1. Why This Decision Usually Goes Wrong
The build-versus-buy decision is rarely lost on the technical merits. It is lost on who is in the room.
If you ask a development agency whether you should build custom software, the incentive gradient runs one direction. If you ask a SaaS vendor’s solutions engineer whether their platform can handle your workflow, the answer is always a qualified yes — qualified in ways that surface in month four, after procurement has signed.
Neither party is lying. They are answering a different question than the one you asked. The agency is answering "could this be built?" The vendor is answering "could this be configured?" Both answers are almost always yes. Neither tells you what you actually need to know, which is whether the total cost of the path — over three years, including the parts nobody quotes — is worth the difference in fit.
The useful question is not "can we build this?" It is "what does the gap between an off-the-shelf fit and a perfect fit actually cost us per year, and is that number larger than the cost of owning software?"
That reframing does most of the work. It converts an architectural argument into an arithmetic one.
2. The Four Tests
We run four tests in every diagnostic. Three yeses means building is probably correct. One or zero means buy and stop talking about it.
Test 1: Is this process actually differentiated?
Not "is it important" — differentiated. Payroll is important. It is not differentiated. Nobody has ever won a market because their payroll ran better than a competitor’s.
The honest version of this test: if you described your process to three competitors, would they recognise it as unusual? If your workflow is a standard pipeline with renamed stages, you are describing a configuration problem, not a software problem.
Test 2: Does the off-the-shelf gap force human work?
This is the test that carries the most weight, because it is measurable. Count the hours per week your team spends compensating for what the tool cannot do — exporting, reformatting, re-entering, reconciling, chasing.
Multiply by loaded hourly cost. Multiply by 48 working weeks. That annual figure is your build budget’s floor. If four people each lose six hours a week to a tooling gap, you are burning well over a thousand hours a year. That is not a software preference; that is a headcount decision being made by accident.
Test 3: Is the integration you need commercially available?
Sometimes the blocker is not features but connections. If the two systems that must talk have no supported integration, no usable API, or an API priced as an enterprise upsell, the "buy" path quietly becomes a build path anyway — just one where you are also paying licence fees.
Test 4: Will this process still exist in three years?
Custom software is a durable asset with durable maintenance. If the workflow is a response to a temporary contract, a transitional regulatory regime, or a market experiment you have not validated, buy the cheapest thing that works and revisit later. Building for a process you might abandon is the most expensive mistake available in this category.
3. The True Cost of Building
Build quotes are systematically understated, and not usually through dishonesty. They are understated because the quote covers construction and the cost covers ownership.
What is in the quote: discovery, design, implementation, testing, deployment.
What is not, and should be:
- Maintenance. Budget 15–20% of the original build cost annually. Dependencies age, platforms deprecate APIs, security patches are not optional.
- The second wave of requirements. Every system in production generates feature requests within ninety days. This is a sign of success, and it is not free.
- Operational ownership. Monitoring, alerting, incident response, backups. Someone has to be responsible at 3am, and that someone costs money whether or not anything breaks.
- Knowledge risk. If exactly one person understands the system, you have a liability priced as an asset. Documentation and code review are the mitigation, and both take time that rarely appears in a quote.
A three-year total cost of ownership is the only comparison that means anything. Against a per-seat SaaS licence that scales with headcount, custom software often wins that comparison — but it wins it honestly, or not at all.
4. When Buying Clearly Wins
We recommend buying more often than clients expect, and the pattern is consistent:
- Small teams with standard processes. Below roughly fifteen seats on a conventional pipeline, configured off-the-shelf almost always wins on total cost.
- Regulated, commoditised functions. Accounting, payroll, tax. The compliance surface is enormous, it changes constantly, and vendors amortise that cost across thousands of customers. You cannot.
- Anything where the vendor’s roadmap is your roadmap. If the product is already heading where you need it, you get future value for the current price.
- When the real problem is adoption. A surprising share of "we need custom software" turns out to be "nobody was trained on the tool we already own." Rebuilding does not fix a process problem; it re-implements it at greater expense.
If you take one thing from this: the recommendation to buy is a legitimate outcome of a technical engagement, and it should be available to whoever is advising you. If your advisor cannot profitably tell you to buy, treat their build recommendation as marketing.
5. The Hybrid Default
Most engagements do not end in a pure build or a pure buy. They end in a seam.
Keep the commodity systems you already pay for — accounting, email, storage, payments. Build the connective layer and the differentiated workflow on top. This gets you the fit of custom software across the surface where fit matters, at a fraction of the cost of replacing systems that already work.
In practice that means a custom application handling your actual process, integrated to the platforms of record via webhooks and APIs, with a single data model that treats those platforms as sources rather than as the system of truth. It is less satisfying than a clean-slate rebuild, and it is almost always the correct answer.
The diagnostic exists to find that seam. Everything downstream is easier once it is drawn in the right place.
Run the readiness assessment
Eight questions on your data, process, and tooling. Three minutes, no system access, no sales call.