Software Development Cost: What Drives the Number
How much does software development cost in 2026? Pricing by project type, team model and region, with real cost drivers explained. Get a free estimate.
Custom software cost is the product of two numbers: how many days of work the thing takes, and what a day costs. Everything else is detail. The reason quotes for one brief arrive 5x apart is almost never that one supplier is five times more efficient — it is that they assumed different scopes, or they bill at rates from different continents, or both.
So a cost figure with no stated rate and no stated scope tells you nothing. This page covers what moves each of those two numbers, how to read an estimate, and where budgets actually get consumed.
Why the same brief gets wildly different quotes
Four reasons, in rough order of how much they explain:
Different rates. Senior developer day rates vary by region by a factor of three to five. A team in Western Europe or North America, one in Eastern Europe, and one in South or South-East Asia can each be charging a defensible rate for their market and produce quotes that look incomparable. This is the largest single factor and the one least often stated plainly.
Different assumed scopes. Your brief said "user accounts". One supplier costed email and password. Another costed single sign-on against your identity provider, four roles with per-record visibility, an invitation flow, and an audit log. Both read the same sentence.
Padding for unknowns. A supplier pricing an ambiguous brief has to cover what they cannot see. The more vague the brief, the larger the buffer — and you cannot tell a buffer from a scope by looking at the total.
A deliberate underbid. Some quotes are low because the supplier intends to recover margin through change requests. This is visible in the contract rather than the number: look at how change requests are priced and at what counts as a defect versus a change.
The practical consequence: if your quotes differ by more than about 50%, the problem is usually your brief, not the suppliers. A written specification is the cheapest way to make quotes comparable.
What actually drives the day count
These are the variables that move an estimate most, and the ones worth being precise about before you ask for a number.
User roles and permissions. The single biggest multiplier in business software. One role doing everything is a small build. Five roles, each seeing different records, with an auditor who reads everything and an admin who can impersonate, is a different order of work — and it is a data-model decision, not an interface one, so it is expensive to add later.
Integrations. Every external system is work, and the cost depends almost entirely on the quality of the other side's API. A modern documented REST API is days. A system whose integration path is a nightly file drop, or whose documentation is a PDF from 2017, is weeks. Count them, name them, and find out which have sandboxes.
Data migration. Moving records from an existing system is routinely underestimated because the estimate is made against record counts rather than data quality. The work is in the cleaning rules, the duplicate decisions, and the reconciliation — not the transfer.
Reporting. "Plus reporting" at the end of a brief is frequently a quarter of the build. Reports that span objects, or that need historical state rather than current values, are substantial engineering.
Compliance. HIPAA, SOC 2, GDPR, PCI or a sector regulator changes the architecture rather than adding a feature. Audit logging, access control, encryption, data residency and retention have to be designed in. Retrofitting them under audit pressure costs multiples of building them from the start.
Platform count. Web plus iOS plus Android is not one project. Cross-platform mobile reduces it but does not halve it, because store requirements, permissions and device testing still happen twice.
Rough bands, with the rate stated
Any band is meaningless without the rate assumption, so here is ours: the figures below assume a blended senior offshore day rate, which is what most agencies of our kind charge. At Western European or North American rates, multiply by roughly two and a half to four. At the very low end of the freelance market, divide by two — and read the section on what cheap quotes tend to omit.
| Project | Typical range | What moves it |
|---|---|---|
| Internal tool, one role, one data source | $8k – $25k | Number of screens, whether it replaces a spreadsheet or a system |
| Client portal, documents and status | $20k – $60k | Role count, integration with your system of record |
| Custom CRM or operations system | $40k – $150k | Object model complexity, migration, reporting depth |
| SaaS MVP, one workflow and billing | $35k – $90k | Multi-tenancy, payment, how much is deferred from v1 |
| Mobile app with a backend | $40k – $120k | Native vs cross-platform, offline requirements |
| ERP module (first module) | $50k – $150k | Process complexity, integrations to what it does not replace |
| AI extraction or retrieval system | $25k – $80k | Accuracy requirement, validation, human review workflow |
Treat these as the middle of a distribution, not a quote. A brief at the top of a range and one at the bottom can both be described in the same sentence.
How to read an estimate
A usable estimate has four properties. If one is missing, ask for it.
A range, not a single number. A single figure with no range is either padded or will move. A range with stated assumptions is honest about what is not yet known.
