Facebook ads for SaaS companies

Start with the number, because it is the whole page. Facebook reaches 80% of adults aged 30 to 49 against Instagram's 62%, and 74% of 50- to 64-year-olds against 40%. Those are the bands in which somebody owns the business rather than works in it. Now hold that against how software is actually sold. A horizontal, product-led tool — the note-taker, the analytics dashboard, the invoicing app — is bought by the person who will use it, and that person skews to the other app, which is why the Instagram twin of this page is written for them. Vertical software — the routing system for a pest control company, the practice management system for a clinic, the shop system for four bays — is bought by the person who signs, and that person is here. Same industry label, two different companies, two different playbooks.

The consequence shows up first in the objective mix, and it is the only place in this whole cluster where the platform axis moves it. The Instagram batch was pinned as five Traffic, three Leads, two Sales and no Awareness — Traffic-led because a self-serve subscription completes in the ad's own session. This batch is six Leads, two Traffic, one Sales and one Awareness. Nothing about the industry changed. The buyer did, and an owner with staff and a contract does not complete a purchase in a feed.

Second, the shelf, which Instagram does not have. Meta's placement picker for an image ad lists Facebook Feed, in-stream video, video feeds, Marketplace, Stories, ads on Reels, search results, Business Explore and the right-hand column — nine surfaces against Instagram's four — and Advantage+ placements fill all of them from one image and one headline. Their recommendations disagree: Facebook Feed wants 4:5 at 1440 × 1800 with primary text of 50–150 and a headline of 27, and lists no description at all; Marketplace wants 1:1 at 1080 × 1080 with primary text of 125, a headline of 40 and a description of 30. So the working rule is the same one the gym guide arrived at from a different direction — write the headline to 27 and compose the portrait so a square crop out of its middle is still the ad.

The measurement is worth having in front of you, because it is now three Facebook batches deep and it does not move. The ten ads on the Browse page produced fifty headlines: five fit Facebook Feed's 27 characters, and all fifty fit Instagram's 40. That is the same five-in-fifty the gym Facebook batch produced. Nine of the ten lead headlines clear 27 and none clears 40. Every primary text overruns Meta's 50–150 recommendation, and this batch overruns it harder than any before it — 322 to 708 characters, against the gym batch's 218 to 542. B2B copy is the longest copy this pipeline writes, which makes the first clause more load-bearing here than anywhere else on this site.

Finally, the surfaces this app has that the other does not, and what they are actually good for. The right-hand column and search results take a 1.91:1 asset; Instagram Feed accepts that ratio too (corrected across this cluster on 7 September 2026 — the shape is allowed there, the surfaces built to want it are not) but has nowhere to route it. For software that matters more than it does for a local trade, because the comparison ad — "X vs Y", the format a large share of SaaS budget goes into — works best against somebody who has just typed a competitor's name, and search results are the only placement in either app that is adjacent to that. Two things to know before you spend on it. Those surfaces render a wide asset small, so one line of type is the budget; the Browse page's wide compliance ad puts six lines in and is unreadable in the right column. And in a vertical, naming the competitor is exactly where this pipeline goes wrong on its own — see the mistakes section.

The Instagram twin of this page argues for Traffic by default and calls it the reverse of every other industry on this site. That argument is correct and it does not survive the app change, which is worth saying plainly rather than quietly writing a different table. Traffic leads there because a self-serve subscription completes in the ad's own session. Nothing an owner-operator buys completes in the ad's own session, so the table below looks much more like the dentist's than like its own twin's.

The Browse batch reflects it: six Leads, two Traffic, one Sales, one Awareness. The one Sales ad is the one with a published per-truck price and an online signup behind it — which is the test, rather than anything about the industry. If a stranger can finish buying without speaking to you, Sales is on the table. If the purchase involves moving data, training a front desk, or a signature, it is a lead however much your pricing page wishes otherwise.

The place to be careful is Leads itself, because it has two conversion locations here and choosing the wrong one is the most expensive default in this account. An Instant Form collects an email address from somebody who has never seen the software, which for a $150-a-month vertical product is a name and a nurture sequence rather than a prospect. The destination section is about which instrument to use and when; the short version is that on this app the qualifying step you add is the one that pays for itself.

