Every ‘AI in marketing examples’ article catalogues the same six brands. Coca-Cola repurposing ads with generative AI. Sephora’s chatbot. Spotify’s Wrapped. Heinz’s text-to-image campaign. Netflix’s recommendation engine. Starbucks personalisation. Inspiring if you run global advertising at a Fortune 100. Useless if you run marketing at a 50-person B2B SaaS scale-up and your CEO has just asked what you’ll ship this quarter.
So here are ten AI in marketing examples that fit that team. Ten workflows pulled from B2B SaaS marketing functions running them right now, with the prompts to copy, the outputs to expect, and the change against the manual version each one produces.
These workflows come from the marketing managers inside our community. Every one has been run by at least one B2B SaaS team in the last quarter. None require a technical background to get started.
An AI in marketing example, the way we use the term, has to clear three bars.
Generative AI in marketing examples that get cited in the press, the kind you read about in the FT, mostly fail the third test. Heinz running text-to-image prompts to generate ketchup visuals is genuinely interesting, but a B2B SaaS marketing manager with a £3,000 monthly budget and a pipeline forecast to hit cannot copy it.
The ten workflows below were picked specifically to clear all three.
The examples aren’t ranked. They’re grouped roughly by the work they do: content production first, then pipeline intelligence, customer insight, personalisation, forecasting, and internal admin last.
If you want the full framework behind these, including the six jobs AI is doing inside a B2B SaaS marketing function, our complete guide to AI in marketing is worth a read.
Generating a landing page variant for each persona, sector, and use case you sell to.
The before. You sell into multiple verticals. Probably three or four roles per vertical. Maybe two or three job-to-be-done framings on top of that. Multiply those together and you have somewhere between thirty and sixty plausible landing page variants. The team ships three or four. Everyone else lands on a generic page that mentions none of them.
The prompt:
Generate three landing page variants for [persona] in [sector], focused on [job-to-be-done].
For each:
Headline (≤10 words) naming their specific problem
A 2-sentence value prop in their sector's language
One proof element relevant to that sector
Primary CTA (3-5 words)
No superlatives, no clichés. 250 words total.
What comes out. Three variants per run, each leading with a sector-specific problem rather than a category pitch. ‘Audit prep that doesn’t eat your Q4. SOC 2, ISO 27001 and evidence collection in one workspace, built for compliance teams in regulated fintechs. See how it works.’
The after. Variants on key segments routinely outperform generic pages by a quarter to two-fifths in conversion. The volume gain matters more though. Shipping thirty variants in a day rather than three in a quarter is the real unlock.
Where to start. Pick the paid channel with the highest spend and the worst current conversion rate. Build one variant for that segment. Measure for a fortnight before scaling.
Turning a thirty-minute conversation with an internal expert into a 1,200-word first draft.
The before. Customer success knows things your content team doesn’t. Product knows more. The engineers know what buyers ask in every demo. None of that knowledge ever reaches the blog, because writing it up manually swallows most of a week and the SMEs don’t have drafting time.
The prompt:
Source: 30-minute interview transcript with [SME role] on [topic].
Output: 1,200-word first-draft article for B2B SaaS readers.
Open with the SME's strongest specific claim, in their own words
Three H2 subheads framed as questions
Quote the SME directly for the five most distinctive lines
Close with one practical takeaway
Invent nothing the transcript doesn't support
What comes out. A clean first draft with the SME’s actual phrasing preserved through the body. Three question-framed sections, a closer that reads as advice rather than as summary, and a writer who’s editing rather than starting from a blank page.
The after. The output multiplier B2B SaaS teams typically report sits between three and five times per quarter. The harder shift is cultural: the bottleneck moves from ‘we never get round to writing it up’ to ‘we can’t book enough interview slots’. ChatGPT vs Claude for B2B marketing covers which assistant handles longer transcripts better.
Where to start. Pick the SME with the most unwritten knowledge in their head, not the one with the most diary space. The interview is only worth what the SME knows.
