Northstead Coffee is a 14-person specialty roaster based in Bristol, UK. They sell direct-to-consumer subscription coffee with an average order value around £42. Twelve months ago they were stuck — paid social was their largest acquisition channel and their creative production couldn't keep up with the testing volume they needed. Here is the detailed story of what changed, with the actual numbers, the mistakes they made, and the practices that stuck. (Published with their permission. Names of competitors are redacted.)

This is the longest customer story we have published. We wanted to write it because Northstead's path is unusually well-documented and because the team made the kind of mistakes that are easy to see in hindsight but genuinely hard to avoid in the moment. If you are running paid social at a similar scale and feeling the same creative-production bottleneck, this should read like a useful prequel to your own quarter.

Where they started: 12 ads per quarter, 6-week production cycles

In the quarter before adopting the platform, Northstead produced exactly twelve ad creatives. The bottleneck was their freelance designer, who fit Northstead in around three other clients and produced two new creatives every two weeks. By the time a new creative shipped, the previous one had been running for six weeks and was deep into fatigue territory.

Their performance numbers in that quarter:

• Customer acquisition cost: £64.20 (target was £55)
• Creative win rate (variants that beat the previous winner): 17% (industry benchmark is around 25%)
• Average days a winning creative ran before fatigue: 21 days
• Number of creative concepts tested: 12
• Total ad spend: £185,000

The strategic problem was not subtle. They needed roughly five times the creative volume to maintain test velocity, and their cost structure could not accommodate hiring a full-time in-house designer. They had been quoted £8,000-12,000 per month for an agency retainer that would have delivered maybe 25-30 creatives per quarter — better than 12, still not the 50-80 they were modeling as the sweet spot.

Month 1: the false start

They onboarded onto Video AD Builder in early Q2 and immediately made the mistake that almost every new team makes: they tried to generate ads with the same creative process they had been using with the freelance designer. Meet on Monday to define a concept. Write a brief by Wednesday. Run a single generation on Thursday. Review on Friday. Ship the following week.

The cycle was faster than the freelance designer (4 days instead of 14) but the output volume barely moved. They produced 18 creatives in month 1 — slightly more than their previous baseline but not the step-change the platform was supposed to enable. Their CAC actually went up slightly because they were spending platform credits on creatives that were not meaningfully different from what they had been running.

Their growth lead's post-month-one summary, which she shared with us: "We are using a Ferrari to drive to the corner shop. We need to rethink how we work, not just where we generate."

Month 2: the workflow rewrite

The breakthrough came from one specific decision: they stopped doing "creative meetings" entirely. The Monday creative meeting was replaced with a 20-minute Friday "hypothesis review" where they wrote three creative hypotheses for the following week and ranked them. No actual creative was reviewed in this meeting. Just hypotheses.

Each hypothesis became a prompt template — they used the five-slot pattern we wrote about elsewhere on this blog. The marketing manager spent Monday morning generating eight to twelve variants per hypothesis (so 24-36 variants per week total) without any committee review. The variants went into a shared Figma board.

The Friday meeting added a second 20-minute slot: rank the week's 24-36 variants, pick the top four, those four ship Monday morning. Anything that wasn't picked got deleted, not archived. They were deliberately avoiding the "we'll use it later" trap that creates inventory bloat.

Month 2 production output: 87 creative variants tested, 12 winners shipped to scale. Up from 18 in month 1 and 12 in the quarter before that. The CAC dropped from £64 to £58 by the end of month 2, mostly because they finally found three creatives that significantly outperformed the previous champions.

Month 3: the brand kit moment

The early winners had a quality problem: they looked great individually but didn't look like Northstead together. The team had been so focused on volume that they hadn't established a brand kit, and the generated outputs ranged from "Pacific Northwest hipster coffee" to "Italian espresso elegance" to "minimalist Scandinavian." All were beautiful. None of them was their actual brand.

