Cost

What a second language costs

A second language adds roughly one to two delivery days for a chat-only bot and more where a Mini App is involved. Translation is the cheapest part. The cost is right-to-left layout, mixed-direction strings in inline keyboards, and the fact that nobody on the project can proofread every language.

What a second language costs: effort, cost and what moves it

Fixed price: $1,680 USD
Delivery: Awaiting sign-off

Effort: 4 delivery days
Biggest cost driver: Right-to-left support and localisation architecture
Commercial basis: Fixed price agreed before work starts; scope changes quoted separately
Year one, beyond the build: Support retainer, Managed hosting, plus translation services, model usage for multilingual ai

As of 2025-10-01, Telegram Bot API 13.4

Why this costs what it costs

Dubai businesses need at least two languages and often three, and the estimate for that is routinely built as though the work were translation. It is not. Getting the strings translated is a service you can buy by the word from several competent local agencies, and it is a small line in this table. What costs money is everything the second language does to the interface.

Arabic is the expensive one, for reasons that are specific and fixable. Text runs right to left, which means Mini App layouts mirror rather than merely restyle. Strings that mix Arabic and Latin content — a price, a booking reference, a phone number inside an Arabic sentence — render in an order that surprises people who have not built for it, and getting them right requires explicit direction marks rather than hoping. Inline keyboard buttons have limited width and Arabic labels do not fit where English ones did, so button text has to be rewritten rather than translated. None of this is difficult; all of it is work that does not exist in a single-language build.

Russian is cheaper because the mechanics are the same as English — left to right, familiar rendering — but it carries a different cost: length. Russian strings run noticeably longer than their English equivalents, which breaks buttons and wraps messages in places the design did not anticipate. In this market Russian is frequently the second-most-valuable language after English, ahead of Arabic for some segments, and it is worth choosing on your actual customer base rather than on assumption.

The hardest problem is quality assurance, and it is organisational rather than technical. Somebody has to read every string in every language in context and confirm it is not merely correct but appropriate — the right register for a clinic, the right formality for a broker. On most projects nobody on either side can do that for all three languages, and the gap is filled by a machine translation nobody checked. Budgeting a named reviewer per language, on your side, is the single highest-value thing you can do for this line item.

The breakdown

Work	Days	Cost	What it covers
Localisation architecture	1	$420	String extraction, a language preference stored per user, fallbacks when a string is missing, and a structure that lets a fourth language be added later without a refactor.
Second language integration	1	$420	Wiring the translated strings through every flow including error paths, and the language switch itself. Assumes translation is supplied rather than commissioned.
Right-to-left support	1	$420	Arabic only. Mirrored Mini App layouts, direction marks on mixed strings, and inline keyboard labels rewritten to fit rather than translated into overflow.
Per-language content review	0.5	$210	Per language. Reading every string in context with your reviewer and correcting register, formality and terminology. The part that decides whether it reads as native.
Notification localisation	0.5	$210	Reminders, receipts and alerts in the user's language rather than the business's. Frequently forgotten, and the most visible failure when it is.
Total	4	$1,680	Fixed on signature. Scope changes are quoted separately, never absorbed silently.

What moves the number

Inline keyboard buttons have limited display width and no wrapping.

Arabic and Russian labels overflow where English fits. Button text is rewritten per language rather than translated, which is a copywriting task rather than a localisation one.

Mixed-direction strings require explicit direction marks to render predictably.

A reference number or price inside an Arabic sentence appears in the wrong position without them. It is subtle, it looks like carelessness to a native reader, and it is invisible to a team that does not read Arabic.

Telegram exposes a user's client language code, which is a hint rather than a preference.

It is a good default and a bad assumption — plenty of people run an English client and want to be spoken to in Arabic. An explicit switch is required regardless of how good the detection is.

An AI assistant needs a separate corpus and evaluation set per language.

Retrieval is not translatable. Multilingual assistants cost close to one assistant per language, which is the largest multiplier in this whole area and the one most often missed in a budget.

Russian strings are materially longer than their English equivalents.

Layouts that fit in English wrap awkwardly in Russian. Cheap to handle when anticipated, and a scattering of small visual defects when it is not.

Nobody on the project reads all three languages fluently.

A named reviewer per language on your side is the practical quality gate. Without one, unreviewed machine translation ships and your customers are the ones who find it.

What year one actually costs

Running cost	Year one	Detail
Support retainer	$3,480/yr	Every copy change becomes a change in every language. The monthly allowance absorbs that; without a retainer, one language quietly drifts out of date behind the others.
Managed hosting	$660/yr	Unchanged by language count. Strings are small and add nothing meaningful to the running cost.
Translation services	billed by the provider	Billed per word by whichever agency you use, and the smallest cost in this whole area. Use a human agency for customer-facing copy in a market this competitive.
Model usage for multilingual AI	billed by the provider	Applies only to the assistant tier, and rises with language count because each language carries its own retrieval context rather than sharing one.

Buying the same thing elsewhere

Where you buy it	What the money buys	What you carry
English only	The lowest cost and the fastest delivery, and in several Dubai segments — B2B, tech, much of the professional services market — genuinely sufficient. Do not buy languages you cannot justify.	Excludes customer groups where Arabic or Russian is the language of trust rather than of necessity. The loss is invisible in your analytics, because those customers never start a conversation.
Automatic machine translation at runtime	Every language at once for a small per-character cost and almost no build time. Tempting, and occasionally the right answer for a purely informational bot.	Register and terminology are wrong in ways native speakers notice immediately, and errors reach customers with nobody having read them. For anything transactional, this reads as carelessness rather than as reach.
A separate bot per language	Complete independence, no shared string management, and each bot free to differ. Occasionally correct when the audiences and offerings genuinely differ by language.	Every feature is built and maintained several times, and the versions drift apart within months. Usually a decision made to avoid localisation work that ends up costing more than the work would have.

Questions about the money

Which languages do businesses here actually need?

English is the baseline. Whether the second is Arabic or Russian depends genuinely on your customer base — a clinic in Deira and a crypto community in Business Bay have different answers. Look at where your existing enquiries come from rather than at a generic demographic split.

Why does Arabic cost more than Russian?

Direction. Right-to-left means mirrored layouts in Mini App views, explicit direction marks on any string mixing Arabic with Latin characters or digits, and inline keyboard labels rewritten to fit a narrower space. Russian shares English's mechanics and only brings length.

Can we add a language later?

Yes, and cheaply, provided the localisation architecture was built at the start — which is why it is the first line here rather than an option. Retrofitting it into a bot with strings scattered through the code is a rewrite of every flow.

Do you supply the translations?

We can commission them, but the better arrangement is that you supply them or nominate the reviewer, because register matters more than accuracy in customer-facing copy. Your team knows how your customers expect to be addressed and we do not.

How does the bot know which language to use?

It defaults to the client language code Telegram provides and offers an explicit switch that persists. The default is a reasonable guess and nothing more — plenty of people run an English interface and prefer to be spoken to in Arabic.

What about an AI assistant in two languages?

That is close to two assistants rather than one bot with two dictionaries, because retrieval is not translatable — each language needs its own corpus and its own evaluation set. It is the largest multiplier in this area and it belongs in the budget from the start.