One long-form piece converted into the social posts, short-form video scripts, email snippets, and sales enablement copy that distribute it.
The before. You spend two days on a pillar piece. You plan to spend another day distributing it. Day three lands, the next project starts, and the article gets two LinkedIn posts and a Tuesday newsletter mention. The other eight or ten distribution assets never make it out the door.
The prompt:
Repurpose this piece for B2B SaaS distribution:
5 LinkedIn posts (≤250 words), each leading with a different insight, each opening with a scroll-stopper
3 nurture email snippets (≤100 words each)
1 sales one-pager: 3 claims, buyer-language phrasing, 2 objection responses
1 60-second video script with on-screen text cues
Each piece stands alone (no references to the original article). Voice: [paste 2-3 voice rules].
Source: [paste]
What comes out. Ten or more distribution-ready assets per run, each with its own opening, angle, and length discipline. The LinkedIn posts read as if a human had read the article and written them, because that’s effectively what happened.
The after. The distribution layer compresses from roughly five hours to under thirty minutes. Team output multiplies five-to-six-fold against the pre-AI baseline, without adding anyone.
Where to start. Take the last pillar article you published. Run it through the prompt this afternoon. You’ll recover ten posts you could have shipped last month.
Auto-built research dossiers for ABM target lists, ready overnight.
The before. ABM works when the research has been done. The economics rarely allow it. A list of two hundred accounts, with two or three analyst-days each, produces nearly a year of research before the first email goes out. Most teams compromise: the top thirty get briefed properly, the rest stay aspirational.
The prompt:
Build an ABM dossier for [target company, URL]:
1. Snapshot: sector, size, revenue, recent funding or leadership moves (90 days)
2. Tech stack relevant to [our category]
3. Three in-market signals (hiring / public statements / launches), most likely first
4. Buying committee + likely decision owner
5. Three outreach angles, each tied to one of the signals above
6. Risk read: anything saying don't bother
≤500 words. Skip filler. Lead with the strongest signal.
What comes out. A structured dossier per account. Outreach angles tied to specific signals rather than generic personas. The risk read does as much work as the positives, because it stops the SDR from chasing accounts the data says won’t close.
The after. Target lists shift from aspirational logos to qualified accounts, each with a signal, a likely committee, and an opening angle. The internal ABM conversation stops being abstract. HubSpot Breeze vs standalone AI tools covers whether to run this inside the CRM or alongside it.
Where to start. Run it on the top ten accounts before scaling. The output shape will tell you whether the prompt needs adjusting for your category.
Last quarter’s sales calls fed through an AI assistant to extract recurring objections and the words buyers actually use.
The before. Sales has theories about why deals slip. Some are right. Some are convenient. Marketing reads the loss summary, agrees with the obvious diagnosis, adjusts a slide. Without the call data, none of this is verified. It’s two functions agreeing on a story.
The prompt:
Analyse [N] sales call transcripts (mixed: won, lost, no-decision).
Extract:
The 3 most-frequent objections, with occurrence counts
The 1 objection most-correlated with lost outcomes
5-10 verbatim phrases buyers used to describe the problem
The demo moment where engagement most often dropped
The proof point most often associated with won deals
Output: one-page memo for content and sales enablement. Flag pattern vs anecdote. Quote buyers directly.
Transcripts: [paste]
What comes out. Verbatim buyer phrasing with call counts attached. ‘The implementation will eat my Q4’ appearing across thirty calls, correlating heavily with no-decision outcomes. The memo becomes the source document for content briefs and homepage copy.
The after. Conversion improvements compound across the homepage, demo follow-up, and pricing page. The mechanism is dull: messages built from real buyer language outperform messages built from internal positioning, consistently, across every team we’ve watched run this workflow.
Where to start. Begin with the last twenty calls. Attempting a full year of calls on the first run produces synthesis that’s too abstract to act on.
Hero, proof, and CTA on landing pages reshaped to the visitor’s sector, company size, and source.