Stated assumptions. Each figure should say what it depends on — "assumes the client's API provides order history", "assumes three roles", "assumes no data migration". These are what let you see why a number would change.
Explicit exclusions. What is deliberately not included. This section prevents more disputes than any other and its absence is a warning sign.
A defect-versus-change definition. What counts as a bug the supplier fixes at their cost, and what counts as a change you pay for. Leaving this vague is how projects end in argument.
Where budgets actually get consumed
Not usually where clients expect. In rough order:
- Integrations that turned out to be harder than the API documentation implied. The most common overrun by a wide margin.
- Permission and visibility rules that were discovered during the build rather than specified before it, forcing data-model changes.
- Edge cases in data migration — the records that do not fit the rules, which are always more numerous than the audit suggested.
- Rework after the first real users see the workflow. This is not waste; it is the point of shipping early, and a budget with no allowance for it is unrealistic.
- Reporting requirements that arrived late, from someone who was not in the scoping conversation.
Core feature development, which is what the brief describes, is usually the part that goes most to plan.
How to spend less without building less
Buy before you build. The cheapest custom software is the custom software you did not need. If a product covers eighty percent of your process, configuring it and building the remaining twenty is almost always cheaper than building all of it. A supplier who never recommends a product is worth a second opinion.
Cut scope, not quality. One workflow finished properly beats four half-built. Half-built workflows teach you nothing and still cost money.
Pay for a specification first. A short discovery phase that produces a written scope, data model and estimate costs a fraction of a build and makes every subsequent quote comparable. It also occasionally concludes that you should not build at all, which is the highest-return outcome available.
Phase it. Deliver in slices that go live and earn their keep, with a decision point after each. This converts one large irreversible commitment into several small reversible ones.
Fix the rate question honestly. If budget is the binding constraint, an offshore or nearshore team at a lower rate is the largest lever available — larger than any efficiency claim. The trade is time-zone overlap and communication overhead, and you should ask any supplier to state their guaranteed overlap in writing rather than describe it vaguely.
Fixed price or time and materials?
Neither is universally right, and the honest answer depends on how well the scope is known.
Fixed price suits well-specified work with stable requirements. You buy certainty and you pay for it — the supplier prices the risk, so a fixed price for a vague brief is always higher than the expected cost. Changes are change requests, which can become adversarial.
Time and materials suits work where you expect to learn something, which is most product development. It is cheaper when things go well and it transfers risk to you, so it requires either trust or close involvement.
A hybrid is what we would usually propose: fixed price for a discovery phase, then time and materials for the build against the scope discovery produced, with a cap you can stop at.
Common questions
Is offshore development cheaper overall, or just cheaper per hour? Per hour, substantially. Overall, usually, but less than the rate difference implies — communication overhead and time-zone gaps are real costs. The projects where offshore goes badly are typically ones with no technical lead on the client side, because then nobody is making decisions quickly enough.
Why would I pay for a discovery phase rather than get free quotes? Because free quotes against a vague brief are not comparable, and the padding you cannot see usually exceeds what discovery costs. Discovery also produces a specification you own, which you can take to any supplier.
What should a maintenance budget be? Plan for ongoing cost rather than treating launch as the end. Fifteen to twenty percent of the build cost per year is a reasonable planning figure for hosting, dependency updates, OS and platform changes, and small improvements. Software with no maintenance budget degrades.
Does AI make development cheaper? It makes some tasks faster, and it has not changed the parts that dominate cost — deciding what to build, modelling the data, integrating with systems that resist integration, and handling edge cases. Treat large claimed savings sceptically.
How accurate are estimates? Phase one, specified in most detail, is usually reliable. Later phases are broad ranges and should be presented as such. Anyone quoting a precise figure for twelve months of work does not know something you do not.
If you want a number for your own project, the fastest route to a useful one is a written scope. Our software discovery phase produces exactly that — specification, data model, architecture and a costed plan — as a fixed fee, and the document is yours whether or not you build with us. If you already have a specification, send it and we will quote against it.
Want this scoped properly?
Tell us what you need and we will come back with scope, timeline and a rate. No obligation to proceed.
Keep reading
Software Requirements Document Template (And What Makes One Unquotable)
Free software requirements document template with examples. Define scope, features and acceptance criteria before you hire developers. Download it now.
Best AI Automation Agencies in 2026: How to Compare Them
A buyer's guide to choosing an AI automation agency in 2026, with the agency types that exist, what each suits, and the questions that separate them.