How to Choose a Facebook Ads Agency (Malaysia)

The test, in one paragraph

Stop looking for the right agency and start looking for evidence you can gather yourself. There are three pieces of it, and all three are available before you take a single sales call. The first is public: open the free Meta Ad Library, look up the agency's own Page and the Pages of the clients it names, and read what is genuinely in the feed today. The second is structural: confirm that the ad account and the Pixel will sit in a business portfolio you own, with the agency holding access you can revoke. The third is contractual: read the fee model and work out which way the fee moves when your account changes. An agency that comes through all three may still be the wrong fit for your business, and one that leaves a question open on any of them is a risk you understand in week one rather than discover in month seven.

Disclosure: AdPlay.ai sells ad-creative software, so we are not a disinterested party here.

Why a ranking cannot be built from public data

Type "best Facebook ads agency Malaysia" into a search box and you will get lists. Before you use one, ask what data any list could possibly be ordering. The result that matters — what an agency achieved inside a client's ad account, at what cost per lead or per purchase — sits in Ads Manager, visible only to the client and to whoever the client granted access. It is not published, not licensed and not available to a researcher, a directory or a journalist. Meta's Transparency Center describes additional information such as spend, reach and funding entities for ads about social issues, elections or politics; commercial ads carry none of it. So whatever an ordering of Malaysian Meta agencies reflects, it cannot be measured results, because the measurements are not public in the first place.

That is a statement about the evidence, not about the people compiling it. It applies to every number you will read on this subject, including the fee figures further down this page: each of those comes from a company that sells the service, because the only Malaysian pricing we could find published for this market comes from companies selling it. We could not find a Malaysian industry body publishing an independent management-fee benchmark, or any neutral, dated survey to check published figures against. Treat every band as a figure the company publishes for its own service, dated and attributed.

The alternative to a ranking is not cynicism, it is a change of question. Instead of asking who is best, ask what you can verify. That question has real answers. You can verify whether an agency is running ads at all, what its clients' ads look like and how often they change, who will own your ad account and your Pixel, and how a fee moves when your spend moves. None of that produces a league table, and all of it produces a decision you can defend to whoever signs the cheque. If you want the landscape itself, our Malaysian agency directory is deliberately unranked and grouped by the specialism each company publishes for itself.

Look up the agency's own Page in the Ad Library

The Meta Ad Library is a searchable public archive of ads running across Meta's platforms. When Meta relaunched it in March 2019 it announced that the library now includes "all active ads any Page is running — not just ads related to politics or issues". That single sentence is what makes this workflow possible for an ordinary Malaysian buyer. You do not need an agency relationship, a login or anyone's permission to see what a Page is advertising right now.

Start with the agency itself, because it costs you two minutes. Set the country filter to Malaysia, search the agency's Page name, and look at what comes back. There is no correct answer here, only informative ones. An agency running its own lead-generation ads is spending its own money in the channel it is selling you, and you can read the offer, the hook and the landing destination it chose for itself. An agency running nothing is not disqualified by that; there are ordinary reasons a business does not advertise in a channel it sells, and the only way to learn the reason is to ask it directly on the call.

While you are on the Page, open its transparency panel and note when the Page was created and where its managers are located. Meta's help page describing exactly which fields appear did not render for us when we checked in July 2026, so read that list off your own screen rather than treating any published description of it as a specification. The purpose is not to catch anyone out. It is that any mismatch between what a Page shows and what a pitch deck says is a conversation you would rather have before signing than after.

Save what you find on the day you find it. Screenshot the ads, note the date you looked and record the exact Page name you searched, because Pages can be renamed and paused ads leave the view. A dated folder of screenshots is the difference between a verifiable note and a half-memory of something you saw last month. Keep that folder alongside your quotations, so that when three providers are on the table you are reading one set of your own notes rather than three sales pages written to three different templates.