The before. A logistics ops manager clicks a paid search ad you’re paying £18 a click for. They land on a homepage that mentions ‘forward-thinking teams’ and shows three customer logos from sectors that aren’t theirs. They bounce. The click money is gone. Your CAC line ticks up by a fraction.
The prompt:
A visitor matched [sector] / [size band] / [traffic source] / [keyword] arrived on the homepage.
Generate three variants:
Hero (≤12 words) naming their sector explicitly
Subhead (≤25 words) naming the job-to-be-done their sector cares about
CTA (3-5 words)
Skip company size mentions unless the visitor is enterprise (1,000+).
What comes out. ‘Compliance reporting built for regulated fintechs. SOC 2, ISO 27001, and audit prep in one workspace, used by 47 UK fintech compliance teams. See how it works.’ Three variants per run, all sector-grounded.
The after. Conversion lifts of a tenth to a quarter against generic pages are typical. The bigger compounding gain shows up on CAC, because paid traffic gets cheaper when the landing experience matches the ad.
Where to start. Personalise the highest-traffic paid landing page first. Variant testing across five low-traffic pages is statistically useless.
Predictive forecasts that translate the current quarter’s funnel into a band of plausible numbers for the next.
The before. The finance director wants Q3 numbers. The standard marketing answer is a single-point estimate calculated in five minutes, weighted heavily toward optimism, with no acknowledgement of which signals point the other way. The forecast survives until week three of the quarter, when reality catches up with it.
The prompt:
Funnel data, last 12 months: MQL by month, MQL→SQL, SQL→opp, opp→won, avg deal size.
Forecast next quarter's closed-won. Output three scenarios:
P25: recent declining trends continue
P50: rolling 6-month conversion averages hold
P75: current quarter momentum holds
For each: name the 2-3 assumptions that must be true. Flag any anomalous data point that could skew the result.
Data: [paste]
What comes out. A P25/P50/P75 forecast with explicit assumptions for each scenario, plus flagged anomalies in the underlying data. Board-ready output rather than a five-minute extrapolation.
The after. Board pushback drops. The questions shift from ‘are these numbers reliable?’ to ‘which assumption are we testing?’ Forecast accuracy against actuals improves measurably within two quarters of running this workflow consistently.
Where to start. Build the manual version once first. You need to know what good output looks like, otherwise you can’t catch a confidently wrong AI forecast at the point where it matters.
Daily AI monitoring that flags meaningful funnel deviations within 48 hours rather than within the monthly review.
The before. Trial-to-paid drops twenty percent on a Tuesday. Nobody notices until the monthly metrics review, three weeks later. By then it’s six weeks of revenue gone, the cause is buried under three weeks of subsequent activity, and the conversation is about damage control rather than root cause.
The prompt:
Today's funnel snapshot vs trailing 30-day average: [paste].
Flag any metric moving:
±20% vs the 30-day baseline
±10% for 3 consecutive days
Clear pattern break
Per flag: metric, size of move, 3 plausible causes ranked, the first check the team should run, ping-now or wait?
Output: a single Slack message, ≤200 words.
What comes out. A Slack message naming the metric, the move, the three most likely causes, and the first investigation step. Same-day visibility on a deviation that would otherwise stay buried.
The after. Detection lag drops from weeks to days. The board phrasing shifts away from ‘why didn’t we catch this earlier?’ The team spends less time forensically reconstructing what happened and more time actually intervening.
Where to start. Pick three pipeline-critical metrics first (typically MQL volume, MQL-to-SQL, and demo-to-opportunity). Trying to monitor everything produces noise the team ignores within a fortnight.
A weekly Monday morning update generated from CRM, GA4, LinkedIn Ads, and email data.
The before. Monday morning: data from four systems, compared against last week, rewritten as narrative, reformatted for the slide deck. Two hours minimum. The marketing manager arrives at 10am to a half-empty calendar and leaves the desk at noon with one task done.
The prompt:
This week's pull: [CRM / GA4 / LinkedIn Ads / email] + last week's update for comparison.