Their solution took two weeks of work and was the highest-leverage thing they did all year. They:

1. Shot a single batch of 14 reference photos: six SKU photos (their main bean varieties), two packaging shots, three lifestyle shots in their actual roastery, three texture shots of coffee at various stages.
2. Wrote a 280-word tonal brief covering color palette, lighting style, typography rules, and the "kind of person" who appears in their ads.
3. Defined four recurring "characters" (early-morning commuter, weekend host, focused remote worker, weekend brewer) that all generated lifestyle ads would draw from.
4. Set those four characters and the reference photos as defaults in every generation.

The next month's output looked dramatically more coherent. Same volume — 92 variants, 15 winners shipped — but the campaign as a whole started feeling like one brand voice instead of seven. Their growth lead noticed that returning customers were starting to recognize their ads in the feed and engage with them at higher rates than cold ads.

Month 4-6: the compounding phase

Once the workflow and the brand kit were stable, the platform's compounding effects kicked in. Each month they produced more variants, but more importantly, their prompt library grew. By month 6 they had a database of 87 prompts tagged with the campaign performance they had generated. New prompts borrowed structures and language from previously winning prompts, with a hit rate that climbed from the initial 17% to over 30%.

Vertical video was the unlock for months 5 and 6. They had been running mostly 1:1 banner ads because that was what their freelance designer produced. The platform's per-format generation let them spin up 9:16 versions of every winning concept in two clicks. Their TikTok and Reels placements, which had been a tiny slice of spend, became their highest-ROI channel within two months.

Their CAC by month 6: £39.80. Their creative win rate: 32%. Average creative lifespan before fatigue: 34 days (longer because they were testing meaningfully different variants instead of micro-iterations). Total ad spend was up to £247,000 quarterly — they had reinvested most of their CAC savings into more spend on the now-more-efficient channel.

The full before-and-after

Twelve months in, the comparison looks like this:

The headline numbers tell most of the story but not all of it. Two side effects matter at least as much.

First, the marketing manager who took on the prompt-writing role describes the work as substantially more interesting than her previous role of "brief writer." She is now making creative decisions every week instead of writing briefs that someone else would interpret. Retention of marketers improved noticeably for Northstead during this year — they did not lose anyone on the marketing team, which is unusual at their scale.

Second, the company's product team started using the same platform for landing page hero imagery, product packaging concepts, and email header art. The marginal cost of adding a new image use case was nearly zero, so they kept finding new uses. By month 9, "we can probably just generate it" had become an organizational reflex that touched far more than paid social.

What didn't work

For honesty, here are the things Northstead tried that did not pan out:

Daily creative meetings. They briefly experimented with a daily 15-minute creative standup. It became a status meeting nobody enjoyed and didn't speed up decisions. Killed after three weeks.

Generative ad copy. They tried LLM-generated body copy for the ads and found that the AI-written headlines tested 20-30% worse than human-written headlines on the same visuals. They moved back to the marketing manager writing copy by hand. (Worth noting: this may be a category-specific finding. Coffee is a strongly voice-driven brand. Other categories see different results.)

Influencer-style UGC generation. Generated UGC-style videos performed dramatically worse than actual UGC footage they sourced from real customers. They now generate the polished campaign work and source the UGC content separately.

Letting the AI pick its own thumbnails. Auto-selected thumbnails for videos consistently underperformed manually-selected thumbnails. Spend the extra 30 seconds per video picking the frame yourself.

What this case does and doesn't generalize to

Northstead is a particular kind of company: a brand-driven DTC business with a marketing team that wanted to grow into creative responsibility, a product simple enough to be generated reliably, and a CAC level that could absorb experimentation budget. The playbook will not translate verbatim to a B2B SaaS company with a 9-month sales cycle, or to a brand whose product is uniquely difficult to render (we are looking at you, hot sauce in clear glass bottles).

The pattern that does generalize: the platform was the small part. The big part was the workflow rewrite, the brand-kit investment, and the institutional decision to value creative volume and learning velocity over individual-asset polish. Teams that adopt the platform without the workflow rewrite get modest improvements. Teams that adopt both get the kind of compounding shown above.

If you are at a similar inflection point — paid social is your biggest channel, creative production is a bottleneck, and you are considering an agency retainer to fix it — Northstead's twelve months are a reasonable proof-of-concept that there is a cheaper, more interesting path. We are happy to introduce you to their growth lead if you want to ask her anything we didn't cover.