Why nobody wants to write this down
Because “it depends” is true, and because a single number in a title is how you get the wrong kind of enquiry. Still: buyers search how much does custom software cost, software development cost, and bespoke software cost — and vague answers waste everyone’s time. Here are bands we can stand behind for work done properly — design, build, test, deploy, handover — not a weekend prototype.
| What you are buying | Typical shape | Ballpark |
|---|---|---|
| Paid discovery / prototype | 2 to 4 weeks, written scope at the end | Small, fixed |
| Focused MVP or internal tool | 8 to 16 weeks, one main workflow | Medium |
| Business system / custom application | 4 to 9 months, integrations | Larger, often phased |
| Multi-module platform (ERP-shaped) | 9 months and up | Programme, not a project |
We are deliberately not printing a rupee figure that will be wrong for your integrations. If a vendor gives you a precise custom application development cost before they have seen your data and your other systems, treat that as a marketing number. A serious estimate names the factors below.
How custom software cost is calculated
Software development cost India buyers see on the web is usually a logo and a range. Useful pricing is a stack of decisions. Change any row and the total moves. Ask a vendor to score these before you argue about the bottom line.
| Factor | What moves the number |
|---|---|
| Project complexity | One workflow vs many; happy path only vs exceptions; regulated data vs ordinary ops |
| Number of users | Five people on one desk vs branches, dealers, or public self-service |
| UI / UX | Admin screens staff will tolerate vs polished product UI and mobile layouts |
| Backend / API | CRUD behind login vs jobs, queues, webhooks, and third-party APIs |
| Database | Clean greenfield vs ten years of Excel with duplicate customers |
| Authentication | One role vs SSO, MFA, dealer logins, and records that must not leak across tenants |
| Integrations | None vs Tally, payments, mail, WhatsApp, weighbridge, supplier portals |
| AI | None vs extract-then-workflow with a human in the loop (models are a line item forever) |
| Hosting | Shared app service vs dedicated, backups, environments, who pays the cloud bill |
| Maintenance | Handover only vs a change allowance after staff actually use it |
| Support | A PDF vs a named thirty- or ninety-day window and a way to undo a bad deploy |
Project complexity
Complexity is not “number of screens”. It is how many ways the job can go wrong and whether two people can write the same rules. A leave form with three exceptions is cheaper than “make sales faster”. If the process is still an argument, you are buying process design first — same honesty as when to automate a process.
Number of users
Five internal users on one workflow is a different product from fifty dealers who must never see each other’s prices. Concurrent load, training, and permission design scale with who logs in — not with a licence slider on a brochure.
UI / UX
Staff tools can be honest and plain if the job is clear. Customer-facing or mobile-first work raises design and polish cost. “Make it look like our brand” mid-build is a change request, not a free theme swap.
Backend / API
Forms that save to a database are the cheap end. Background jobs, rate limits, idempotent webhooks, and APIs other systems will call are where calendar weeks go. If the sentence of done includes “and the other system updates”, price the API work as first-class, not as “integration TBC”.
Database and data quality
A clean schema is cheap. Migrating ten years of Excel with duplicate customers is a project of its own. Budget migration as a named slice or it will steal the last month of the build.
Authentication and roles
One shared password is not a system. Five roles that must not see each other’s records is a design problem. Dealer portals, SSO, and MFA add build and ongoing identity cost. Skip this line and you either overshare data or rebuild login in phase two.
Integrations
Talking to Tally, a payment gateway, a legacy desktop app, or a supplier portal is often more work than the screens. Each named system should appear on the quote with an assumption (API exists, test account available, who owns the other side when it is down).
AI
AI is not a free upgrade to a form. It earns its keep when input is messy — PDF, email, scan — and a person still owns the send or post click. Model usage is a recurring cost. Fixed steps with clean forms should stay ordinary workflow; see AI business automation and AI integration.
Hosting
Demo-on-a-laptop is not go-live. Hosting, SSL, backups, staging, and who gets the invoice when the site is live belong in the estimate. Omitting them makes the first number look small and the second month look like an emergency.
