36 minutes read
Send the same requirements to several vendors and you’ll get back completely different numbers. Some of them can even say that they will only be able to send numbers after you commit to discovery.
All these companies have their reasons and, with this blog, we will help you understand them. We will also lead you through the factors that impact the pricing – scope, compliance work, team seniority, and location, and give you the ranges we see in real deals.
Most custom software development projects cost between $50,000 and $500,000. A simple internal tool starts near $30,000. A complex regulated enterprise platform can run past $1M.
That range is so wide because “custom software” covers numerous services, from a single-purpose web form to a multi-tenant SaaS platform. Asking how much does custom software cost in general gets you an average, and averages hide things that actually set the number. The useful question wouldn’t be how much does custom software development cost in general, but what does your scope, at your compliance level, built by this team cost.
| Project size | What it includes | Team | Timeline | Typical range |
|---|---|---|---|---|
| Small tool | One workflow, basic admin, one integration | 2–3 people | 6–12 weeks | $30,000–$75,000 |
| MVP / first release | Core feature set, auth, one platform | 3–4 people | 3–5 months | $60,000–$150,000 |
| Mid-size product | Several modules, roles, 3–5 integrations | 5–7 people | 6–9 months | $150,000–$400,000 |
| Enterprise platform | Multi-tenant, reporting, legacy migration, audit trails | 8–12 people | 9–18 months | $400,000–$1.5M |
| Regulated build | Any of the above plus HIPAA, FDA, PCI or SOC 2 evidence | +2 roles | +25–40% time | $250,000–$2M+ |
*The ranges assume a mid-market vendor with a blended rate of $45 to $70 an hour – where most Eastern European and Latin American teams sit in 2026. For US firms, the scope will triple. **The rows don’t include integrations and compliance.
When a vendor says that the price depends, they’re usually right, but you can force the vagueness out of the conversation by asking what it depends on. As mentioned in the intro, four variables change the cost of custom software development:
Scope, integrations, and data volume set the size of the build; the regulatory bar you have to clear and the technology stack you’re building on set the multiplier. Let’s look at each of them more precisely.
Compliance changes how you build every feature, test it, and document. Architecture gets stricter: encryption at rest and in transit, tenant isolation, role-based access down to the field level. Every action needs an audit trail, which means extra data models and extra storage.
QA expands, because you now test evidence as well as behaviour, and someone has to write the test protocols for auditors. The paperwork itself becomes a deliverable, including policies, risk assessments, and validation documents that need engineering input.
When we speak of external costs, the first SOC 2 Type 2 programme in 2026 runs roughly $28,000 all-in for a 25-person startup, around $75,000 for a 100-person company, and $180,000 or more for an enterprise. Penetration testing adds $8,000 to $20,000 per application. HIPAA and FDA work carry their own documentation and validation overhead on top.
As a planning figure, add 15% to 40% to the same product cost without the regulatory bar. The low end covers GDPR-style data handling and a security review. The high end covers FDA-regulated clinical software, where validation and traceability are integrated into every release.
If a vendor quotes your regulated build at the same price as an unregulated one, they haven’t priced the compliance work, and you’ll pay for it later.
If your company already runs Microsoft 365, Entra ID, and Azure, a .NET build inherits identity, security policy, and hosting you’re already paying for. You won’t need a separate identity provider, new SSO integration, or a second set of security controls to explain to your auditor.
We did it for the client management system we built for UK financial advisory firm SKYNEXE. For them, staying within the Microsoft stack was a hard requirement, because their partner firms already held the licences and the compliance story had to match.
The second saving is scope you don’t build. Parts of an internal system often fit the Power Platform, and admin tooling that would cost around $40,000 in custom screens can sometimes ship as a model-driven app in a fraction of the time.
On the SKYNEXE project, the adviser-facing app runs on Power Apps with Dataverse and Power Automate, while the client-facing portal is a Blazor app on a shared .NET backend. Two very different cost profiles, one system.
