Ads in Social Media: A Practical Planning Guide

TL;DR

Ads in social media are bought through a platform auction that weighs bid, predicted action rate and creative quality, so weak creative raises cost even when budget rises.
Social advertising creates demand rather than capturing it; search still collects existing intent faster, so judging social by search conversion rates misreads both channels.
No platform publishes how many social ads exist, and the number would be useless: read competition from your own CPM, frequency and unspent budget instead.
The number of ad sets you should run equals total budget divided by the weekly spend each set needs to exit the learning phase, which for small accounts is often one or two.
No reliable public CPM, CPC or cost-per-lead benchmark exists for Thailand; derive a ceiling from your own margin and close rate, and record a weekly baseline from your own account.

Ads in social media are paid placements bought inside social platforms, where a real-time auction decides which ad each person sees rather than a buyer booking a fixed slot in advance. You choose an objective, a budget, an audience signal and a creative, and the platform's delivery system spends that money against the people it predicts are most likely to do whatever your objective names.

The thing to understand up front is that almost nobody scrolling a feed is looking for you at that moment. Attention has to be manufactured rather than collected. That single fact shapes nearly everything that follows: how the copy is written, how results are measured, and what a reasonable expectation looks like in month one.

Social advertising means ad inventory bought directly through a platform's own ads manager. The unit sits inside content people already came to consume: the main feed, stories, the short-video surface, in-app search, or the message inbox. Delivery is decided one impression at a time. Every time somebody opens the app and a slot becomes available, an auction runs to decide which ad wins that slot.

An organic post reaches whoever the recommendation system decides deserves it, based on the post's engagement signals and each user's relationship with the page. You control very little about who sees it or how many. Paid placement gives you three things organic cannot: control over which group is reached, control over how often a person is reached in a week, and control over when to accelerate or stop. Organic has its own advantage, since it costs nothing per person reached and builds page credibility that money does not buy directly. The two do different jobs and are usually run together.

How it differs from search advertising

Search advertising works on demand that already exists. Somebody types a query because they want something at that second, and the matching signal is the query itself, which is an unusually clear declaration of intent. Social advertising has no equivalent signal. What it has instead is in-app behaviour history, whatever data you send back through a pixel or conversions API, and the response patterns of people who resemble those who converted before. Search pulls people who already raised a hand; social interrupts people who have not. Both have a place, but importing conversion-rate expectations from one channel to judge the other is the most common planning mistake.

How the auction decides

The highest bid does not automatically win. The major platforms describe the same three-part score: the amount you are willing to pay, the estimated probability that this particular user takes the action your objective names, and the quality or relevance of the creative as judged by user reaction. The practical consequence is that a creative people stop for and act on buys inventory more cheaply, while one people scroll past or hide gets more expensive over time even if the budget goes up. Creative is therefore a cost variable, not a decorative choice.

The other widely misread mechanic is the learning phase. Every time a new ad set is created or something material is changed in an existing one, the system re-enters an exploratory state while it works out which kind of person responds to this creative. During that window the cost per result swings hard and should not be used to make decisions. Meta's help documentation states that an ad set generally needs roughly fifty optimisation events per week to exit learning. That is a platform mechanic rather than a market benchmark, and the real status should always be checked in your own account.

The reason so many Thai businesses lean on social advertising is not that it is cheap. It is that social is one of the few channels that creates demand rather than harvesting it. If your product is something people do not know exists, nobody searches for it, search volume is effectively zero, and money poured into search has nowhere to go. Social is the one place where a product can be put in front of people who never thought about it, at volume, with frequency you can control.

It is worth being blunt about the other side so that expectations stay honest.

It is strong at building broad awareness at a low cost per person reached, at explaining products that have to be seen to be understood, at accumulating audiences to retarget, and at opening conversations that get closed in chat.
It is mediocre at selling expensive, slow-consideration purchases on first contact, because the viewer is not in a decision-making frame and needs several more layers of retargeting and content.
It is weak at capturing demand that already exists. If people are typing your service name into a search box every day, search advertising will collect them faster and more cheaply every time.
It cannot compensate for a product nobody wants or a broken destination. Ad budget only makes an existing problem visible sooner.

At the level of a business plan, social ads fill the top of the funnel and intent channels harvest the bottom. Cutting social because its reported cost per acquisition looks worse frequently produces a drop in branded search volume a few months later, which is a consequence the platform's own reporting cannot see.

People usually think of ad types as a list of formats, but there are two axes to choose on at once, and the second one affects results considerably more than the first.

