Every cost guide in this market ends on the same line: budget 15–25% of the build per year for maintenance. The figure is real and it is repeated because it is roughly right on average. It is also the least useful way to plan, because it ties your running cost to what you paid rather than to what you built.
Two AED 150,000 projects can have running costs an order of magnitude apart. The percentage cannot see the difference. What follows is what the difference is actually made of.
Two different budgets wearing one name
"Maintenance" is usually two unrelated things billed as one. Separating them is the single change that makes the number planable.
| Keeping it alive | Continuing to change it | |
|---|---|---|
| What it is | Dependency and OS updates, certificate renewals, store resubmissions, monitoring | New features, new reports, new integrations |
| Driven by | What you shipped | What you decide to do next |
| Optional | No. External change forces it | Yes. Entirely |
| Behaves like | A fixed annual floor | A budget line you open and close |
| Right shape | A small support arrangement or a few days a year | A monthly partnership, or nothing at all |
Most disappointment about maintenance cost comes from paying a monthly retainer sized for the second column while only consuming the first. That is not a vendor being dishonest. It is a mismatch nobody named at signing.
What sets the floor
The unavoidable annual work is a function of how many external parties your software depends on. Each one changes on its own schedule and does not ask you first.
- A mobile app in the stores. Apple and Google each ship a major OS release every year, and both periodically raise the minimum SDK you must build against to stay publishable. Apple charges USD 99 a year for the developer account, Google USD 25 once — taken from Apple's and Google's own published fees rather than from a market guide.
- A payment gateway. Gateways deprecate API versions and rotate credentials on their own timetable, and a payment path that stops working is not a bug you get to schedule.
- Personal data. PDPL rights such as erasure and portability are features that have to keep working as the schema moves underneath them, which means they need testing after changes that appear unrelated (reconn).
- Anything with a certificate or a domain. Small, boring, and the most common cause of an outage that looks catastrophic for an hour.
- Servers. The one line every guide gives a range for and nobody decomposes. Skimbox publishes AED 300–1,200 a month during a build and AED 2,000–12,000 a month at scale (2026) — a forty-fold spread, which is the next section.
What the infrastructure line actually is
Hosting is the one running cost that does not need estimating, because the vendors publish their list prices and anyone can read them. So rather than quote a market range, here is the bill decomposed into the line items it is actually made of, at the shapes of project we ship. Every figure below is a published vendor price converted at 3.6725 dirhams to the dollar, the rate the dirham has been pegged at since 1997 (Central Bank of the UAE). The bands are ours and derived: the arithmetic is checkable, the choice of stack is a judgement.
| What you shipped | The line items | Per month |
|---|---|---|
| Internal tool, no public traffic | Serverless platform free tier to USD 5, managed Postgres free tier | AED 0–18 |
| Customer-facing web product with a database | Serverless platform USD 5 or Vercel Pro USD 20, managed Postgres from USD 25 | AED 92–165 |
| The same, published in both app stores | Apple USD 99 a year, Google Play USD 25 once | + AED 30, plus AED 92 once |
| Ten times the traffic on any of the above | 90 million requests beyond the 10 million included, at USD 0.30 per million | + AED 99 |
| A managed support or compliance tier | Same database, team plan instead of pro, from USD 599 | AED 2,200 |
Read the last two rows together, because they are the whole point. Multiplying your traffic by ten adds about a hundred dirhams a month. Buying a support tier multiplies the bill by more than twenty. The published AED 2,000–12,000 at scale is real, but it is not the price of being popular — it is the price of a contract, and it is a decision rather than a consequence.
Where the published figure does hold
For a customer-facing mobile or web product with payments, accounts and personal data, 15–25% of build cost per year is a reasonable planning figure and both Netguru and Skimbox publish it independently. Skimbox works the example through: an AED 60,000 build carrying AED 9,000–15,000 a year (2026).
Note what that number does not include. It is the cost of the software continuing to exist. It is not the cost of anyone using it — published UAE guidance puts marketing at 30–50% of development cost, which is the larger number and the one more likely to be missing from the plan entirely (Netguru, 2026).
What to ask before you sign the build
- What is the annual floor for this specific scope, in dirhams, assuming we ask for no changes at all?
- Which third parties can force work on us, and what is the notice period each of them typically gives?
- What happens if we want nothing for six months and then something urgent — are we buying availability, or buying work?
- Who holds the accounts, the certificates and the store listings? If the answer is the vendor, the running cost includes a dependency you did not price.
A vendor who can answer the first question with a number rather than a percentage has thought about your project specifically. That is most of what you are trying to find out.