The flipside is that Azure consumption is an operating cost that grows with usage, and Power Platform licensing is per user per month, so a system with 5,000 external users prices very differently from one with 50 internal. Model the licence cost for several next years, not just one. When done properly, Azure lowers total cost for Microsoft-centric companies.
The numbers below assume a mid-market team at a blended $45 to $70 an hour, with discovery included, and with no heavy regulatory burden. Add the compliance multiplier from the section above if it applies to you.
A first release with one user role, one platform, and the smallest feature set to the idea. For example, four people for four months at 160 hours a month and a $50 blended rate, which is 2,560 hours and $128,000. Anything quoted under $40,000 would be no more than a prototype with a login screen. It can be the right buy, but only if you know that’s what you’re getting.
A working internal or customer-facing system with multiple roles, reporting and two or three integrations. Most of the spread comes from integrations and permissions, not from screen count. Six months with five people is the common shape.
Multi-tenancy, billing, subscription logic, onboarding, admin tooling and per-customer configuration. Billing alone is often $20,000 to $50,000 once you handle proration, tax, and failed payments. Plan for a second wave of spend after launch, when you get valuable questions and feedback from your first customers.
Cost here is based on what you’re replacing. Data migration, parallel running, integrations to finance and HR systems, and change management for hundreds of users usually outweigh the feature work. 12+ months is a normal duration for such projects. Consider a phased rollout as a safer and cheaper option than a single cutover.
Cross-platform with Flutter or React Native runs roughly 30% to 40% cheaper than two native builds. That difference gets smaller if you need deep hardware access. Add store review cycles, device testing and a backend if you don’t already have one.
A retrieval-based assistant over your own documents sits at the low end. A trained model with a labelled dataset, an evaluation setup and monitoring sits at the high end. The often neglected line is inference: token or GPU cost is an operating expense that scales with usage, so it should be budgeted for at least two to three years.
Two companies’ prices can differ by 60%, purely because of the sector they operate in. They will ask you to produce different evidence: the failure cost of a bug and how much of the budget goes to people who are not involved in writing code. Here’s what we see across the industries we work in.
Healthcare builds carry three cost drivers at once: protected data, clinical risk, and old integration standards. HL7 and FHIR interfaces are rarely plug-and-play, and every hospital runs a slightly different configuration of the same system. Expect a 25% to 40% uplift over an equivalent unregulated build, most of it in QA, documentation, and integration work.
We saw all three on the Site Software platform we built with Selux Diagnostics. It’s a US company running rapid antibiotic susceptibility testing. Our team worked on machine learning infrastructure, database interfaces, and the LIS integration over HL7, alongside Selux’s own engineers, under FDA rules to help clinicians get results in about 5.5 hours and start targeted treatment one to three days sooner.
Five years on, we’re still on that platform, which tells a lot about the real lifecycle of regulated healthcare software.
Financial software prices at the top of the market because everything is traceable. Every action needs a log, every calculation needs to be reproducible, and access control has to survive an auditor’s questions. KYC, risk profiling, and regulatory reporting are often larger than the core product.
The previously mentioned UK financial advisory technology company, SKYNEXE, came to us with client onboarding running on email threads and PDF forms. The functional ask was to allow advisers to collect client data and for clients to fill in their own details.
Advisers and clients needed separate access paths into the same records. Every change to a client’s risk profile had to be attributable to a person and a point in time. The reporting layer had to produce numbers an FCA reviewer could trace back to the source record without taking anyone’s word for it.
That is why it became two applications instead of one. Wealth Pilot runs on Power Apps and Dataverse for advisers. Wealth Portal runs on Blazor and .NET for clients. The two stay synchronised in real time on Azure, with FCA-aligned risk reporting through Power BI. Read the full SKYNEXE case study.
