Win & Scale
IndustriesSaaS & B2B

Meta ads for SaaS and B2B, run by your AI agent.

Find the ads that bring demo requests worth a call, and launch the next one paused. Your agent judges every ad on your lead goal.

7 days free with every feature. Cancel anytime.

Works with

Try asking
  • Which hooks produced demo requests?
  • Compare ops lead and founder ads by angle.
  • Brief three ads for the new feature.
  • Launch the walkthrough test, paused.
An ad from Corvane, a fictional brand, in a phone, with the agent's work around it: Values tagged (One home for projects, Ops lead), a winning creative, a launch built paused and a prompt waiting for the owner to activate it. B2B ad for Corvane, a fictional project software, showing a dashboard card under the headline One home for every project.

What does a SaaS brand need from Meta ads?

B2B buyers are a small slice of Meta's audience, so every demo counts, and a cheap one that never closes is the costliest.

  • “Demo requests from people who will never buy.”

    Win & Scale counts website leads and form leads. Which of them closed is a CRM question, and qualified leads from your CRM are the missing deeper goal.

  • “We say the same thing to everyone.”

    A message that lands with one persona is invisible to another. Persona and angle sit in ad names like Demo_B2B_v5, if they are recorded at all.

  • “Nobody has time to brief the next test.”

    An ad that worked in spring runs through the autumn because rotating it is nobody’s job. The next hook is rarely written before the last one tires.

Judge each ad on the lead it was bought for.

Your agent judges a lead page on website leads and an instant form on form leads, each against the rest of the account. It says plainly that it can’t tell a qualified lead from a curious one.

  • meta_get_evidence_health checks the numbers are fresh, and lists the outcomes Meta cannot see, such as qualified leads.
  • meta_get_history_winners judges on website leads or form leads, not on Meta’s “Results”.
  • meta_get_audience_breakdown shows which ages and placements still produce demo requests.
A live test for Corvane, a fictional brand, day 4 of 7, question frozen, two observations. The testimonial card has 2 of the 10 leads its rule asked for, so the reading is inconclusive and too early. Below it, one proposed learning from a concluded test: a delivery date in the first line beats a discount for returning buyers. It counts only once you approve it.

Say the right thing to the right persona.

Win & Scale stores what each creative is as Values: its angle, persona, format and hook. The ad name stays a hint, so a lesson survives the rename.

  • meta_tag_creatives tags each walkthrough, customer story and founder video with its angle, persona, format and hook.
  • meta_get_cross_analysis compares two of them, such as persona by angle.
  • meta_transcribe_creative reads what each demo clip actually says, line by line.
A Library creative, Corvane's Benefit card (a fictional brand), a static read as an image: its Values are the angle One home for projects, the persona Ops lead, the format Benefit card and the hook One home for every project, with the line the agent read, “One home for every project.”, and each trait marked Winner, In line or Low delivery, inconclusive.

Brief the next test before the last one tires.

Your agent turns the next idea into a test, with its question frozen up front: what changes, what stays, and how it will be judged. Results are added to the record, never rewritten.

  • meta_suggest_next suggests which persona or hook deserves the next test slot.
  • meta_create_test writes down the question, and how it will be judged, before any result comes in.
  • meta_conclude_test waits for enough results, then files the lesson for the next launch.
The Briefs window on its Tests tab. A suggested test for Corvane, a fictional brand: Testimonial card for the Ops lead, set against the benefit card. It says what could mislead it and what would prove it wrong. Committing it calls meta_create_test.

Brief the demo ad. Launch it paused. You turn it on.

Your agent writes briefs from your brand voice and the proof you can claim, then builds the launch with nothing spending. Activating it is a separate call that waits for you.

  • meta_save_brief keeps feature and customer claims to what your brand actually states.
  • meta_launch_ads builds the campaign, ad set and ads paused, and flags copy your strategy doesn’t back.
  • meta_activate spends nothing until you say go. Your agent asks you first.
An agent calls meta_launch_ads for Corvane, a fictional saas brand, and gets back a paused launch: the campaign, the ad set and three ads, all Paused, with Activate waiting for the owner.

What can a SaaS brand ask its AI agent?

Ask the way you would ask a media buyer who knows B2B. Your agent checks the evidence before it answers.

  • Evidence checkWhich numbers from the last quarter shouldn’t I trust yet? Don’t change anything.
  • Which hooksWhich hooks produced demo requests over the last 180 days, judged on website leads?
  • Compare personasCompare persona by angle across the last 180 days. What have we never tried?
  • Launch pausedAnswered in the conversation: Launch the ops-lead walkthrough test. Leave it paused for me.
  • Build a formDraft an instant form for demo requests that asks for name, work email and company.
  • Brief the featureBrief three ads for the new feature from what produced leads, in our brand voice.
Claude Code · winandscale connected

Launch the ops-lead walkthrough test: three cuts, one ad set. Leave it paused.

  1. meta_get_evidence_healthsynced 2h ago
  2. meta_get_history_winnersQuestion hook · Winner, 90d
  3. meta_launch_adsPaused
  4. meta_activateWaiting for you

Built 1 campaign, 1 ad set and 3 walkthrough ads, all paused and linked to the ops-lead test. I judge them on website leads, not on which leads close.

What does your agent tag on every SaaS ad?

Every ad is tagged with Values: its angle, persona, format and hook. Results compare even when ad names don't.

ValuesYour account’s Values come from your own creatives.

meta_list_taxonomy

  • Angles

    Why someone should book now

    • Time saved
    • Fewer tools
    • Faster onboarding
    • Security
    • Switching story

  • Personas

    Who the ad speaks to

    • Ops lead
    • Founder
    • Finance lead
    • IT admin

  • Formats

    What the creative looks like

    • Walkthrough
    • Customer story
    • Founder talking head
    • Product clip
    • Static comparison

  • Hooks

    The first three seconds

    • Question
    • Manual-work callout
    • Feature reveal
    • Customer quote

Where to go next

Questions, answered

Everything else is in the docs.

Can an AI agent run Meta ads for a SaaS brand?
Yes, an AI agent such as Claude Code or ChatGPT can run a SaaS brand’s Meta ads with Win & Scale. Your agent reads the ad account, tags each creative and builds the next launch paused.
Which goal is a SaaS ad judged on?
Win & Scale judges a demo-request page on website leads and an instant form on form leads. The goal is the event the ad set was bought for, never Meta’s “Results” column.
Does Win & Scale know which demo requests become customers?
No, Win & Scale has no CRM pipeline attribution. Qualified leads from your CRM are the missing deeper goal, and the evidence check lists them among the outcomes Meta cannot see.
Will my agent launch ads without asking?
No, every launch is created paused. Turning one on is a separate call, meta_activate, and your agent asks you first.
How much does Win & Scale cost for a SaaS brand?
Win & Scale starts with a 7-day free trial that includes every feature. After it, each ad account pays a flat monthly price by ad spend, from $49 a month for up to $10k.
Which AI agents work with Win & Scale?
Any AI agent that connects over MCP works with Win & Scale, including Claude Code, Claude, ChatGPT, Cursor and Codex. Agent calls cost 0 AI credits, because they run on your own plan.

Find the ads that bring demos. Launch the next one paused.

Start a 7-day free trial with your website, connect your AI agent, and keep every launch paused until you say go.

Works with