Custom software costs less to build than to own. A first release typically runs $20,000–$100,000 depending on how many surfaces it has, and then 15–25% of that figure every year to run, support and change it. Budgeting the build alone is why software projects get approved and then quietly regretted in year two.
This guide covers the whole picture: what a build costs by product shape, complexity, phase, feature and region; what AI features add; how long it takes; what recurs annually; and why the same requirement gets quotes that differ by a factor of three.
Key takeaways
- Build is roughly half of a three-year total: Run and change make up the rest and are almost always omitted from approval documents.
- Surfaces drive price more than features: A second user type or a second system to stay in sync with moves a budget further than ten more screens.
- Region explains most price gaps: Comparable work bills at roughly $15–$160 per hour depending on where the team sits.
- 60% of software projects exceed their initial budget: Almost always because the first number covered build only.
What a build costs by product shape
The shape of the product, how many separate interfaces it needs, moves the number far more than the length of the feature list.
| Product shape | Typical range | What that includes |
|---|---|---|
| One custom integration | $20,000–$32,000 | Two systems connected, field mapping, error handling, reconciliation reporting |
| Mobile app plus backend | $25,000–$45,000 | One cross-platform app, API, database, admin panel, store submission |
| Web application plus backend | $30,000–$55,000 | Responsive web product, API, database, admin panel, one or two integrations |
| Multi-surface platform | $60,000–$100,000 | Admin web, user mobile, user web and a shared backend delivered as one product |
Two things drive movement inside a band: the number of user roles, and the number of systems you must stay synchronised with. Adding a third role or a second ERP integration costs more than adding ten screens.
What a build costs by complexity
A second way to place a project, useful when the surfaces are not yet decided.
| Tier | Range | Characteristics |
|---|---|---|
| Simple | $20,000–$35,000 | Standard screens, one user type, no real-time behaviour, at most one integration |
| Moderate | $35,000–$60,000 | Several roles, payments or scheduling, two or three integrations, an admin panel |
| Complex | $60,000–$100,000 | Multiple surfaces, real-time or offline behaviour, heavy integration, strict permissions |
| Regulated or enterprise | $100,000+ | Health, finance or public sector obligations, audit requirements, formal change control |
Published figures elsewhere sit higher: commonly $40,000–$100,000 for a simple app, up to $200,000 for a medium one, and $400,000 or more for complex products, with sector averages cited between $60,000 and $250,000. Compare what is delivered before comparing totals.
What each phase costs
Costs are not spread evenly, and the shape tells you quickly whether a proposal is balanced.
| Phase | Share of build | Typical spend on a $50,000 project |
|---|---|---|
| Discovery and definition | 10–15% | $5,000–$7,500 |
| Design and prototyping | 10–15% | $5,000–$7,500 |
| Engineering | 50–55% | $25,000–$27,500 |
| Admin and back-office interface | 15% | $7,500 |
| QA, testing and launch | 10% | $5,000 |
If discovery is under 10%, scope risk is being carried into the build, where it costs several times more to resolve. If the admin interface is not a named line, it is missing rather than free.
What region and day rate contribute
Day rates are the other half of the arithmetic. Agencies in this segment commonly bill $300–$450 per person-day, and rates for comparable work range from roughly $15 to $160 per hour worldwide.
| Region | Typical hourly rate | Notes |
|---|---|---|
| North America | $100–$160 | Highest rates; closest time zone for US buyers |
| Western Europe | $80–$130 | Strong regulatory familiarity for EU products |
| Eastern Europe and Türkiye | $35–$60 | Common mid-market choice; EU-adjacent time zones |
| Latin America | $30–$60 | Nearshore option for US buyers |
| South and Southeast Asia | $15–$40 | Lowest rates; widest quality variance |
A rate difference is not a scope difference. The same product can honestly cost three times more in one region than another, which is why a cheap quote and an expensive quote often describe identical work.
What individual features add
Some features carry disproportionate cost because they introduce infrastructure rather than screens.
| Feature | Typical addition | Why |
|---|---|---|
| User accounts and roles | $2,000–$6,000 | Permissions, recovery, session handling |
| Payments | $3,000–$9,000 | Provider integration, failure states, reconciliation |
| Real-time or chat | $5,000–$15,000 | Sockets, delivery guarantees, presence |
| Offline capability | 20–40% of build | Sync logic, conflict resolution, testing |
| Maps and geolocation | $3,000–$10,000 | Provider costs, permissions, battery behaviour |
| Third-party integration | $2,500–$8,000 each | Data model, failure behaviour, reconciliation |
| Video calling | $6,000–$18,000 | Managed media provider, fallbacks, recording rules |
Offline is the one buyers most often underestimate, because the difficulty is not storage but deciding who wins when two people edit the same record while disconnected.