Cost here concentrates in two places. First, integration breadth: carriers, warehouse systems, ERP, customs data, and telematics all have different formats, and each one is a small project. Second, field conditions. Software used on a loading dock or at a remote site needs offline behaviour, sync conflict handling, and interfaces that work on a cracked screen in bad light. That calls for a lot of design and QA effort, but is not always mentioned in first estimates.
Our work in this area included a field maintenance solution for a construction client and a parts cross-reference tool for industrial automation, both built around technicians working away from a desk. The mobile side was relatively cheap, but not the data reconciliation.
Ecommerce has the lowest entry cost of any category on this list, because Shopify, BigCommerce and their app stores cover most of what a store needs. Custom development earns its place outside the platform offerings: complex pricing rules, B2B account hierarchies, marketplace logic, ERP-linked inventory across warehouses, or a checkout that has to behave differently by region.
Budget accordingly. A custom storefront on a headless setup runs $60,000 to $180,000. A custom order and pricing engine behind it runs far higher. If your requirement is a catalogue and a checkout, we advise you buy the platform and spend your development budget on differentiating from your competitors.
Energy software costs more because the data never stops and the downtime is extremely expensive. Real-time ingestion from thousands of sensors, alarm rules, historical analysis, and both cloud and on-premise deployment options add architecture work you don’t need in a typical web app. Security review is heavier, and on-premise deployment means you’re supporting environments you can’t reach remotely.
We’ve worked with Sensia, a US provider of integrated automation for the oil and gas sector, for more than ten years on their web-based monitoring platform. Surveillance engineers use it to track well parameters in real time, catch problems such as stuck pumps or gas interference early, and shut down processes before faults become expensive. Deployment runs either in the cloud or on-premise, which is often a procurement requirement in this sector.
Biotech sits at the expensive end for the same reason as clinical healthcare. Results have to be reproducible, data lineage has to be provable, and instruments have to be integrated one at a time. Lab software also has a long life, so the cost conversation is about a decade, not a couple of years.
Our cell line research data platform has been in development for over four years, and our herd health management work is now past eleven. In biotech, the platform is the investment.
Where your team sits changes the cost to develop software more than any other single factor, which is why the average cost of software development is a global average of nothing in particular.
| Country | Median back-end developer pay, 2026 | Equivalent hourly |
|---|---|---|
| United States | $175,000 | $84 |
| United Kingdom | $108,913 | $52 |
| Germany | $87,011 | $42 |
| France | $71,929 | $35 |
| India | $22,086 | $11 |
| All countries surveyed | $79,742 | $38 |
Source: Stack Overflow Developer Survey 2026, 49,009 respondents with 23,928 answering the compensation question; the dataset is published under an open licence. Hourly figures are the annual median divided by 2,080 working hours.
The US Bureau of Labor Statistics puts the median wage for software developers at $135,980 a year, or $65.38 an hour, with the tenth percentile at $82,460 and the ninetieth at $214,670. That’s lower than the survey figure because BLS measures the whole occupation and counts straight-time pay, while the survey skews senior and includes bonuses.
Now the part the tables never show: a salary is not a rate. To get from one to the other, add employer taxes and benefits, which run 20% to 35% depending on country, then divide by billable hours, since no engineer bills 2,080 hours a year once you account for holiday, sick leave, training, and internal work. Add overhead and margin on top.
In practice, a vendor’s hourly rate lands at roughly two to two and a half times the salary-derived hourly figure. A German engineer at $42 an hour of wage cost bills somewhere near $85 to $105. An Indian engineer at $11 bills near $22 to $28.
Custom software development rates are the easiest thing to compare and usually the most misleading. Take a 2,000-hour project. At $80 an hour, it’s $160,000. At $30 an hour, it’s $60,000, and the saving looks great.
Now consider that the cheaper option usually skips the discovery phase, so 15% of the work will eventually be rebuilt. Two extra hours a day of your own management time. Or turnover that puts a new developer on your codebase months after signing. All of it later impacts your calendar, schedule, and financial department.