The Instagram twin organises this section around a role and a moment, because its reader is a practitioner having an ordinary evening. That is the wrong unit here. An owner is not being asked to recognise their Tuesday; they are being asked to recognise a fact about their business — a cost, a contract, a dependency, a thing only they can do. So each row below pairs an owner with a decision they have already half-made and are avoiding.

Ten of the twelve rows have a finished ad behind them, which is a first for this cluster's SaaS pages. The Instagram batch used only six of the ten angle families and named the four it skipped — Offer & Economics, Product & Mechanism, Scarcity and Founder POV — as the four hardest to do honestly for software. This batch ran all ten, and those four turn out to be the easy ones once the buyer is an owner-operator: there is a published price to divide, a migration to draw, a contract with a renewal date, and a founder who used to do the job. The two rows without an example are without one on purpose and say why.

One structural note that holds across all of them, inherited from the twin and unchanged: the first line names the reader, because the ad set cannot. "Owners of three to eight bay independent shops" is a superb targeting instrument and it is not in the picker.

The good news first, and it is the same good news the Instagram twin opens with: software is not a special ad category. Meta accepts exactly four — housing, employment, financial products and services, and social issues, elections and politics — and none of them is an industry list you belong to. Age, geography, lookalikes, exclusions and custom audiences are all yours. Note that you must still declare something: since 31 March 2020 every campaign created through the Marketing API has had to specify a value, and businesses offering none of the four specify NONE.

The bad news is the second half of that sentence, and it is new to this page rather than inherited. The category attaches to what an ad offers, not to what your company is filed under. A software ad whose creative offers financing, a payment advance, a deferred first invoice, or a lending feature inside the product is a financial-products ad, and that category has been required for advertisers based in or reaching the United States since January 2025 — Meta's own reference gives two dates on the same page, 14 January in its banner and 21 January in its milestones table. Its restrictions are identical to housing's: fixed 18–65+ age, no lookalikes, no saved audiences, a forced minimum radius. Declaring it costs reach; not declaring it costs the campaign, because Meta reads the creative regardless and treats working around a category as evasion. If your product line includes anything that moves money before the customer earns it, that is a decision to make deliberately at campaign level rather than discover at review.

Then the practical problem, which is the twin's problem with a different answer. You want people who own a business of a particular kind and size, and the ad set cannot ask for that any more than it can ask for a job title. Meta has narrowed detailed targeting repeatedly and its own documentation on what survives is not safe to paraphrase, so do what the twin says: open the picker in your own account and plan from what is actually in it. The recommendation does not depend on the answer. Go broad and let the first line name the business — three to fifteen trucks, four bays, two centres — which excludes everybody else for free and keeps the audience large enough for delivery to work with.

Where this app genuinely gives you something the other does not is geography, and it is worth using rather than defaulting past. Vertical software is often sold market by market — a state's subsidy rules, a province's compliance regime, a county's licensing — and an owner-operator audience is dense in exactly the places a broad national buy wastes money on. Set the geography to the markets you can actually support and invoice in, exclude your own customers and current trialists, and seed lookalikes from paying accounts rather than from trial signups.

One trap that is in the account rather than in the ad, and that only bites part of this industry. If your customers are clinics, Meta's Business Tools terms prohibit sending data derived from health information — and that prohibition explicitly covers the names given to events, conversions and custom audiences. A pixel event called physio_claim_recovered, or a custom audience called "dental practices", is health information in the field nobody audits. Name events after the commercial action and nothing else.

The Instagram twin's verdict is a landing page almost always, and Instant Forms only for a genuinely sales-led motion. Here the sales-led motion is the normal case, so the whole recommendation moves — and the reason is not the platform, it is that an owner buying a system for a business with staff does not start a trial at eleven at night and discover the product alone.

So the default is a form, and the type matters more than the fact. Meta's two Instant Form types differ by one step: More volume submits in a thumb movement from pre-filled profile data, and Higher intent adds a review screen. On this app that screen is a filter worth paying for, because a cheap lead from an owner who half-remembers tapping something costs you a sales call rather than an email. Add one qualifying question that changes what happens next — how many trucks, how many locations, which system you are on now. The last of those three is the most useful question in vertical software and the one people are most willing to answer.

The call is a real destination here and it is under-used. Call ads are mobile-only, deliver on Facebook Feed, Marketplace, Reels and video feeds, and Ads Manager reports calls placed, 20-second calls and 60-second calls separately. For a buyer who runs a business by phone all day, that last column is a qualification metric no form gives you: a 60-second call is somebody describing their current setup. If you have anyone who answers during the hours you advertise, test it against the form rather than assuming the form wins.