What the Ad Library will show you

For each ad in the archive you get a consistent and useful set of fields. You see the creative itself, whether that is a static image, a carousel or a video. You see the primary text, the headline and the call-to-action button. You see which platforms it is delivering on, so a Facebook-and-Instagram buy is distinguishable from one that also runs Messenger placements. And you see the date the ad started running.

That is more than it sounds. Taken across a dozen ads it tells you what a team can actually make. If every ad in a client's account is a static graphic with the promotion written across it, the creative capability on offer is a designer and a template, which may be exactly what you need. If you see vertical video, talking-head clips, product demonstrations and carousels inside one account, someone is producing for the placement rather than resizing a single asset. You can also read positioning: which offer leads, whether the copy speaks Malay, English or both, and whether the call-to-action sends people to WhatsApp, an instant form or a website.

Do this across three or four of the agency's clients and you have something a portfolio deck cannot give you, which is unedited work. A deck shows the campaign someone chose to show you. The archive shows everything that is live, including the ordinary Tuesday work that makes up most of any account. For the mechanics of searching, filtering and capturing, the Meta Ad Library guide covers the tool in order.

What the Ad Library will never show you

Now the discipline, because everything above is worthless if you over-read it. Meta's Transparency Center states that "for ads about social issues, elections or politics, Meta provides additional information, including spend, reach and funding entities". For every other ad, which is to say all of the commercial work you are inspecting, there is none of that. No spend. No reach. No impressions. No funding entity. No targeting parameters, no audience definitions, no clicks, no conversions, no cost per result and no return on ad spend.

Two further limits matter. The Ad Library API, which would let you export at scale, is described by Meta as enabling deeper analysis of ads about social issues, elections or politics, plus ads delivering to the European Union and public branded content, with access restricted to authorised users. Ordinary Malaysian commercial ads fall outside all of those categories, so bulk export through the official API is not on the table, and any offer of bulk Malaysian commercial ad data is coming from somewhere else, which is worth asking about. Retention is similarly narrow: political ads are stored for seven years and EU ads are archived for a year after their last impression, while Meta documents no equivalent archive for a commercial ad in Malaysia. Assume the record disappears when a campaign is paused, which is another reason to capture on the day.

Hold one rule firmly for the rest of this process: nothing in the Ad Library is a performance claim. If a long-running client ad is offered to you as proof that an approach works, the accurate response is that Meta publishes no figure that could establish it either way, and that the same ad is consistent with a large budget or a tiny one. This is not scepticism for its own sake. It is the difference between evidence you can rely on in a negotiation and an impression you cannot.

Reading cadence and format mix as activity, never as performance

Within that limit there is still real signal, and it comes from patterns rather than individual ads. Cadence is the first. Line up the start dates across a client's live ads and ask how often new creative enters the account. A set of ads with start dates spread across recent weeks is the footprint of a team that keeps introducing new work. A set that all share a single start date tells you when those ads began and nothing about what has happened since, because anything paused has already left the view.

Format mix is the second. Count how many distinct creative concepts are live rather than how many ads, since one concept resized four ways is still one idea. Look at whether video and static coexist, whether the same offer appears with different hooks, and whether the copy changes between them. Several genuinely different angles running at once is what deliberate testing looks like from the outside. Fifteen near-identical ads is one idea uploaded fifteen times, which is a production choice rather than a fault.

Message variation is the third, and in Malaysia it usually shows up in language. A brand whose Malay ad carries a hook written for a Malay-speaking audience is investing in that segment. One running a word-for-word rendering of an English headline is doing something cheaper. That distinction is visible in the primary text, free of charge, before anyone quotes you anything.

Now the warning, and it is the most important paragraph on this page. Every one of those readings is evidence of activity, not of results. High cadence shows that someone is working, not that the work is profitable. A rich format mix shows production capacity, not conversion. Duration shows only that an ad has not been switched off, and any duration you calculate is right-censored by the day you looked and by how often a third-party archive re-checks. Use these signals to understand what a team produces and how often. Do not convert them into a score, and do not let anyone else convert them into one for you.

Look up the clients they name

Take the client names from the agency's own website and run each one through the same process. Three questions are worth answering.