The four models below cover almost every custom software development pricing arrangement you’ll be offered, and each one shifts risk between you and the vendor in a different direction.
According to this model, you agree a scope, a price, and a date. The vendor carries the delivery risk, which is often the safest choice. However, a vendor quoting a fixed price on an uncertain scope adds a contingency of 20% to 40%, and you pay it whether or not the risk materialises. Fixed price also makes change expensive by design, because every adjustment goes through a change request, and the commercial incentive is to interpret the spec narrowly.
Fixed price works well when the scope is indeed fixed; when you have a defined integration, a migration with known endpoints, a redesign of an existing flow, a discovery phase. It works badly for anything where you expect to learn from users.
We always say on the consulting calls that fixed price is a good fit if you could write the acceptance criteria today and would still agree with them in six months.
This model implies that you pay for a team every month and direct their work yourself. Rates are per person, commitments are usually quarterly, and the scope can change without a contract amendment.
Most product companies opt for this model because the cost of a change request in the fixed-price situation exceeds the cost of flexibility here. A five-person team in Eastern Europe typically runs $22,000 to $38,000 a month all in, including project management and QA. You get predictable spend, people who accumulate domain knowledge, and the ability to reprioritise every two weeks.
In return, you need to provide a product owner who can make decisions at your end. A dedicated team without a decision-maker on the client side burns budget while being idle.
Full-cycle delivery, where the vendor owns discovery, design, build, release, and support against business outcomes rather than a task list. Pricing is usually a monthly team cost plus a defined delivery roadmap, sometimes with milestone gates.
The reason to choose it is not a price, but a capacity. If you don’t have a technical leadership team, someone has to make architecture calls, prioritise the backlog, and decline features. Buying that as part of the engagement costs more per month than a staffed team, but less than the cost of months spent building the thing you don’t need.
Our product development engagements usually start with a paid discovery phase precisely so that we map and price a real roadmap.
You pay for hours worked at agreed rates. It’s the fairest model on paper and the hardest to govern in practice. Time and materials works when it comes with guardrails: a monthly cap, a two-week planning cycle, and a burn report you actually read. Without those, it becomes an open tab. Ask for a not-to-exceed figure per sprint and a written escalation rule for when a task overruns its estimate by more than 25%.
When someone asks how much does software development cost, they usually mean the coding. Development is the smallest share of the budget that most buyers expect to be the largest. Here’s the split we use for planning, applied to a $200,000 build.
The ranges in the table overlap and won’t sum to exactly 100%. A compliance-heavy project pushes QA to the top of its band and design to the bottom. A consumer product does the opposite.
| Phase | Share of budget | On a $200,000 project | What you get for it |
|---|---|---|---|
| Discovery | 8–12% | $16,000–$24,000 | Scope, architecture, estimate you can hold someone to |
| Design | 12–18% | $24,000–$36,000 | Flows, UI, design system, usability testing |
| Development | 40–55% | $80,000–$110,000 | The working software |
| QA | 12–18% | $24,000–$36,000 | Test coverage, regression, release confidence |
| Project management | 8–12% | $16,000–$24,000 | Planning, reporting, decisions unblocked |
| DevOps | 5–10% | $10,000–$20,000 | Environments, CI/CD, monitoring, deployment |
Shares are ranges, so the low and high ends do not sum to exactly 100%. The actual split shifts with product complexity: integration-heavy builds pull budget toward development and DevOps, while regulated builds pull it toward QA and documentation.
The phase people cut first and regret most. Discovery turns a wish list into a scoped backlog, a technical approach, and an estimate with named assumptions. On a $200,000 project that’s $16,000 to $24,000 and three to five weeks. Skipping it doesn’t save you money. It moves the same work into the build, where it costs more and lands as a schedule slip.
A vendor willing to give you a firm price without discovery is either guessing or padding.