Axis one: the format

Format is what the user sees. A static image is the fastest thing to produce and suits testing one sharp claim. Short vertical video suits stories and short-video surfaces, where attention keeps shifting. A carousel lets people swipe through several cards in one unit, which fits multiple products or a sequence of steps. Collection and catalogue-driven ads assemble themselves from a product feed you upload, so the specific item a person already viewed can be shown without producing a creative for every SKU. In-app lead forms and click-to-message ads both remove the website from the path entirely, which changes conversion rate and measurement at the same time.

Axis two: the objective

The objective is the instruction telling the system what to learn from. Pick awareness and it hunts for the cheapest impressions. Pick traffic and it hunts for people who habitually tap links, who are not the same people who habitually buy. Pick engagement and it finds people who like and comment. Pick leads and it finds people who fill forms or open chats. Pick sales and it leans toward people with a history of buying online. This is why traffic campaigns so often deliver plenty of clicks and few orders. The creative is not necessarily bad; the system was told to go and find clicks, and it did so faithfully.

The table below summarises which formats commonly pair with which objectives and what each combination is genuinely good at. Treat it as a planning starting point rather than a rule.

Axis two: the objective
Format	Objective it usually pairs with	What it is good at
Short vertical video	Awareness, video views or traffic	Introducing a product to people who have never heard of it and collecting viewers to retarget
Static feed image	Traffic or engagement	Testing a single message or offer at low production cost
Carousel	Traffic or sales	Showing several products in one unit, or explaining steps that need an order
Catalogue or collection	Sales	Serving the exact item a person already viewed or added to a cart
In-app lead form	Leads	Collecting contact details without sending anyone to a website
Click-to-message	Leads or messages	Starting a conversation in the inbox where the sale actually closes

Six social ad formats you can run: short vertical video, a still feed image, a carousel, a product catalogue, an in-app lead form, and a click-to-chat ad.

One pattern is worth remembering: the formats that make a result easiest to produce also tend to produce the lowest-quality results. An in-app form completed in two taps yields many records, but the share of people who answer the phone afterwards is usually lower than for a web form somebody had to type into. Comparing cost per lead between the two without looking at what happens downstream leads to the wrong conclusion.

Strategy here looks different from a few years ago because granular audience selection has been steadily demoted. Platforms have consolidated interest options, removed sensitive categories, and pushed advertisers toward broad targeting paired with automated delivery. What is genuinely left under your control is two things: which signals you feed in, and which creative you send out.

Creative has become the targeting

When the audience is set broad, the creative does the selecting. A video that opens on a kitchen pulls in people interested in kitchens, the system learns from that behaviour, and it expands toward similar people. In other words, the hook and the first three seconds now function as a more powerful targeting instrument than any interest checkbox. Teams that can produce several genuinely different angles tend to reach lower costs than teams with one beautiful angle.

The creative testing loop

Testing works when it is a repeatable loop rather than a change made whenever the current ad feels stale.

Write the hypothesis as one sentence before production, for example that this group cares more about time saved than about price.
Produce three to five executions that differ on the angle of that hypothesis, not on button colour or font.
Run them inside the same ad set so they compete under identical conditions, and let them run until enough data accumulates.
Judge at the downstream outcome, not at click-through rate. A high-click ad that never closes is pulling the wrong people.
Keep the winning angle as the spine, produce the next generation from it, and retire the losing angle by changing the hypothesis rather than the artwork.

There is no reliable public number for how often creative should be refreshed. What does work is reading frequency alongside cost per result in your own account. When frequency climbs and cost per result moves with it, that creative has exhausted this audience. The point where that happens differs enormously between a business with a wide audience and one with a narrow one.

Funnel sequencing and the ceiling on retargeting

The usual structure has three layers. Prospecting goes after new people using broad or lookalike audiences. Retargeting goes after people who already interacted: finished a video, visited the site, or started a chat and went quiet. Retention works on existing customers for repeat and expansion purchases.

The common failure is putting nearly all the budget into the second layer because its numbers look best. Retargeting has a low cost per result because it harvests interest the first layer created, not because it creates new demand. Reduce prospecting and the pool flowing into retargeting shrinks with it, and within two to four weeks total volume falls even while the unit economics still look healthy. The simple check is to track the size of each retargeting audience weekly. If it keeps shrinking, prospecting is being starved.

Budget structure

With a limited budget, the most damaging habit is splitting it across many ad sets. Each ad set has to accumulate its own events to exit learning, so six ad sets means feeding six learning phases from the same pot. Everything stays in an exploratory state, costs swing constantly, and no single ad set gathers enough data to support a decision. The safer approach for a small account is to consolidate into one or two ad sets and create variety at the creative level instead.

