Ask for the operating model with the quote
The better question is not just how much it costs to build. Ask what it will cost to operate, change, recover and eventually replace. If those answers are absent from the proposal, the low initial number may reflect missing scope rather than efficient delivery.
Separate four cost categories
Implementation includes discovery, design, development, migration, testing and training. Operation includes hosting, vendor usage, monitoring, support and the internal person who administers the system. Change includes new requirements, dependency updates and adjustments when connected services evolve. Exit includes exporting records, transferring ownership and moving to an alternative.
Use your own expected usage and support requirements. Avoid multiplying a generic maintenance percentage by the build price and calling the estimate complete. Ask which costs are fixed, usage-based, optional or unknown. Require the assumptions to remain visible so you can revisit them as the business changes.
The illustration below shows how a $12,000 implementation can become $33,000 of first-year ownership once operating costs, support, internal staff time and planned changes are included. These are hypothetical planning inputs, not STKD fees or market benchmarks. Replace them with your scoped quote and price any relevant migration, incidents or transition work separately.
Do not compare a prototype with a production system
A public LinkedIn discussion by Ali Sadat distinguishes a prototype, a departmental application and an enterprise replacement. That distinction is useful when evaluating apparently contradictory build-cost claims. We do not adopt the post's productivity assumptions or cost outputs as benchmarks. Read the discussion.
A prototype can demonstrate a workflow using test data and limited access. A production system has to handle real permissions, messy records, failed requests and a person who needs help when it stops working. A critical replacement also needs a transition plan that protects ongoing business.
Ask every vendor to mark which stage its quote covers. Is migration included? Are test records merely representative, or has the team examined the real data shape? Does support include investigation and repair, or only a response acknowledging the issue? Ambiguity here produces expensive disagreement later.
Understand what AI does and does not remove
AI can assist with implementation, documentation and tests, but the commercial promise should still be a verified result. GitHub explicitly warns that AI-generated suggestions may contain errors or security issues and recommends careful review and testing. GitHub responsible-use documentation.
Request evidence of acceptance tests and review, not a claim that the application was created unusually quickly. Ask how access boundaries were tested, how secrets are managed and how the system behaves when a connected service fails. A polished interface is not evidence that the underlying workflow is safe.
Human time also belongs in the estimate. Your staff may need to explain exceptions, clean records, approve decisions and learn the new process. If a vendor assumes those tasks are free and instantaneous, its delivery plan is not describing the full project.
Price ownership and recovery explicitly
Leon Merkel's public account of moving from a self-built CRM back to purchased software centers on maintenance competing with other work. The relevant question for your business is who gets interrupted when this system needs attention, and what they will stop doing. Read the discussion.
Ask for a named maintenance owner, a support channel, expected response arrangements and the distinction between defect repair and new scope. Define who pays for third-party services and who controls their accounts. Make sure another qualified team could understand and operate the system from its documentation.
Recovery needs more than the sentence we have backups. Specify what is backed up, how restoration is tested, what recent work could be lost and what manual process is available during an outage. The appropriate recovery target depends on the business consequence; it should not be invented by a developer in isolation.
The quote-comparison checklist
Use the same list with each supplier. A transparent unknown with a discovery plan is preferable to a confident fixed number that hides unresolved requirements.
- What exact workflow and acceptance tests are included?
- Which migration, training and internal staff tasks are assumed?
- What recurring infrastructure, usage and support charges are expected?
- What is covered by maintenance, and what becomes a change request?
- Who owns code, accounts, documentation and the ability to export data?
- What happens if the supplier is unavailable or we decide to leave?
When custom software is not worth it
If a supported product solves the important problem with reasonable configuration, the freedom of custom software may not justify ownership. If the workflow changes constantly and nobody can agree on its purpose, discovery should come before a build. When the advantage is real, fund the operational discipline around it. The cheapest useful system is the one your team can rely on, understand and change without becoming trapped.