Covers user flows, interface design, a design system and usually a round of usability testing. The cost driver is the number of distinct user roles, not the number of screens, because each role needs its own journey. Internal tools sit at the low end. Anything customer-facing, where drop-off costs you money, sits at the high end. A design system pays back from the second module onward, so it’s worth funding early on a multi-year product.
The engineering itself: backend, frontend, integrations, data layer. If a quote puts development above 60% of the total, ask what’s missing, because it’s usually QA, project management, or DevOps hidden inside the developer hours or dropped entirely. The other thing worth checking is seniority mix. Two seniors and one junior will outdeliver four mid-level engineers on complex work, and cost about the same.
Test planning, automated coverage, regression runs, and release verification. Regulated projects push this to 20% or more, because you’re producing evidence as well as confidence. Defects found in production cost several times what they cost in a sprint, and in clinical or financial software they cost your reputation as well.
Sprint planning, reporting, risk tracking, and keeping decisions moving. This is the line clients most often ask to remove, and the removal becomes visible very quickly after the launch. When it happens, the developers absorb the coordination work at developer rates, or nobody does it at all, and the project drifts. On a distributed team, this phase isn’t optional.
Environments, pipelines, infrastructure as code, monitoring and alerting. Underfunding it is cheap for one release and expensive for the next fifty, because every manual deployment is a recurring cost and a recurring risk. Get monitoring in place before launch, not after your first incident.
Four patterns account for most overruns we see, and none of them are technical.
The first is deciding late. A two-week wait for a decision on a distributed team costs roughly the same as two weeks of that team’s time, because people either stall or work on something they’ll rebuild.
The second is building features nobody uses. It’s known that about 50% of features in a typical system are never used, and another 20% are rarely used. That’s most of a budget spent on software that sits there.
The third is treating integrations as small. A third-party API with poor documentation and a slow sandbox approval process can absorb weeks of a developer’s time before anything works.
The fourth is a too-late QA squeeze. Cutting test time to hold a date moves the cost into support, where it’s paid at a worse rate and in front of customers.
Published custom software development costs almost always describe the build alone. The build price is what you negotiate. The total cost of developing software is what you actually pay. Add up these five items before signing a vendor; together they routinely add 40% to 60% to the quoted figure.
The industry planning figure for ongoing maintenance is 15% to 25% of the original build cost annually. On a $200,000 build, that’s $30,000 to $50,000 a year for security patches, dependency upgrades, bug fixes, minor improvements, and framework updates that arrive whether you want them or not.
It isn’t optional spending. Unmaintained software becomes unsupportable software, and the rewrite costs more than a decade of maintenance would have. Ask every vendor for their post-launch support model and rate before you compare build prices.
Hosting, storage, backups, environments, and traffic. A modest business application on Azure or AWS runs $500 to $3,000 a month; a data-heavy platform with real-time processing runs far more.
Non-production environments are often half the bill, because staging and QA run all week. And cost scales with usage, so the pricing you approve at 200 users is not the pricing you get at 5,000. Ask for a projected infrastructure cost at your planned volume.
Payment processing, identity, mapping, email, SMS, document generation, data enrichment and now model inference. Each has its own pricing model, and several are per-transaction, which means your operating cost rises exactly when your business succeeds. A single mapping or verification service can easily become a $ 2,000-a-month line.
List every external dependency during discovery and price each one at expected volume, not at the free tier.
Moving history from an old system is its own project: extraction, cleaning, mapping, reconciliation and a parallel-run period where both systems are live. On enterprise replacements this often reaches 10% to 20% of the total. Training and rollout support add more. A system nobody adopts has a return of zero regardless of build quality, so budget for the humans, not just the code.
AI has changed custom software development costs less than the marketing suggests. What it has changed is the allocation of hours.
The most useful number for budgeting comes from Stanford University’s Software Engineering Productivity research, cited in DORA’s 2026 ROI of AI-Assisted Software Development report: AI delivers a 35% to 40% productivity gain on simple, greenfield tasks, and 10% or less on complex legacy code. So the discount is real on a fresh MVP and close to nothing on the old system you need to fix.