As for how much to start with, there is no trustworthy public benchmark for the Thai market. No platform publishes a national CPM, CPC or cost per lead, and the figures that circulate in articles usually come from a handful of accounts in one industry. The answer that holds up is derived from your own economics. Start with the profit you make on one sale, multiply by the rate at which leads become sales, and you have a ceiling on what a lead can cost. Multiply that ceiling by the number of events the system needs per week and you have the minimum weekly budget that lets one ad set function. If that number is above what you have, the fix is to optimise toward a more frequent event, such as a conversation started rather than a purchase, instead of running underfunded.

Measurement and numbers that disagree

Platform-reported conversions and your own back-end numbers will not match, and they disagree for structural reasons rather than because somebody misconfigured something. Platforms report on a user basis and look backwards to an ad the person clicked or saw inside a configured attribution window. Your back end usually credits the last session that delivered somebody to checkout. A person who sees an ad on a phone and two days later types the brand name on a laptop is an ad conversion in the platform report and a search visit in your analytics.

Three concrete responses help. First, pick one source of truth for money decisions, normally the system that records actual revenue. Second, use the platform report for internal comparison, such as which creative beats which, because that is the job it does well. Third, record the ratio between the two numbers every month. When the ratio is stable you can use platform reporting for daily decisions while knowing roughly how much it overstates. When the ratio moves sharply, that is a signal that tracking broke, not that performance dropped.

This gets searched often and has no credible answer. No platform publishes the total number of ads running on its system. Some public ad libraries let you look up ads advertiser by advertiser, but they do not provide totals, and their coverage varies by country and by ad category. Anyone quoting a global total for social ads is estimating without a source.

More to the point, even a true total would be useless. You are not competing with every ad in the world. You compete with the ads bidding for the same slot, in the same country, against the same audience, at the same hour, which might be a few dozen advertisers. Two better questions replace it.

First: how crowded is your auction

Competitive pressure can be read from your own account without knowing anything about system-wide ad counts. A few signals do the work. A CPM that rises across the whole account at once while creative stays unchanged usually means more advertisers entered the same pool, not that your ads got worse. Frequency climbing unusually fast means the system cannot find new people inside the audience you defined, which is an audience-size problem rather than a competition problem. Budget going unspent during certain hours means your bid is too low to win slots when contention peaks. Separating these three symptoms matters, because the remedies point in different directions.

The habit worth building is recording your own account's weekly CPM across a full year. After a couple of months you have a personal baseline more accurate than any published benchmark, because it reflects your industry, your geography and your seasonality. Festival periods and the double-digit e-commerce campaign dates show clear price movement, and that is planning information rather than cause for alarm.

Second: how many ads should you be running

The right number does not come from instinct or from how many ideas the creative team produced. It comes from budget divided by the data volume each ad set needs. If an ad set needs roughly fifty optimisation events a week to leave learning, and your account's cost per event sits at a certain level, multiplying the two gives the weekly budget per ad set. Total budget divided by that figure gives the maximum number of ad sets the account can actually support. For most small businesses the answer is one or two, which is far fewer than expected.

Inside a single ad set, three to five genuinely different creatives is usually enough. Beyond that the system concentrates delivery on whichever looks good early, the rest barely serve, and you learn nothing about them. Adding more creatives does not buy more testing when the budget stays the same.

The examples below are structural patterns that a small or mid-sized business can copy. They are illustrative patterns, not case studies, and no results are claimed for any company. What matters in each is what the pattern is trying to do and how you would judge whether it worked.

Click-to-message for a Bangkok service business

This pattern uses an image or short video that states the service and the service area plainly, with a button that opens a chat window instead of a web page. The first message is pre-filled so a person can send it with one tap. The point is to shorten the distance between interest and conversation, because most service businesses close in chat rather than on a page. The metric to watch is not how many people tapped through, but the share who reply after your team responds and the number of appointments that actually get booked. Plenty of taps followed by silence usually means the ad promised something the chat did not continue.

Catalogue ads for an online store

This pattern pulls products from a feed synced to the store and assembles creative automatically around the items each person viewed or added to a cart. The point is to remind people while the intent is still alive. What usually breaks it is not the creative but the feed: product titles that do not contain the words people actually use, stale prices, or out-of-stock items still showing. The metrics that matter are average order value and return rate, not cost per order in isolation.