Does the client have live ads at all? A logo wall is a claim about the past and says nothing about whether a relationship is current. This is not a gotcha, since campaigns pause for good reasons, budgets move quarterly and a seasonal advertiser is dark for months at a time. But if none of the named clients has a single live ad on the day you look, that is a straightforward thing to raise on the call and an easy thing for an agency to explain.

Does the live work match the case study? Compare what is in the archive against what is in the deck. Where they line up, you have corroboration you did not have to take on trust. Where they differ, you have learned what an ongoing month looks like next to what a pitch showcases, which is useful information regardless of which one impressed you more.

Does the work look like your business? An agency whose visible portfolio is property launches and aesthetic clinics has a genuine competence, but it may not be the competence a Shopee seller or a B2B manufacturer needs. You can read the pattern yourself instead of relying on a specialism claim. It is worth running the same search across your own category first, so that you arrive at the meeting knowing what is normal in your market and what an unusual idea would look like.

Own the ad account, own the Pixel

This section will save more Malaysian businesses more money than everything above it. Before any work begins, establish that you own the assets.

Create your own business portfolio in Meta Business Suite, then create the ad account and the Pixel or dataset inside it, and keep the Page, the catalogue and any WhatsApp Business account there too. Verify your domain from that portfolio. Then add the agency as a partner with access to exactly those assets, and to nothing else. When the relationship ends you remove the partner, and everything stays where it is: campaign history, conversion data, custom audiences and the learning that took months of spend to accumulate.

The alternative is that the assets are created inside the agency's portfolio instead. What you have then is access rather than ownership, and the difference only becomes visible at the exit. We could not confirm Meta's current documented behaviour on moving an ad account between portfolios, because its help pages did not render for us when we checked in July 2026, so treat that as something to verify in your own Business settings rather than something to assume in either direction. Either way the safe order is identical, and it costs nothing to follow it.

The Pixel deserves its own paragraph because it is the asset people forget. It accumulates the conversion history that makes optimisation work, and it is installed on your website, not the agency's. If the dataset sits in someone else's portfolio, you can end up without tracking on a site you own, on a day you did not choose. Set it up under your own portfolio from the start; our Business Manager setup guide walks through the order.

One related question belongs here rather than in the contract: whose payment method funds the media. If the ads run on the agency's card and are rebilled to you, ask what happens to the campaigns on the day an invoice is disputed, and ask to see the Meta billing receipts alongside the agency invoice. If the ads run on your card inside your ad account, you can read the charges directly at source.

What the contract has to say

Once ownership is settled inside the platform, write it down. A short annexe covering four points is enough for most Malaysian SME engagements.

Ownership. The ad account, Pixel or dataset, Page, catalogue and any creative produced under the engagement are yours. The agency holds revocable partner access. Say explicitly that access ends on termination and that no copies of your audience data are retained.

Handover. On exit the agency provides creative source files rather than exported JPEGs alone, plus audience definitions, campaign structure notes and the last twelve months of reporting. Agree this while everyone is friendly, because nobody negotiates it well during a separation.

Data. State the agency's role in relation to your customers' personal data, because this is no longer boilerplate in Malaysia. Since 1 April 2025 a data processor is directly bound by the Security Principle in section 9 of the Personal Data Protection Act 2010 in its own right, and a data controller must ensure its processors comply with it. The section 12A duty to appoint a data protection officer came into operation on 1 June 2025 and applies to controllers and processors alike, and the maximum fine for contravening the Personal Data Protection Principles is RM1,000,000. An agency that uploads customer lists or configures your Pixel is handling personal data on your behalf; our guide to Pixel consent under the PDPA covers the notice side of the same obligation.

Notice period. Read the notice period together with any minimum term and with the handover clause, because those three together decide how long an exit actually takes and what you walk away holding. If a clause is declined, ask why and write the answer down; the reasoning tells you how the engagement will be run.

What each fee model rewards

Fee models are incentive structures wearing a price tag, and the useful question is arithmetical: as your account changes from one month to the next, which way does the fee move?

