"What does an app cost?" has no single answer — but there is a method behind the number, and it can be explained. Below is how we estimate internally. You can use it to judge any vendor's quote, not just ours.

Cost comes from three parts

Part What it covers Notes
Engineering time Requirements, design, development, testing, launch The bulk of the cost, and it grows faster than scale does
Third-party services LLM APIs, SMS, maps, payment channels, push notifications, identity verification Usage-based; charges begin only after launch
Infrastructure Servers, bandwidth, storage and databases — plus GPUs for on-premises deployments Ongoing, budgeted annually

Most arguments over a quote come down to the first item. The other two have published pricing — estimate the usage and you can calculate them.

Engineering time is driven by things you can count

Instead of asking "roughly how much?", count the items below — the range narrows immediately:

Dimension How to count it Effect on time
Screens Customer-facing plus admin — count every distinct screen The most direct measure of scale
API endpoints How many distinct actions need to talk to the backend Drives backend and integration work
Roles and permissions How many roles, and whether permissions differ field by field Each extra role multiplies the test combinations
Platforms Mini-program only; both mobile platforms; either of those plus an admin console The backend and design carry over to a second platform; the client build and its testing do not
Payments and reconciliation Do you take payments, split settlements, or handle multiple currencies? Payments add noticeably to both testing and compliance
Data source Your own data, or integration with the client's existing systems Third-party integration carries the most uncertainty
Deployment Cloud, or on-premises On-premises adds environment setup and acceptance work

Which end of the range are you on?

Once you have counted, use the table below to place your own project:

Dimension Bottom of the range Top of the range
Platforms One platform Two platforms, or two plus an admin console
Roles and permissions One or two roles, permissions set per module Several roles, permissions set field by field
Payments No payments, or single-payment checkout only Split settlement, refunds, reconciliation, multiple currencies
Data source All data is your own Integration with existing client systems
Deployment Cloud On-premises, including environment fitting and audit
Compliance No sensitive personal data Personal data, finance or healthcare

If most rows land on the left, budget for the bottom of the range. If two or three land on the right, budget for the top. The right-hand column carries more weight for these reasons:

  • Payments and reconciliation. Taking money is not the hard part; refunds, split settlement, reconciliation and exception-order handling are — together with the testing each of them adds.
  • Multiple roles and permissions. Going from three roles to six is not twice the work — what grows is the number of permission combinations that have to be tested.
  • Integrating the client's existing systems. The quality of the other side's API documentation, its stability and its response times are all outside our control. This is the single largest schedule risk.
  • On-premises deployment. Fitting the client's server environment, network policy and audit requirements makes the acceptance cycle longer than it is for a cloud deployment.
  • Regulatory requirements. Where personal data, finance or healthcare are involved, the compliance review is itself a phase of work.

Timelines for common project types

These are our time ranges for common project types; where a given project falls in its range depends on the dimensions above. These are timelines, not prices — but timeline is the main driver of cost.

Project type Phase Timeline
Mini-program (with admin) MVP 3–5 weeks
AI application (LLM integration) MVP 2–4 weeks
E-commerce app (both platforms + payments) MVP 6–12 weeks
AI agent Proof of concept 1 week
AI agent MVP / full version 2–4 weeks / 6–12 weeks
Custom AI skill Scoping, build and delivery 1–2 weeks
Enterprise knowledge base Standard edition A few weeks (depending on document volume and number of data sources)
GEO foundation work First round 2–4 weeks

Every figure above is for an MVP or a first round of delivery; a full version is quoted per module on top of the MVP. The spread within each row comes from the table in the previous section. Whether an e-commerce app lands at six weeks or twelve depends on multi-merchant support, whether reconciliation is built in-house, and how many currencies are involved.

Estimating annual running costs

After launch, budget these four items separately; together they are your annual running cost:

Item Basis What moves it
LLM calls Token usage Daily active users, conversation length, whether retrieval is involved
Cloud servers Instance spec and runtime hours Concurrency, and whether autoscaling is needed
Bandwidth and storage Traffic and capacity Images and video, and whether a CDN is used
On-premises One-off hardware plus hosting Model size determines GPU memory requirements

A small cloud project and an on-premises deployment with GPUs can differ substantially in annual cost, depending on model size and concurrency. We estimate against your actual expected usage rather than quoting a generic figure.

How we quote

  1. Requirements review. Work through the dimensions above and produce a scope inventory.
  2. Fixed quote. Price and timeline are fixed against that inventory. We do not bill open-ended day rates.
  3. Changes go through assessment. Small adjustments that do not affect the architecture are usually covered by the project; anything that adds a module or changes the technical approach is assessed, quoted, and scheduled once you approve it.

We do not publish prices on the site, for exactly the reason this page describes: a number detached from scale tells you nothing and can mislead your budget. For a number you can actually use, send us your situation and we will quote a fixed price after the review.

Our position on acceptance criteria, source-code ownership and the maintenance period is in the FAQ.

Related cases

Changzhongbao — AI Digital-Human Companion App AI App

Changzhongbao — AI Digital-Human Companion App

A digital-human companion app for older adults: a family member clones their face and voice from one short video and one voice recording, so the elderly user can talk to a familiar “family” avatar and follow courses inside the app, with consent and compliance prompts throughout.

数字人形象克隆声音克隆 TTSiOS / Android实名与授权合规
Taurus Payment Gateway Enterprise Systems

Taurus Payment Gateway

A payment aggregation gateway integrating major Indonesian banks and Southeast Asian e-wallets, giving cross-border businesses one place to collect, reconcile and settle.

Question not covered here?

Send us your situation. We reply within one business day, at no charge and without the run-around.