Lead forms for a high-consideration purchase

For expensive purchases such as property, insurance or long programmes, in-app forms remove a lot of friction, and they also invite people who are nowhere near ready. The counterweight is one or two qualifying questions hard enough to filter idle submissions, such as an approximate budget or an intended timeframe. Judge this pattern on the contact rate and the rate at which contacts become appointments, not on cost per lead alone.

Short vertical video for cold prospecting

This pattern exists to open the funnel. The ones that work usually start with a problem or situation the viewer recognises within the first three seconds and reach the product afterwards, rather than opening on a logo. The honest metrics for this layer are not same-day sales but mid-point view rate, cost per person reached, and the weekly growth of the retargeting pool. If those three improve while total sales stay flat, the problem is normally in the next funnel layer rather than in the video.

Planning ads in social media for Thai consumers carries conditions that make a plan lifted from a Western market fit badly. The first is platform mix. Thai users run several apps at once and use different apps for different jobs. Chat sits at the centre of buying to a degree that surpasses many markets, so paths that end in a conversation frequently outperform paths that end in a web form across a lot of categories.

The caution is that platform share moves, and figures quoted in articles are usually older than they look. Rather than trusting a number in any article, including this one, verify with the platforms' own planning tools in the month you are planning, and with the audience data in your own account, which tells you more directly where the specific customers you want actually are. Treat platform choice as a question to test rather than a fixed answer.

In practice, Facebook advertising remains the base layer for many businesses because of its mature catalogue and retargeting tooling. TikTok advertising suits cold prospecting through short video and usually needs different creative from what runs on a static feed. LINE advertising holds a particular position in Thailand because it connects to follower bases and messaging systems businesses already use for existing customers. Choosing does not mean choosing one; it means knowing which app carries which job in the funnel.

The third condition is language. Translating ad copy literally from English into Thai generally performs worse than writing it in Thai from the start, because sentence rhythm, politeness level and the words people actually use for a product all differ. Direct sales phrasing that reads fine in English often lands stiffly enough in Thai that people scroll past. Keep the structure of the offer, have somebody who writes natural Thai rewrite it, and mine the words customers type in your chat inbox, which is the best language source available and already sitting in your possession.

The last condition is reply speed. When most paths end in an inbox, a good cost per lead means nothing if the team answers too late and the person has gone elsewhere. Before raising budget, check that the people answering chats can absorb the extra volume, because this bottleneck sits outside the ad system and does more damage to results than anything inside it.

The difference is the level of control and the objectives you can choose. Boosting is a shortcut with limited options, usually optimised toward engagement and unable to configure measurement fully. Building a campaign in the full ads manager lets you pick an objective that matches a business outcome, control placements, build retargeting audiences and split-test creative. For a first experiment on a very small budget boosting is acceptable, but the moment sales or leads need measuring, move into the full system.

How much budget should you start with per month

There is no public benchmark for the Thai market that answers this honestly. The way to find the answer is to work backwards: start from the profit you can accept per acquired customer, convert that into a ceiling for cost per lead or per order, then multiply by the number of results per week the system needs in order to learn. That gives the minimum budget for one functioning campaign. If the figure exceeds what you have, optimise toward a cheaper and more frequent event rather than spreading the budget thinner.

How many ads should run at the same time

Let the budget decide rather than the number of ideas. Calculate the weekly budget one ad set needs to accumulate enough data, then divide total budget by it. The result is the maximum number of ad sets the account supports, which for small businesses is often one or two. Inside each, run three to five creatives that differ on message angle. Adding ad sets without adding budget leaves all of them stuck in learning simultaneously.

Why do platform conversions not match the back end

Because the two systems count differently by design. The platform looks backwards to an ad the person clicked or saw within a configured window and can connect activity across devices when the user is logged in. Most web analytics credit the last source before the purchase. That gap will never close. What you can do is choose one source of truth for money decisions and watch the ratio between the two monthly so you notice the moment tracking breaks.

Yes, but the expectation about timing has to change. B2B decision cycles run much longer, and somebody who sees an ad today may buy months later, so demanding sales inside the first month usually leads to shutting campaigns down prematurely. A structure that holds up better uses social to build recognition and accumulate interested audiences, supports it with content that answers real buyer questions, and lets intent channels harvest the demand when buyers are ready.

If you are weighing up which platform to start with, or already running and finding the numbers too unstable to decide anything, sort out account structure and measurement before adding budget. A planning view of social advertising is a reasonable place to work out how to sequence the funnel and split budget for a business of your size.