Under a flat monthly retainer it does not move at all. That makes budgeting simple, and it means the fee is identical in a month of rapid growth and a month of standing still. Under a percentage of ad spend the fee rises with the media budget and falls with it, which is comfortable while the budget is climbing and less comfortable in the month when the sensible decision is to spend less. Banded pricing, where the fee steps up with your spend range, produces that same movement in coarser steps, and it can put a single large jump at the boundary between two bands, so ask what happens in a month that lands just above a threshold. A per-project fee does not move either, and it stops when the project stops, which suits a launch and leaves the following quarter unowned unless somebody agrees to own it. None of these is dishonest and none is best; they price different things, and the point is to know which thing you are buying before you sign.

The table below is an unranked, non-exhaustive, alphabetical selection of Malaysian companies that publish figures on their own websites, reproduced as each company states them and checked in July 2026. It is here to show the shapes a published fee can take, not to compare providers. Inclusion is not a recommendation and exclusion means nothing at all: this is a selection of pages that happen to carry a figure, and a page without one tells you nothing about the service behind it. Every row comes from a business that sells the service, which is the only kind of pricing information we could find for this market.

Read the table for structure rather than for price. A published figure can refer to a management fee, to the ad spend a page suggests a business plans for, or to a one-time setup charge, and those are three different quantities that happen to be printed in the same currency. Work out which one a figure refers to before you set any two figures beside each other, including the figures above; the third column exists to state what each row's number is attached to. Whatever shape a quote eventually takes, ask for one thing in writing: that the management fee and the media budget are separate lines, and that the media budget is paid to Meta rather than marked up in transit. Our breakdown of published Malaysian agency fees reproduces each of these published pages at greater length.

How to read a published fee page

A published figure is a starting point for a conversation, not a quote, and reading one carefully takes about five minutes.

Start with what the number refers to. Management fee, ad spend and total marketing outlay are three different things, and a page can publish any of them. If your total budget is RM6,000 a month and the fee is RM2,500, a little under half of your money is buying reach, which is a ratio worth deciding on deliberately rather than discovering later.

Then look at whether the figure is a floor or a fixed price. Wording such as "as low as" or "from" signals a starting point that scope can move; a flat monthly figure printed against a named package is a different kind of statement. Neither is better, but they behave differently once your requirements are on the table.

Then read the scope printed beside the number. How many ad accounts, brands and Pages does it cover? How much creative is produced each month, and does that include video? Who installs the Pixel and the Conversions API, and is that inside the monthly fee or a separate setup charge? What is the reporting cadence, and is anything promised about response times? A monthly retainer with no defined creative volume leaves the monthly output undefined, so ask for the volume in writing, in units you can count: so many static concepts, so many video edits, so many rounds of revision.

Then check the one-time charges and the term. A one-time setup fee spread across a twelve-month engagement changes the monthly comparison, and a stated minimum contract changes what happens if the fit turns out to be wrong. Ask whether the published figure includes service tax, and ask for a written quotation with a validity date, because a quotation and a published price are different documents.

Then turn the conversation into a document you can hold. Ask for the quotation to state, on one page, the monthly management fee, what is delivered for it each month, any setup fee and what it buys, the minimum term and notice period, the tax treatment, who holds the assets, and how the media budget is paid. Every item on that list is something you would otherwise have to reconstruct from memory in month three. A provider that puts them in writing has made the engagement easy to run, and any item left off is simply an item to ask about before you agree the rest.

Finally, convert everything into a single monthly line you can compare like for like: management fee, plus any setup fee divided across the term, plus the media budget, set against what is delivered each month. That number is the one to take into the meeting. Do the same arithmetic for every provider you speak to, on the same assumptions about budget and scope, or you will end up comparing answers to different questions.

Reporting, and the questions to ask before you sign

Agree the reporting cadence in writing before the work starts. A monthly read-out on its own is long enough for a month of budget to run before anything is written down, so ask for a brief weekly signal as well: spend to date, cost per result, what changed and what is being tested next. Five lines in a message is enough, and the discipline of writing them tends to surface problems early.