Draft a Monday leadership update:
1. Headline: what mattered most this week (1 sentence)
2. Numbers: 3-5 metrics, week-on-week change
3. Narrative: 100 words on what's moving and why
4. Watch-list: 1-2 items for next week
5. Focus: 2-3 priorities
≤300 words. Confident. Don't invent numbers.
Data: [paste]
Last week's update: [paste]
What comes out. A 300-word update in the same structure every Monday. Clear headline, matching numbers, a narrative that names what’s moving rather than burying it under bullets.
The after. Production compresses to roughly fifteen minutes of editing on top of a generated draft. The Monday morning meeting starts on time. Leadership reads the narrative because it reads as narrative.
Where to start. Get the data pull working before automating the narrative. The hard part of this workflow is the data integration, not the writing.
A weekly AI agent that scans competitor sites, pricing changes, changelogs, and LinkedIn activity, then produces a one-page digest.
The before. Competitive intelligence usually lives in a Notion page someone last updated in Q1. Then the price-review meeting hits and three people spend a Wednesday scrambling through competitor pages, screenshotting changes, reconstructing a timeline from memory. By Thursday someone has built a slide. By Friday it’s already out of date.
The prompt:
Detected changes from competitors [A, B, C] this week: [website / pricing / changelog / LinkedIn].
Produce a one-page weekly digest:
1. Headline change of the week (one paragraph)
2. Pricing moves + implications for our pricing
3. Feature launches + the customer problem each addresses
4. Messaging shifts: louder, quieter
5. Three team actions: product marketing / content / sales enablement
≤400 words. Cosmetic changes get a one-liner, not a section.
Changes: [paste]
What comes out. A one-page digest published every Friday afternoon. Pricing changes flagged with the implication called out. Three concrete actions distributed to the right team owner. The price-review meeting becomes a half-hour conversation rather than a Wednesday-Thursday rebuild.
The after. Competitive moves surface inside the week they happen rather than the month. Product marketing learns about a pricing change from the digest rather than from a prospect. Sales enablement has the language ready before the first deal where the change matters.
Where to start. Limit to three competitors initially. More produces noise, not signal. You’ll see diminishing returns past five.
Three traits show up across every workflow above, and they’re what distinguish AI in marketing examples that work from ones that quietly die after six weeks.
They’re narrow. Each one solves a single task, not ‘AI for marketing’. The output is small enough to judge quality at a glance. Teams that try to ‘deploy AI across the marketing function’ without picking one workflow to start with almost always stall right here.
They’re measured. Every workflow has a clear before-state and after-state, in numbers. Time saved, output multiplied, conversion lifted. Projects that never define the success metric drift, because nothing tells the team whether to keep iterating or pull the plug.
They’re iterated. The first output of any of these prompts is roughly 70% useful. The team running the workflow refines the prompt across two or three iterations until the output is 95% useful. Teams that publish the first version unedited, decide AI ‘doesn’t work for us’, and stop, usually quit at exactly this point.
The pattern lines up with the six-job framework in our complete guide to AI in marketing. It also lines up with the limits we wrote about in The Bit AI Can’t Do: AI is a remix engine that compounds your existing direction. It doesn’t generate the direction itself.
Q: What is an example of AI in marketing?
Q: What’s the best AI marketing example for a small B2B SaaS team?
Q: Can I copy these AI workflows for free?
Q: Do these AI marketing examples actually save time, or is that just the claim?
Q: How do these examples link to AI in marketing more broadly?
If you want the framework behind these ten (the six jobs, the 16 use cases, the 30-day starter plan), our complete guide to AI in marketing for B2B SaaS is the one you need.
If you want the workflows the community is running this quarter, the prompt libraries iterated weekly, and the monthly live Q&A with founder Danny Asling on what’s working, join us at community.saastrix.uk. The Claude AI for Marketers sprint is the natural entry point for teams starting from these workflows.
.svg){kind=icon}