Five structures cover almost every SaaS video ad worth running: outcome first, before and after, workflow, objection, and migration. Each one opens differently, needs different footage, and suits a different stage of intent, which is why testing across structures beats testing five hooks on the same structure. What does not work in any of them is a feature list, and the reason is structural rather than stylistic.
Below is each structure as a script skeleton, the footage it requires, and how to run a test that tells you which one your market responds to.
The five structures at a glance
| Structure | Opens on | Footage it needs | Best stage | Fails when |
|---|---|---|---|---|
| Outcome first | A quantified result | A customer on camera, or a caption plus product footage | Cold prospecting | The number is rounded or unattributed |
| Before and after | The mess they currently work in | The old process, screen recorded, plus the new one | Cold and mid funnel | The before state is a strawman |
| Workflow | A named task, start to finish | Continuous screen capture, clean account | Mid funnel and retargeting | It becomes a feature tour |
| Objection | The reason they have not bought | A practitioner or founder speaking directly | Retargeting, pricing page visitors | The objection is one nobody actually has |
| Migration | The competitor by name | Screen capture of the import, plus a switcher testimonial | High intent, competitor keywords | You overclaim on how easy switching is |
Outcome first
Opens on the result and works out the rest afterwards. The default structure for cold traffic.
- Beat 1, 0 to 3s: the number. "Month end close went from nine days to three."
- Beat 2, 3 to 8s: who got it, with a title on screen. "Finance lead, 200 person manufacturer."
- Beat 3, 8 to 16s: the one mechanism that produced it. Show it.
- Beat 4, 16 to 22s: the next step, proportionate to a cold viewer. A calculator or a benchmark, not a demo booking.
Footage: a customer interview is ideal. Without one, a screen recording of the mechanism plus a captioned claim works, provided the claim is attributed to a named customer on screen. Get the usage rights in writing before it runs.
Before and after
The strongest structure when the current workaround is visibly awful, which in SaaS it usually is.
- Beat 1, 0 to 3s: the before, shown not described. The spreadsheet with 40 tabs, the group chat, the paper form.
- Beat 2, 3 to 6s: one line naming the cost. "Four hours a week, every week."
- Beat 3, 6 to 15s: the same task in your product, same pace, same task, no cuts that hide steps.
- Beat 4, 15 to 20s: the delta as a caption, then the next step.
Footage: two screen recordings of the same task, and it must genuinely be the same task. A rigged before state is spotted instantly and held against you. If the honest comparison is 40 minutes to 12 rather than four hours to nothing, run the honest one.
Workflow
One task, start to finish, in real time or close to it. This is the structure that converts warm traffic and bores cold traffic.
- Beat 1, 0 to 3s: name the task and the role. "Reconciling driver paperwork, if you run a depot."
- Beat 2, 3 to 20s: the whole task, uncut, narrated as it happens. Cursor visible, real data, real latency.
- Beat 3, 20 to 25s: the time it took, on screen.
- Beat 4: trial or self serve next step.
Footage: continuous screen capture from a clean demo account, recorded at a resolution that survives phone size. Zoom into the region that matters rather than showing the full screen, and never use a full desktop capture: at 1080x1350 on a phone, a whole dashboard is unreadable. A product demo video ad lives or dies on that legibility decision.
Objection
Opens on the reason they have not bought yet, which makes it the best retargeting structure available.
- Beat 1, 0 to 3s: the objection said out loud. "You think migrating this will take a quarter."
- Beat 2, 3 to 10s: concede the part that is true. "It takes about two weeks with our help."
- Beat 3, 10 to 18s: show the evidence. The import screen, the timeline, a customer who did it.
- Beat 4, 18 to 24s: the next step, which for this audience can be a demo.
Footage: a founder, a solutions engineer or a customer speaking straight to camera, plus whatever proves the answer. Conceding the true part is what makes this structure work, and it is the beat teams cut. Pick objections from sales call recordings, not from a brainstorm, and run one objection per ad.
Migration
For high intent viewers already using a competitor. Legally and reputationally the most careful of the five.
- Beat 1, 0 to 3s: name them. "If you are on a legacy scheduling tool and paying per seat."
- Beat 2, 3 to 8s: the one specific difference that matters, not a comparison table read aloud.
- Beat 3, 8 to 16s: the import running, in real footage, with the actual duration on screen.
- Beat 4, 16 to 22s: what happens to their data and contract, then talk to sales.
Footage: screen capture of the import, and ideally a switcher testimonial with a title on screen. Comparative claims need to be true and current, so date them and re-check them. Overclaiming on switching effort produces trials that churn, which is worse than no trial.
Why do feature lists fail in feed?
Because a list transfers the work of prioritising to the viewer, in a muted three second window where they will not do it. Six features with equal weight is six things to evaluate, so the viewer evaluates none of them and scrolls. One claim they are accountable for does the prioritising for them.
A feature list is also written for the person comparing vendors on a spreadsheet, who is at the end of a process, while prospecting ads reach people at the start of one. The comparison stage wants a page, not a feed video.
How to test across structures
Test structure first, hook second. Structures differ by more than a hook does, so structure is where the large effects are, and hooks are where you optimise a structure that already works.
A sequence we use:
- Ship one ad in each of the five structures, same offer, same audience, same budget split. Judge on hook rate and cost per qualified click, not on impressions.
- Give it two weeks and enough spend that each ad clears a few hundred clicks. Our 2026 fatigue benchmark puts a typical active campaign at 8 to 20 live variations, so five is a starting grid rather than a full rotation.
- Take the two best structures and build three hook variants each. That is the second week's work, and it is where most of the gains land.
- Retire on frequency rather than on gut feel. Small B2B audiences accumulate exposures fast: 150 a day at a typical 40 CPM against 20,000 people is roughly 2.6 average frequency by day 14, and higher for the responsive core.
- Keep the losing structures in the file. A structure that failed cold often wins in retargeting, particularly objection and workflow.
Producing five distinct structures at once is the awkward part, which is why teams test five hooks on one structure instead. Genyad's SaaS video ads workflow exists for that step: upload your demo recordings, customer interviews and screen captures once, and each variation is a fresh script, shot selection, voiceover, caption set and export drawn from that library rather than a re-cut of one timeline. Our ad script generator drafts to these five skeletons. One variation is one credit, five are free with no card.
What we do not do: AI avatars or synthetic presenters, importing your product from a URL, rendering from a product feed or CSV, predicted performance scores, or publishing to LinkedIn and Meta for you. You need real footage and you upload the exports yourself.
Frequently asked questions
Which SaaS ad structure should I start with?
Outcome first for cold prospecting, provided you have one real customer number you can attribute. If you do not, start with before and after using an honest screen recording of the current workaround, which needs no customer participation and qualifies the viewer just as sharply.
How many SaaS ad variations do I need live at once?
Eight to twenty for an active campaign, which our benchmark report puts as normal, and more if your audience is under 20,000 people. Spread them across at least two structures rather than twenty hooks on one, because structure differences are larger than hook differences.
Can I use demo footage directly as an ad?
Not unedited. A demo explains in order and an ad opens on the result, so a demo recording is source material rather than a finished ad. Pull the single most visually legible moment, put the outcome in front of it, and keep one capability per ad.
Do migration ads risk trouble with competitors?
They carry real risk if the comparative claims are wrong, out of date or overstated, so keep each claim factual, dated and re-checked, and avoid implying a switch is effortless when it is not. Naming a competitor is legitimate in most markets; misdescribing their product is not.