Judge the report by what it leads with. One that opens with impressions, reach and engagement rate and never arrives at cost per lead or cost per purchase is describing activity. A useful one states the business outcome and what it cost, explains what changed and why, and names the next test. Ask to see a real redacted example from an existing client before you sign, then keep that example as the agreed template so nobody has to relitigate the format in month three.

Agree the definitions at the same time as the format. A lead in a report is whatever the report counts as a lead, and a purchase is whatever the Pixel was configured to fire on, so settle in writing which events count, whether they are counted on click or on view, and over what attribution window. Two reports built on different windows are not comparable, and neither is your own month-on-month history if the setting changes halfway through the year.

Then ask the questions that tell you how the work will be planned, and listen to the shape of the answers rather than their polish.

• How many distinct creative concepts would you run in the first month, and how do you decide a concept is finished?
• What would you do in week three if nothing was working? A diagnostic order is a substantive answer; a single lever is not.
• How do you decide a result is real and not noise?
• Are there conditions under which you would advise me to spend less, and what would they be?
• Who writes the copy and who edits the video? In-house or freelance, and can I meet them?
• What do you need from me to do this well?

That last question matters most. Testing new angles runs on product knowledge, offers, customer objections and access to the people who answer your enquiries, so ask what they need from you, write the answer down, and check it against the work you actually want done. If what you want is a steady flow of new angles, the inputs that produce angles have to appear somewhere in the written plan, along with who supplies them and by when. That is a conversation to have before the first invoice rather than after it.

Checks that sit outside the archive

A few final checks have nothing to do with Meta. Ask for the registered company name and registration number, and check that the entity on the contract is the entity on the invoice. Ask who the named account contact is, whether the people in the pitch will do the work, and what happens when that person is on leave. Ask for a reference you source yourself rather than one selected for you, and when you speak to them, ask what the third month was like rather than the first.

Agree the escalation path before you need it: who you call when spend spikes, when a Page restriction lands or when an ad set stops delivering the night before a campaign. Ask how account access is secured, whether two-factor authentication is enforced on the people who will hold partner access, and how access is removed when a staff member leaves their team. These are unglamorous questions with short answers, and the answers are easier to get before a contract than after one.

Write down what you were told, by whom and on what date. Not as a trap, but because a selection process that runs over three weeks and four conversations will otherwise blur together, and the specific undertaking you half-remember is usually the one that turns out to matter. A one-page note per provider, filled in on the day of the call, is enough to keep the comparison honest.

The scorecard you take into the meeting

Put it on one page and fill it in before you talk to anyone.

Verification: the agency's own Page checked in the Ad Library, three or more named clients checked, cadence, format mix and message variation noted, and all of it recorded as activity rather than performance, with screenshots and the date you looked. Ownership: business portfolio, ad account, Pixel or dataset, Page and catalogue confirmed as yours, agency added as a revocable partner, payment method decided. Contract: ownership, handover, data role, notice period and minimum term written down. Money: fee model understood, media budget separate and unmarked-up, setup fees and tax clarified, and you can say out loud which way the fee moves when your spend moves. Process: reporting cadence agreed, metric definitions and attribution window agreed, a real report seen, and the six questions asked.

An agency that clears all five is worth your time. One that leaves ownership or the fee model unresolved is not thereby a poor choice, but you now know which risk you would be accepting and can price it into the decision instead of meeting it by surprise. Fill the same page in for every provider you speak to, on the same day and with the same questions, so that what you are comparing is your own notes rather than five differently written sales pages. If you want the same weekly jobs described from the in-house side, the guide on running Facebook ads yourself sets out that routine; the jobs have to happen either way, whoever ends up doing them. With 23.0 million Malaysians reachable on Facebook and 16.1 million on Instagram in late 2025 (DataReportal, 2026), the constraint was never the size of the audience. It is judgement, and the test above is how you find out who is bringing some.