The same report adds a timing problem. Teams adopting AI hit a J-curve: a dip in productivity before the gain arrives, caused by the learning curve, the verification tax of reviewing generated code, and downstream processes like testing and release approval that have to absorb more code than before. DORA’s research also finds AI adoption still correlates with higher delivery instability, and their own worked example prices that instability as a cost.
Don’t take a large AI discount as a headline, because a vendor offering 40% off is discounting the review and testing work that AI produces more of. And ask how AI is governed: who reviews generated code, what happens to your proprietary data, and how they handle dependency risk.
So, how much does it cost to develop software? To answer this, you need to settle the scope. Here’s the sequence we use with clients.
1. Write the outcome in one sentence, with a metric. “Cut client onboarding from nine days to two” is a scope you can price and later prove. “Build a client portal” is not. Every feature that doesn’t move the metric is a candidate for phase two.
2. Cut to the smallest releasable slice. Ask what the first version needs to be usable by real people. Removing two user roles and one integration from version one commonly takes 30% off the price, and you learn more from shipping it than from weeks of planning.
3. Buy discovery separately, before the build. A paid discovery phase of three to five weeks produces an architecture, a backlog, and an estimate with named assumptions, for 8% to 12% of the eventual cost. It also lets you evaluate the vendor with real money at low risk. If discovery goes badly, you’ve spent $20,000 finding out instead of $200,000.
4. Price the same scope three ways. Get quotes from three vendors on the same written scope, and compare total cost of the outcome, staffing plan, and assumptions. If one is 60% below the others, the difference is almost always in what they didn’t include.
5. Hold contingency and a change budget. Fifteen to twenty percent, ring-fenced, spent deliberately. This is the single most reliable predictor of finishing on budget that we see.
Beyond the process, the biggest cost reductions come from buying instead of building. Authentication, payments, notifications, reporting, admin tooling, and document generation all have mature off-the-shelf options. Build the 20% of the system that only you could build, and assemble the rest.
Before you commit to a custom build, check whether you need one. We sometimes get enquiries that are better served by configuring an existing product, and we always say so.
| Comparison criterion | Custom build | SaaS subscription | Low-code platform |
|---|---|---|---|
| Year-one cost | $75,000–$400,000 | $10,000–$60,000 | $30,000–$120,000 |
| Ongoing cost | 15–25% of build per year | Rises with seats and tiers | Rises with users and limits |
| Time to first release | 3–9 months | Days to weeks | 6–14 weeks |
| Fit to your process | Exact | Approximate | Close, within platform limits |
| Ownership | Your code, your data | Vendor’s product | Partial, with platform lock-in |
| Breaks down when | Requirement is common and small | Your process is your advantage | Volume, complex logic, deep integrations |
Year-one cost covers licences or build plus implementation, data migration and integration work. None of the three options is correct by default: the deciding factor is whether the process being automated is a commodity or a source of competitive advantage.
A custom software cost that looks high compared to a SaaS subscription often looks different over time, once seat growth and configuration limits are priced in. The decision rule is as follows: if the process you’re automating is why customers choose you, build it. If it’s a process every company in your sector runs the same way, buy it. If it sits in between, and especially if it’s internal, low-code often wins on total cost.
One thing is worth mentioning about low-code. Costs are low at small scale and can climb sharply with external users, because most platforms price per user per month. Model your multi-year licence bill and check what an exit would involve.
Most quotes are hard to compare because they’re describing different things. A custom software development price is only meaningful once you know what it’s based on. Use this checklist to spot custom software pricing that hides risk instead of pricing it.
A firm price with no discovery means someone is guessing, and the guess will be corrected later at your expense. It’s a massive red flag. Another one is a quote that’s dramatically below the others. Send the same scope back and ask what it excludes.
