AI marketing in practice: what it actually does on a Tuesday
Every webinar promises the same thing: AI will run your marketing for you. It won’t. What it actually does is thirty small time savings a week that only add up if your data and your judgement are already decent. That gap between the pitch and the desk is where most businesses get burned.
Lilach Bullock, an AI implementation consultant with over two decades of marketing experience, laid out what this actually looks like on a real account. The breakdown below is drawn from her working notes, not from a vendor slide deck.
️ What a real morning looks like
The example comes from a small skincare brand based in Leeds: £900,000 annual turnover, team of four. Here is the actual schedule from a recent Tuesday:
- 8:15am: Last week’s Meta ad data goes into ChatGPT with a prompt asking it to flag the three underperforming ad sets and explain why, using cost per click and hook rate. Takes four minutes. A human doing this by eye takes closer to forty.
- 8:30am: Klaviyo’s predictive analytics flags 340 customers as likely to churn in 60 days, based on purchase gaps and open rate decay. That segment gets a different email flow, not the generic newsletter.
- 9:00am: Twelve subject line variants go into Claude for an abandoned cart sequence. Three survive after a human filters out the ones that sound like a corporate LinkedIn post pretending to be friendly.
- 9:30am: Google’s Performance Max is already bidding and swapping creative automatically. No input needed unless you want to exclude a placement.
- 10:00am: A landing page brief goes into a drafting tool. About 60 percent gets rewritten. The bones are useful. The voice never is, not yet.
Prediction, drafting, sorting, and testing at speed, with a person still making the calls that matter. That’s it.
The six jobs AI is actually doing in marketing
Strip away the buzzwords and it comes down to a short list. If a tool or agency can’t tell you which of these it’s doing for you, that’s a signal worth paying attention to.
- Prediction: Who’s likely to buy, churn, or click, based on patterns in your data. This is where Klaviyo, HubSpot’s predictive lead scoring, and Meta’s Advantage+ audiences earn their keep.
- Generation: First drafts of ad copy, emails, blog outlines, image variants. Useful for volume and speed, not for finished work.
- Personalisation at scale: Showing different creative or copy to different segments automatically, something that used to need a team manually building fifteen versions of a campaign.
- Summarising: Turning 400 customer reviews or a call transcript into three bullet points of what people actually want. Saves real hours weekly.
- Bid and budget optimisation: Platforms shifting spend toward what’s working in near real time. One of the strongest use cases because it’s a pure numbers problem, not a creative one.
- Sorting and triage: Which leads get a human call first, which support tickets are urgent, which comments need a reply now.
None of that is magic. It’s pattern-matching at a speed and scale a person can’t hit.
The case study that changes how you sell this
Two years ago, a mid-size fitness brand had its cost per acquisition stuck around £42 for months. Moving a chunk of the budget into Meta’s Advantage+ audience targeting, rather than hand-built lookalikes a media buyer had maintained for a year, dropped CPA to £29 within three weeks.
The uncomfortable part of the story is what happened next. The media buyer spent two months anxious he’d made himself redundant. He hadn’t. His week shifted from building and tweaking audiences to reading the results and deciding where the actual budget changes should happen. He got more useful, not less needed.
Nobody wants to hear that AI marketing often means the same headcount doing a more senior version of their old job. But that’s what the data shows, repeatedly, on real accounts.
⚠️ What it does not mean, whatever the pitch deck says
It does not mean set-and-forget. Performance Max and Advantage+ still need a human deciding on offer, audience exclusions, and creative direction, or they’ll spend your budget efficiently against the wrong goal. One campaign reportedly performed brilliantly by every metric Meta surfaces while driving customers who returned everything within thirty days, because nobody told the algorithm that returns mattered.
It does not mean your content gets better by default. A generated blog post reads fine and ranks for nothing, because search engines and readers can both tell when there’s no actual point of view behind it. An AI draft is a fine starting skeleton and a terrible finished product.
It does not mean cheap. The tools run £20 to £200 a month for most small business needs. The real cost is the time to set them up, train them on your data and tone, and audit what they’re doing wrong. That’s the part almost nobody selling AI tools mentions, because it’s not a great sales line.
The uncomfortable truth about mediocre strategy
AI marketing makes a mediocre marketer’s output look tidier and faster. It does not make a mediocre marketer good. If your targeting logic was wrong, your offer was weak, or your emails had nothing worth saying, AI will produce that same weak strategy at ten times the speed and with better grammar.
Businesses that have gotten worse after going all in on AI tools are usually optimising toward a goal nobody set correctly, and the tool did that badly-set job with total confidence.
The businesses winning with this aren’t the ones with the most tools. They’re the ones who already understood their customer, their margins, and their offer, and then used AI to do that understanding faster.
Strategy first, tools second, in that order, every time. Or you’re just automating a mistake.
️ How to start without wasting two months
- Pick one repeated task that eats hours. Subject line writing, ad reporting, first drafts of client updates. Not “marketing” broadly. One specific task.
- Feed it your real data. Past campaigns, actual customer language from reviews or support tickets. Not generic prompts. This is where quality lives.
- Set a two-week time budget to test it before deciding if it works.
- Measure the actual metric, not “it feels faster.” CPA, open rate, hours saved, conversion rate. Write the before number down first.
- Keep a human on every output that touches a customer before it goes out. Every time that step gets skipped, there’s regret within a week.
When This Works
This approach works when the underlying strategy is already solid, the customer data is clean, and the person running it can tell the difference between an AI output that’s good enough and one that’s quietly wrong. It also works well for businesses with no marketing team at all: automated reminders, review requests, and first-line chatbot answers all sit cleanly in AI’s lane.
When It Does Not
It does not work when tools get turned on before fixing the strategy underneath them. It does not work when nobody audits what the algorithm is optimising toward. And it does not work when the business assumes that faster production means lower value: if your judgement is now the scarce part, your rate should reflect that, not drop because the drafting got quicker.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