Maintenance and support
Handover of source is not the end of cost. Budget a change allowance for stories you only see once staff use the thing, plus a named support window. A system with zero maintenance line starts rotting while the spreadsheet beside it grows. Automation-shaped slices have their own bands on business automation cost.
Build versus buy, in cost terms
Off-the-shelf looks cheaper until you add licences, modules you do not need, and the consultant who customises it until it is a unique system you still do not own. Custom looks expensive until you add the years of workarounds you are already paying for in staff time. The comparison belongs on custom software versus off-the-shelf, not in a single cell of a spreadsheet.
How to keep the number honest
Start with an MVP: one workflow, real users, a written definition of done. Time and materials with a cap, or a fixed phase with a change process, both work. What does not work is a fixed price on an unfixed problem.
What we actually see on quotes in India
A focused internal tool — one login, one workflow, hosted, handed over — is often in the low-to-mid lakhs if the integrations are few and the data is clean. A system that replaces several spreadsheets and talks to Tally, a website and a branch is a multiple of that, usually phased. A “whole ERP by December” quoted as a single fixed number before anyone has seen the data is not a quote. It is a wish.
We will not print a rupee figure that will be wrong for your weighing scale, your dealer portal, or your GST exceptions. What we will say: if two vendors differ by 3× on the same one-page brief, one of them has not priced the integrations, the migration, or the handover. Ask them to show which.
Line items that disappear from the first email
- Discovery that produces a written scope you can argue with.
- Test accounts and a week of the real users clicking, not the vendor demoing.
- Hosting, backups, SSL, and who gets the bill when the site is live.
- Thirty or ninety days of support after handover, named in writing.
- Training that is not “we will send a PDF”.
A cheaper build that omits those is not cheaper. You will buy them later, in a hurry.
How we price so the number can move without a fight
A short paid discovery, then a fixed first slice with a change log. Extra work is written down and priced before it is done, or it is refused. Time and materials with a cap works when the problem is still being understood. What we will not do is lock a twelve-module system to a date and a rupee figure on a first call.
Hours you will spend, even if you hire well
Someone from the business sits with the build every week. They answer “what happens when the supplier is new?” They look at the first real data. They introduce the people who will click it on Tuesday. If that person does not exist, the quote is not low enough to save you. You will pay the hours later, in rework.
Plan for a day a week from an operations lead on an MVP, more on a system that replaces several spreadsheets. That is not the vendor being difficult. That is how you avoid encoding a guess about GST or branch stock.
How to read two quotes that are not the same job
Put them in a table: discovery included or not; first slice defined or “the system”; Tally / payments / mail named or “integrations TBC”; handover of source and hosting; support window after go-live; your hours per week. Price is the last column. A low number with empty cells is how you fund a change request in week six that was the actual project.
If you only have a one-line brief, both quotes are fiction. Write the brief. Then the 3× gap usually collapses, or it becomes obvious who left out migration. That is the only useful meaning of “competitive tender” on custom work.
What “medium” and “larger” actually contain
A medium MVP is one login, one workflow, hosted, tested with real users, handed over with source and a way to change a rate. It is not twelve screens of wireframe. A larger programme is the same thing, repeated: each phase has its own definition of done, its own parallel run if it touches money or stock, and a stop rule. Treating the programme as one fixed price is how you get a number that moves in week six.
India quotes often hide the difference by listing modules. Count workflows, integrations, and who will sit with the build. Three modules with one integration and a clean data set is still medium. One “module” that talks to Tally, a weighbridge, and three branches is not. If you cannot tell which you were quoted, ask them to redraw the quote as slices. The rupee figure then has a place to sit.
What you keep paying after go-live
Hosting, backups, SSL, a way to change a rate, and a named window of support. A cheaper build that omits those shows up as an emergency in month two. Budget a change allowance — small stories you will only see once staff use the thing. A system with zero change budget is a system that starts rotting while the spreadsheet next to it grows.
Bring a one-page brief that scores the factors above. That is how you get a usable custom software development cost — not a magic ₹X from a rate card. If you are still choosing who to trust with the estimate, use how to choose a software development company. For .NET-shaped systems, the work itself sits under .NET development.