What AI features cost
AI is priced differently from everything else on this page: there is a build cost and then a usage cost that scales with volume.
| Approach | Setup | Ongoing |
|---|---|---|
| Hosted model API | $5,000–$20,000 | Per-token usage; grows directly with traffic |
| Retrieval over your own content | $10,000–$30,000 | Usage plus storage and re-indexing |
| Self-hosted or fine-tuned model | $30,000+ | Dedicated capacity billed whether used or not |
Most business use cases belong on a hosted API. Self-hosting is justified by data residency requirements or very high sustained volume, and dedicated capacity is the trap: it bills continuously while your traffic is intermittent.
Budget for evaluation and monitoring too. A per-conversation cost that looks trivial in testing becomes a real line item at production scale.
How platform choice changes the number
| Approach | Relative cost | Best for |
|---|---|---|
| Cross-platform (Flutter, React Native) | Baseline | Forms, lists, content, most business apps |
| Two native codebases | +30–50% | Heavy graphics, deep device integration, platform-specific design |
| Progressive web app | −20–40% | Occasional use, discovery through search, no store requirement |
Cross-platform is the default for business software. The real limits are advanced camera and sensor work, reliable background processing, and same-day support for new OS features.
How long it takes
| Project shape | Typical timeline |
|---|---|
| Single integration | 4–8 weeks |
| Mobile app plus backend | 3–5 months |
| Web application plus backend | 3–6 months |
| Multi-surface platform | 6–10 months |
Compressing a timeline costs money rather than saving it. Adding people to a late project raises coordination overhead, and rush work commonly carries a 15–25% premium while producing more defects.
The costs that do not appear in the quote
A build price is a one-off. The following recur, and together they usually exceed the build within three years.
| Category | Typical annual cost | Notes |
|---|---|---|
| Hosting and infrastructure | $1,200–$9,000 | Scales with traffic and data volume |
| Third-party services | $600–$12,000 | Payments, messaging, maps, identity, monitoring, email |
| Maintenance and security | 10–15% of build | OS and framework updates, dependency patching, store requirements |
| Support | $0–$18,000 | Internal or contracted; depends on user count and hours covered |
| Change and new features | Whatever you fund | Never zero for a product in real use |
Beyond these sit the costs nobody quotes: app store developer accounts, SSL and domains, security review, accessibility remediation, legal and licensing, and user acquisition. Individually small, collectively a meaningful share of year one.
For a $60,000 build, expect roughly $9,000–$15,000 a year to keep it healthy before adding a single feature. Over three years that is an $87,000–$105,000 commitment, not a $60,000 one.
Build a three-year figure, not a project price
Take the build quote, add the annual run cost, multiply by three, then add a change budget of 10–20% of build per year for a product you expect to develop.
For a mid-sized platform: $60,000 build, $12,000 a year to run, $9,000 a year to change, around $123,000 over three years. That is the number to compare against an off-the-shelf subscription, and the one that survives review, because it is what you will actually spend.
Why the same product gets three very different quotes
Across regions, rate explains most of the difference. Between three proposals on the same shortlist at similar rates, it almost never does, there the gap is what each supplier assumed.
The admin interface is the most common omission. Every customer-facing feature creates staff work behind it, and that interface is typically a quarter to a third of the real build. A quote that does not name it has not removed the work; it has deferred it to a change request at a worse commercial moment.
The others: integrations priced as one line rather than individually, data migration assumed to be your job, testing and store submission excluded, and the first month of post-launch fixes treated as billable. Normalise all of these before comparing any numbers.
Day rates also translate into person-days, which is the more revealing question. A $30,000 project is around 70–100 of them; asking a supplier for their assumed day count tells you more than asking their rate.
When buying beats building
Custom software is justified when the process is what you compete on, when no product supports your rules, or when licence and workaround costs already exceed a build.
It is not justified when your process resembles the market standard and a product fits with modest configuration. Compare three years on both sides: licence, implementation, integration, per-user growth, internal administration, and the annual cost of the manual work a product cannot absorb. That last line decides many of these comparisons and is almost always left out.
Reducing cost without damaging the product
Cut scope, not quality. A narrow product that works produces clean evidence and real usage; a broad one that half works produces support load and data you cannot learn from.
The reliable savings: one cross-platform codebase instead of two native ones, buying rather than building anything commodity, billing, authentication, notifications, search, and deferring every feature that does not serve the first release's stated goal. The unreliable saving is skipping testing, which returns as production defects at several times the cost.
Standardising an unstable process before encoding it saves more than any technical choice. Software encodes rules; if the rules change monthly, you will pay to rebuild them.
Ask for the total before you sign
Any supplier should be able to give you a three-year figure covering build, hosting, third-party services, maintenance, support and a realistic change budget. Ask which part of their estimate they are least confident about, a team that names one is estimating, and a team that names none is selling.
Related guides
- Alongside Total Cost of Software: Build, Run, and Change, continue with How to Build a Business Case for Custom Software.
- Alongside Total Cost of Software: Build, Run, and Change, continue with How to Choose a Digital Transformation Partner.
If the work prompted by Total Cost of Software: Build, Run, and Change leads to a funded initiative that needs product strategy, design, engineering, or integration support, Discuss Your Software Project.
Ali Boran Gazel