The trial page belongs to the one case that passes the Sales test — a published price, an online signup, no migration in the way — and to nothing else. And there are two nevers. Never the pricing page: an owner who has not yet decided the problem is solvable is being asked to evaluate cost before value. And never a demo booking that is really a sales call in a calendar widget; if the next step is a person, say so in the ad, because an owner-operator's tolerance for a bait-and-switch is zero and they will tell the front desk of every other business in town.

Two rules carry over from the twin unchanged and one inverts. Unchanged: do not make the product screenshot the hero, and let the first line name the reader because the ad set cannot. The Browse page has exactly one dashboard ad in ten and it escapes the rule for a specific reason — it is not showing the product, it is showing one tile at poster size with a browser window arranged around it. That is a stat-hero in disguise, and it is the only defensible way to use a UI on this app.

What inverts is the evidence rule. On Instagram the risk is that a B2B ad announces itself as marketing and gets flicked past, so the twin's advice is to render a situation. Here the reader is older and comparing terms, and the ad is pulled toward proof — and proof is where this pipeline invents. The gym Facebook batch established that. This one shows the vertical-specific version, which is worse: four of the ten must not run, and the failures are a competitor's real product names rendered in the frame, a real electronic logging device's name reused as a module name, an invented case study stamped "ACTUAL CUSTOMER RESULT", and a first-person founder biography under a face that belongs to nobody.

The most useful single finding is structural rather than a warning. The one ad in ten with no fabrication anywhere is the solid-colour quote card, and it has none because the direction has no evidence slot in it — no chart, no screenshot, no receipt, no face. Where the direction offers a place to put proof, the model fills it. That is a lever: if you are not in a position to substantiate a number this week, pick a direction with nowhere to put one.

Three stages, and the shape differs from the twin's in one important way: the middle stage carries more here, because the gap between "this is a real problem" and "I will move my business onto your system" is measured in months rather than in a browsing session. An owner-operator does not evaluate software continuously. They think about it, do nothing, and then move fast when a contract renews or something breaks.

That argues for patience in the retargeting window rather than intensity in the cold one. Keep the cold stage narrow and cheap — two or three arguments about the business, not a rotating catalogue of features — and put the migration answer, the support answer and the proof in front of the people who came back. The customers' stage is smaller than the twin's because expansion revenue in a vertical usually comes from adding trucks or chairs, which is a conversation your own team has rather than an ad.

The Instagram twin makes an argument worth repeating before contradicting it, because it is correct there and it is the best material on that page. Meta's delivery system needs roughly 50 optimisation events per ad set per week to leave the learning phase. For a dental practice, 50 booked implant consultations a week is not a real number. For a self-serve software product, 50 trial starts a week is entirely achievable — so SaaS is the one industry on this site that can buy the optimisation everybody else is priced out of.

Change the reader and it stops being true. Fifty demo requests a week from owners of three-to-fifteen-truck operations is a dentist-scale number, not a software-scale one, and the advantage evaporates — not because the platform changed but because the business did. This is the clearest illustration in the whole cluster that "industry" is the wrong unit of planning: the same word covers two companies with opposite arithmetic. So plan for learning-limited delivery as the normal state rather than as a failure, which means fewer ad sets, longer reading windows, and a real reluctance to split the budget.

Practically: run one ad set with several ads in it rather than several ad sets, because every extra ad set divides the same thin evidence into thinner piles. Keep the optimisation event as shallow as you can defend — a form submission or a call rather than a closed deal — and accept that the deeper event is a reporting number rather than an optimisation target at this volume. Then leave it alone. Pausing an ad set, or changing its optimisation event, audience or creative, is a significant edit and restarts the learning phase; at fifteen or twenty events a week you cannot afford to restart it monthly.

Judge it on payback months rather than cost per acquisition, exactly as the twin says, but with two changes to the arithmetic. Vertical software has better retention and worse volume than product-led software, so the payback window you can tolerate is longer and the number of chances you get is smaller. And there is a seasonality here the twin does not have: a compliance deadline, a licensing renewal, a school or fiscal year, a trade's own quiet season. Spend into the month before the thing your customers plan around, not into the month of it.

The first is the one that produces the "B2B does not work on Meta" conclusion, and the rest are mostly about what happens when a generator is pointed at a vertical.