The hard part of multi-region corporate communications isn't writing — it's proving every version says the same thing.
A Bursa-listed retail group with hundreds of outlets across Malaysia and the region publishes its quarterly results on a Wednesday morning. The announcement has to go out in three languages — English for the exchange filing, Bahasa Malaysia and Chinese for the press and the public — inside the same tight window. Three people, working from the same approved English statement, produce three versions. By the time someone notices the Chinese release phrases the dividend slightly differently from the English one, the wire has already carried both. This is exactly where AI corporate communications tools are quietly useful.
Nobody did anything wrong — that's the point. And it's worth being precise about where AI for corporate communications helps, because the honest answer is narrower, and more valuable, than "AI writes your press releases."
The problem isn't writing. It's reconciliation
Ask any investor relations or corporate communications lead at a listed Malaysian company what eats their announcement day, and it's rarely the drafting. The board-approved message exists. The problem is everything after: turning one approved statement into several language versions, then proving — under deadline — that all of them say exactly the same material thing, in the right voice.
Two things drift, and they drift in opposite directions:
- Facts drift quietly. A figure gets rounded differently. A percentage is stated as "more than 10%" in one language and "11%" in another. A "final dividend" becomes, through a translator's honest choice of word, something that reads closer to "special." Individually trivial; in a disclosure, material.
- Voice drifts loudly. The English reads like the company. The translated versions read like a translation — flatter, more literal, subtly off. For a brand that has spent years building a recognisable tone, the non-English markets get a weaker version of it, every single time.
For a private company, inconsistency is an annoyance. For a listed one, it's a compliance surface. Bursa Malaysia's Main Market Listing Requirements put continuing, accurate and non-misleading disclosure of material information at the centre of an issuer's obligations — and "we said it differently in the Chinese release" is not a defence anyone wants to test (Bursa Malaysia Listing Requirements). Singapore's regime is built on the same principle: SGX RegCo's continuous disclosure rules require material information to reach the market completely and consistently (SGX Rulebooks). If you operate across both markets — as a growing number of Malaysian and Singapore groups now do — you are reconciling the same message against two rulebooks and three languages at once.
Why "just use machine translation" fails here
The instinct is to throw a translation tool at it. That solves the wrong half of the problem.
Generic machine translation is genuinely good now, but it optimises for fluency, not for fidelity to an approved source under your brand voice. It will happily produce a fluent Bahasa Malaysia sentence that softens a claim, drops a caveat, or renders a regulated term inconsistently across two paragraphs. It has no concept of "this number must be identical to the source," "this disclaimer is non-negotiable," or "this is how our brand says things." And it certainly won't stop and tell you when it's unsure.
There's a reason the localisation industry has known for years that language is a business risk, not a clerical one — CSA Research's long-running "Can't Read, Won't Buy" work has repeatedly found that the large majority of consumers won't act on information that isn't in their own language, and act more readily when it reads natively (CSA Research). Multilingual communication isn't a nice-to-have you can bolt on with a translate button; for a lot of your audience it is the communication.
So the useful question isn't "can AI translate this?" It's "can AI catch the drift a human under deadline will miss, and keep the voice, without ever publishing on its own?"
What an AI corporate communications agent actually does
That framing points at a specific kind of AI system — not a chatbot, and not a translator, but an agent built around a check. The pattern Anthropic calls evaluator-optimizer fits it almost exactly: one part generates, another part evaluates against explicit criteria, and the generator revises until it passes (Anthropic — Building effective agents). Applied to corporate communications, it looks like this:
- It starts only from the message you've already approved — plus your brand-voice guide and your approved glossary. It never invents the source.
- A drafter writes each language version in your house voice, not a literal translation.
- A separate checker extracts every figure, date and material claim from the source and from each draft, and diffs them. A number that reads differently, a claim that's been softened or added, a dividend labelled differently in one language — surfaced, not buried.
- Where the check fails, it revises and re-checks, and loops until every material point matches and the voice holds — or, if it can't resolve something, it stops and flags it.
- Then it routes the whole set to your comms or IR lead with the flags called out. A human signs off. Nothing publishes on its own.
The shift is subtle but the whole value is in it: the machine isn't there to be trusted with the release. It's there to make the reconciliation — the slow, error-prone, deadline-compressed part — fast and auditable, and to turn a drift that used to be caught after publication (or not at all) into a flag a human clears in minutes. We've written up exactly how this is built, use case by use case, in our corporate communications consistency agent breakdown — including the worked example of the "final vs special dividend" catch.
The honest limitations
A few things this does not do, because pretending otherwise is how these projects lose trust:
- It doesn't decide what to disclose. That's your board's and your advisers' call. The agent is a control on consistency, not a substitute for judgment on substance.
- It doesn't remove humans. It removes the reconciliation grind. Your comms team still reviews and signs off — the review is just now minutes on the exceptions instead of a full re-read of every language.
- It's only as good as its guardrails and its evals. Ours don't go live against a real release until they clear a golden set of your past announcements — including the tricky, price-sensitive ones — on every figure and every material claim. If it can't beat that bar, it doesn't ship.
That last point matters more than any feature. An agent that's confidently wrong on a disclosure is worse than no agent. The engineering that earns trust here is unglamorous: extraction accuracy, claim-level diffing, a hard gate, a logged trail you can reconstruct after the fact.
What it costs to think about
The economics are modest relative to the risk. In practice the AI usage for drafting-and-checking a single announcement across a few languages runs on the order of a few Ringgit — call it RM 2–8 — and a company putting out 30 announcements a month is looking at low hundreds of Ringgit in model cost. The build — connecting your brand guide, your CMS or newswire, and your internal channels — is a matter of weeks, scoped after a short discovery.
Set that against the cost of a single inconsistent disclosure that has to be clarified to the market, or a brand that reads like a translation in half its markets, and the sizing question answers itself. If you want to model the numbers for your own volume and in your own currency, our Claude cost calculator will do it in RM or SGD.
Where this is heading
The broader shift is that "agentic" AI is most useful not where it replaces a person, but where it sits between a person's judgment and a risky, repetitive, deadline-bound task and adds a check. Investor relations and corporate communications are close to a perfect case for it: the message is human-owned, the languages are many, the deadline is hard, and the cost of an inconsistency is real. The teams that get value from AI corporate communications here won't be the ones who hand it the keys — they'll be the ones who use it to catch what they'd otherwise miss, and keep the sign-off firmly human.
If your announcements go out in more than one language — and especially if you're filing on Bursa Malaysia, SGX, or both — the question worth asking isn't "should we use AI to write faster?" It's "where in our process does a version quietly drift from the approved source, and what would it take to catch it every time?"
Sources
- Bursa Malaysia — Listing Requirements: continuing and immediate disclosure obligations for issuers
- SGX Rulebooks: continuous disclosure rules for Singapore-listed companies
- CSA Research — "Can't Read, Won't Buy": why consumers act on information in their own language
- Anthropic — Building effective agents: the evaluator-optimizer pattern
This article is general information, not legal or disclosure advice — treat it as a starting point for your own counsel and board, not a substitute for them.
免费咨询
Publishing the same message in three languages?
Tell us what you publish and in which languages, and we'll tell you honestly whether a corporate communications consistency agent is worth building for your team — and what it would take.