A 22-staff B2B SaaS founder I spoke with last month is running her own marketing because the head-of-marketing role keeps slipping out of the budget. Last quarter she shipped four blog posts, one of which took her eleven hours start to finish. Eight email campaigns averaged a 17 percent open rate. Paid Google spend ran at £1,800 a month. There is no formal SEO programme. Her tooling budget is £400 a month covering Mailchimp, Buffer and a ChatGPT Plus seat she pays for on her personal card.
A competitor’s CMO has just posted on LinkedIn about hitting 71 percent visibility growth in eight weeks via AEO. Her chair has forwarded the post and asked why she isn’t doing the same.
She isn’t deciding whether AI belongs in marketing. She’s deciding which slice of the marketing motion to give it first, what realistic ninety-day numbers look like at her scale, and what the ASA’s position on AI-generated content means for the next campaign she ships. That’s the right shape of the question, and it’s what this post is about.
What jobs does AI actually do well in marketing today?
Six jobs are deployable now with quantified evidence. Content generation at scale (around four in five marketers use AI for copy, blog production compresses from nine to thirteen hours per article down to seventy-five to ninety minutes). Email personalisation. SEO and AI-engine optimisation. Display ad creative variation. Paid-search bid optimisation. Social scheduling. Each has a working price point under £150 a month.
Email personalisation is the single most quantified category. Klaviyo’s 2025 benchmark shows the top 10 percent of campaigns hitting 5x the order rate and 7x the repeat purchase rate of the average sender, with multiple cross-vendor case studies reporting a 41 percent revenue increase versus non-personalised sends. eBay deployed Phrasee for AI-generated subject lines and recorded a 16 percent open-rate uplift, with the AI variations beating human copywriter output in A/B tests.
For ad creative, AdCreative.ai’s Creative Scoring predicts performance with over 90 percent accuracy on its own benchmarks, and the UK agency Favoured used HeyGen to push UGC variant production from ten or fifteen pieces a month to fifty or a hundred. AEO matters because AI search traffic is showing 3.2x conversion versus traditional organic, and reasonable content investments have produced 71 percent visibility growth inside eight weeks.
Where are the leaders actually using it?
Look at where the named precedents live, not where the brochure says they live. Jasper’s 2026 survey of 1,400 marketers shows 91 percent now using AI at work, with 51 percent generating multi-asset campaigns from a single brief. One Jasper customer case replaced a £1,500 to £3,000 a month freelancer at £49 to £129 a month plus five to ten hours of internal editing.
The UK-specific evidence is solidifying. Favoured (Hampshire-area digital ad agency) is integrating HeyGen avatars with ElevenLabs voiceover, Ideogram product imagery and Kling b-roll into a modular ad-production pipeline. Lidl ran the largest documented retail campaign of 2025, generating 1.7 million AI visuals in three weeks via Monks.Flow plus Performance Max while keeping brand consistency through automated checks.
For SME-tier tooling, the working stack is mostly UK-priced. Klaviyo Essentials starts at £13 a month, HubSpot Starter £13 to £29, Surfer SEO £89, Canva Pro £119.99 a year, Buffer £5 per channel a month, Jasper £49 to £129. Syrvi AI is the named UK specialist agency operating at £500 to £2,000 a month, with case evidence showing 240 percent lead increases inside ninety days.
Where does AI fall short in marketing?
Seven boundaries matter. Generic copy when prompts are thin (raw output still needs forty-five to sixty minutes of editing per article without rich brand briefing). Brand-voice drift across team members (firms with documented voice guidelines hit 80 to 90 percent consistency, those without hit 40 to 50 percent). Off-brand or hallucinated imagery, including AI product photography landing in the uncanny valley.
Hallucinated factual claims are the biggest one. AI confidently asserts unsupported product benefits, and the ASA applies the CAP code regardless of how content was created. The ASA’s Active Ad Monitoring system has been scaled up to identify non-compliance proactively, and AI-generated synthetic testimonials carry heightened risk under existing guidance on misleading endorsements.
Then the regulatory layer. The ICO’s UK GDPR guidance on AI is explicit on Article 22: solely automated decisions with legal or similarly significant effects need transparency, human review and a DPIA. PECR governs cookies and email consent, and active opt-in is required (pre-checked boxes don’t count). The EU AI Act reaches UK firms whose AI output is used inside the EU, with generative-AI transparency obligations effective August 2026 and penalties up to €35m or 7 percent of global turnover for prohibited practices.
What does a 90-day starter rollout actually look like?
Three phases. Weeks one to four (£100 to £300 a month plus forty to sixty hours): an AI readiness workshop mapping content, email, paid and social workflows; a governance framework covering data access, brand-voice rules, approval flow and a regulated-claims compliance step; deploying ChatGPT Plus or a Jasper trial; activating email-platform AI features. Phase one’s success metric is governance plus one documented pilot, not performance lift.
Weeks five to eight (£150 to £400 a month plus sixty to eighty hours) scales content and email. Weekly blog production via Jasper, Writesonic (£36) or Copy.ai (£36); two or three segmented email campaigns with AI variations; social scheduling automation via Buffer or SocialBee; baseline metrics collected. Realistic targets: four to eight articles a month, 20 to 30 percent open-rate lift on segmented sends, 30 to 40 percent time saved on content production.
Weeks nine to twelve (£150 to £500 a month plus eighty to one hundred hours) bring CRM integration (HubSpot Starter), platform-native paid-media AI (Google Ads Smart Bidding, Meta Advantage+), and predictive segmentation. Total ninety-day software cost £225 to £655 a month ongoing plus £1,000 to £3,000 setup. For a 22-staff services SME, the annual benefit lands at £50,000 to £75,000 against £5,600 to £8,000 quarterly cost, with payback in three to six months.
What should you ask before you commit?
Five procurement questions cover the territory. For email AI, does the platform document send-time and segment decisions, retain consent records, and handle PECR opt-in and opt-out properly? For ad AI, does the system substantiate factual claims pre-publish, flag synthetic testimonials and produce an ASA-defensible audit trail? Active Ad Monitoring catches unsupported claims regardless of who or what wrote them.
For audience-segmentation AI, does the segmentation logic risk triggering Article 22? If the segments influence eligibility for a financial product, employment or essential services, you need transparency, human review and a DPIA. For EU customer exposure, post-August 2026 the generative-AI transparency rules apply: does the platform disclose AI involvement, document training data and pass conformity assessment for high-risk uses?
The fifth question is brand voice. Does the platform support documented voice guidelines, banned-phrase lists, and an editor-review workflow before publish? That’s the difference between AI being a multiplier on a tight marketing team and AI being the reason the brand reads off-key by week three. The deeper procurement checklist sits in the AI vendor due diligence guide, and the trust-erosion question worth holding alongside it is in AI in client communication. If you’d like to sense-check your own slice of this against where you actually are, book a conversation.
