the documented distribution: ~50% odds of running over deadline, budget or scope, ~19% odds of abandonment (PM World Journal, 2026), and exposure to the extreme tail — a mean overrun of 1.8x, up to 280x observed (JMIS, 2022). Scaffolding caps that exposure: a few days of construction and $20/user/month, cancellable.

A rare property of good scaffolding: you know in advance where it stops. Airtable's limits are public and dated — that is a quality, not a flaw.

Read-only collaborators, form submissions and shared links are not billed on Team and Business: the whole team can join the usage validation without multiplying seats. And temporary no-code is not a fringe bet: Gartner predicted as early as November 2021 that 70% of new enterprise applications would use low-code or no-code by 2025.

These ceilings trigger the take-down question: approaching 50,000 records is not a problem, it is a signal planned from day one.

Boehm's experiment (IEEE, 1984) — seven teams, same product — measured ~45% less effort with prototyping, at equivalent performance. The same study documents the flip side: less coherent design, harder interface integration. The prototype validates usage; the production structure gets rebuilt cleanly. This is Eric Ries's build-measure-learn loop: progress is learning validated by usage.

We traveled that road with edorma. The starting point: a weekly Excel/HTML dashboard, maintained by hand every Monday. It validated the essentials: which data mattered, who read it, at what rhythm. The end point: a multi-tenant SaaS isolated by Postgres row-level security, which natively restricts the rows each user can see. What traveled was not the file: it was the schema validated by real usage — the principle detailed in model before you tool (Ownward internal data, 2026).

The word "scaffolding" does not decide how the story ends. When usage validates the tool itself, you do not take it down: you reinforce it. Airtable publishes an official open-source SDK for building extensions inside a base; we run 17 of them in production, built in one year — including a full CRM (telephony, email, SMS, interactive maps). Adoption was immediate: the team stayed in its tool.

Another example: our synchronization engine connects six CRM/ERP systems to a single source of truth, and its mapping configuration lives in Airtable. A new source means a new mapping, not new code. The switch is never ideological — we formalized five criteria for industrializing an automation: observed usage decides, in both directions. That is the core of how we build.

Six questions, 60 to 90 days after go-live, to replace attachment with observable signals.

Scaffolding does not fit everything. Strong regulatory requirements from day one, highly sensitive data, or volumes above the published ceilings from the start justify beginning with the structure. Boehm documented it back in 1984: prototyping alone yields less coherent design — without schema discipline, the scaffolding becomes a maze nobody dares touch. Finally, our proprietary facts (edorma, 17 extensions) are real cases, not statistics: they illustrate the road without proving it works everywhere.