Approving a cost for software development is a separate decision from producing the estimate. Cost is only half the case you have to make internally. Here’s how to build the other half in numbers your CFO will accept.
Start with hours. The cost of custom software development is easiest to defend when you can put a number on the work it removes. Take the process the software replaces, count the people doing it, multiply by hours per week and by a loaded hourly cost, which is salary plus benefits and overhead, usually 1.3 to 1.4 times base pay.
For example, a $180,000 operations platform replaces manual data entry for 14 back-office staff, saving each of them five hours a week. That’s 70 hours a week, 3,220 hours a year at 46 working weeks. At a loaded $38 an hour, that’s $122,360 saved annually. Subtract maintenance at 18% of build, or $32,400, and the net is $89,960 a year. Payback lands at roughly two years, and everything after that is margin.
Then add the effects that are real but harder to count: errors avoided, faster cycle times that let you take on more volume without hiring, staff retention when people stop doing work they hate, and revenue you can attribute if the system touches customers. SKYNEXE’s platform is the clearest version of this. It was built to fix onboarding, and it turned into a product other firms pay to use.
Use conservative numbers, because a too optimistic business case gets challenged and loses. Finally, don’t forget to set the measurement baseline before you start.
We’ve been building software for over thirteen years, mostly in .NET, Azure, and the wider Microsoft stack, with machine learning work alongside it. Several engagements mentioned in this article have run for five, ten, and eleven years. When we price, we quote the first phase of something we expect to still be working on in five years.
Our process starts with a short conversation about the outcome and its metric, then a paid discovery phase where a business analyst, an architect, and a designer produce scoped backlog, a technical approach, and an estimate with assumptions written down. Discovery typically runs three to five weeks and 8% to 12% of the eventual build. You own everything it produces, including the right to take it to another vendor.
After that, we price one of two ways. Fixed price where scope is settled, or a dedicated team with monthly rate if you expect to learn and reprioritise, and which covers most product work. Either way the quote lists all six phases, names the team by role and seniority, and prices post-launch support up front.
Most buyers reach this question by first asking how much custom software development cost overall, then narrowing to the simplest version of it.
A basic custom application with one main workflow, a simple admin area, and one integration typically costs $30,000 to $75,000, and takes six to twelve weeks with a team of two or three. Below $30,000, you’re buying a prototype that demonstrates an idea but lacks the testing, error handling, and security work.
Fixed price transfers delivery risk to the vendor, and they price that risk with a 20% to 40% contingency you pay regardless. It also makes every change a commercial negotiation. Fixed price is reliable when the scope is settled and the acceptance criteria could be written at the moment. For anything you expect to learn from, a monthly team cost usually costs less overall.
Compare the same written scope across three vendors and look at what’s included. Check that all six phases are priced, that the team is named by role and seniority, that assumptions are written down, and that third-party and support costs appear separately. Then check the blended rate against the regional prices.
Often yes. However, a $ 30-an-hour rate against $80 looks like a 60% saving, and rework, extra management time, timezone lag, and turnover usually eat part of it. To get real savings, buy discovery, keep a decision-maker available on your side, and compare vendors on total cost of a defined outcome instead of hourly rates. Cheap plus unmanaged is the most expensive combination.
15% to 25% of the original build cost every year.
On a $200,000 build, that’s $30,000 to $50,000 annually, covering security patches, dependency and framework upgrades, bug fixes and small improvements. Cloud hosting and third-party services are not included.
Yes, in almost every case. Discovery costs 8% to 12% of the project and produces the scope, architecture, and estimate that everything else depends on. Skipping it doesn’t remove that work; it moves it into the build, where changes are more expensive. It also lets you test how a vendor thinks before you commit the full budget.
Because the custom software development price covers a whole design process. The cost includes people’s time, and a typical mid-market team costs $45 to $70 an hour blended.
On top of that, only 40% to 55% of a budget is development. The rest is discovery, design, QA, project management, and DevOps, which are all necessary to end up with the system that works in production.
