A Content Engine for IM8: 5,000 Compliant Assets a Month

Consumer supplements are one of the highest-volume, highest-liability content categories there is; the brand has to ship thousands of ads a month to feed paid social, and every one of them carries FTC and FDA exposure the moment it makes a health claim. Volume and compliance pull in opposite directions: the faster you produce, the more places a stray “cures” or an unattributed statistic can slip through. IM8, the longevity-supplement brand from Prenetics (NASDAQ: PRE), runs exactly this problem at global scale: roughly 200,000 servings a day across 43 countries, an order every 27 seconds, and a content operation expected to produce around 5,000 assets a month.

I designed the content operating system that makes that volume possible without turning claims into a liability: the production architecture, the claims-compliance layer, a performance-creative loop, a five-concept creative slate, the team model, and the AI production pipeline underneath it.

The Challenge

Four things shaped every decision:

• The volume is a math problem, not an effort problem. Eight editors is about 168 editor-days a month. Five thousand bespoke assets would demand roughly 30 finished pieces per editor per day, and that is not a real number; a human crafts one or two masters a day. The operation cannot be an editing queue. It has to be a multiplication system, or it ships late or ships slop.
• Every asset is a compliance surface. Supplements live under FTC substantiation and FDA structure-function rules. A single disease claim or an unattributed trial statistic on asset 3,412 is a legal problem, not a typo. Compliance cannot be a review step bolted on at the end; at this volume, a human eyeballing each asset does not scale.
• Creative has to be the growth lever, not a volume mill. Feeding roughly 3,000 live paid ads a month (per IM8’s operating brief) invites the near-duplicate trap: flood lookalike cuts and the audience experiences one ad while the whole cohort fatigues as a unit.
• The team spans twelve time zones. Media buying sits in Hong Kong; the content lead sits in US Eastern. Without an engineered cadence, that gap becomes a bottleneck where every decision waits half a day.

The Approach

Multiply, don’t edit

The engine starts from arithmetic most content plans skip. If a human can only craft one or two masters a day, then 5,000 finished assets cannot be edited into existence; they have to be multiplied into existence. The production pyramid does that in three tiers each month: about 150 net-new masters (new concepts, films, hero cuts), roughly 850 adaptations built on templates with judgment, and about 4,000 variants and localizations from the template and AI pipeline. One master becomes 25 to 35 finished assets; one ambassador shoot window becomes 300 or more. Humans make masters; systems make volume; NOTHING ships unwatched.

~150 Net-new masters human craft

▼

~850 Adaptations templated, pod-lead QC

▼

~4,000 Variants + localizations AI pipeline, 100% claims-linted

▼

~5,000 Finished assets / month 43 markets, nothing ships unwatched

Editor capacity allocation

60%

25%

15%

Performance 60%Brand + Tentpole 25%Ambassador 15%

Run three lanes, each with its own physics

One workflow spine, three speeds, because failure costs are not equal.

• Performance runs a 48-to-72-hour cycle with delegated quality control, because failure there is cheap and fast; it feeds the always-on paid account.
• Brand and Tentpole (TV, out-of-home, launches) runs stage-gated over weeks, with my review and regulatory sign-off, because failure there is public and permanent.
• Ambassador runs on fixed shoot windows with strict approval chains, planned backward from the date, because talent time is the scarcest asset in the company and the window preempts everything.

Editor capacity is allocated deliberately across the three, with a pre-contracted freelance bench and agency overflow absorbing surge so editors never get pulled off masters mid-cut and the always-on account never silently starves.

Treat claims as architecture, not a checkbox

The part I would defend hardest: compliance is built into the system, not appended to it. A locked claims library, built on day one, holds every approved claim, its substantiation reference, its allowed phrasings, and its per-market flags, seeded from IM8’s real proof stack (the 12-week randomized controlled trial, NSF Certified for Sport, 90 clinically-dosed ingredients across 9 organ systems). Copy may only use library claims. Automated linting runs on 100% of assets and flags drift: disease-claim language, unapproved superlatives, missing attribution on a trial statistic, missing disclosure on creator copy. New claim language is a locked door that requires human sign-off and regulatory review before it enters the library.

I built this muscle running performance marketing inside a $300MM+ SEC-regulated wealth-management firm, where endorsements sit under a federal marketing rule and every public asset is examinable. Supplements swap the SEC for the FTC and FDA, but the discipline is identical: claims are inventory, not improvisation. That is also what makes delegation safe, which is how 5,000 assets a month move without one person approving each one.

Make creative the growth lever

Volume without a learning loop is just noise. Ideas enter hypothesis-tagged (audience by angle by hook family by claim by format), test in clean cells that hold one variable constant, and graduate to scale only after a winner repeats across two audiences. Fatigue is managed by rotating concepts, not flooding variants: the account stays fed with different arguments, not 40 haircuts of the same ad. And before briefing more creative, the system asks where the fatigue actually lives. In one regulated-finance program, mid-flight performance kept sagging in ways new creative did not fix; the real problem was the spending pattern, pulsed budget bursts re-triggering learning phases and starving delivery. Shifting to stable pacing with the same creative pool produced a 6x-plus efficiency gain. Half the “creative fatigue” I have seen was the account, not the ad.

Engineer the time zones

The Hong Kong and US Eastern gap is designed as two working seams instead of one dead zone. A daily anchor at 8:00 to 9:30 a.m. Hong Kong time (my evening) sits at the data seam, right after the full prior US day of spend and creative data closes; kill and scale calls, brief approvals, and unblocks made there give Hong Kong a whole workday to execute. A twice-weekly handoff hands creative asks into the opening US production day, so editors build while Hong Kong sleeps. The rule that holds it together: the lead absorbs the off-hours edge, never the team and never the editors, who keep protected no-meeting production cores. Decision memos with written defaults cover anything that cannot wait the twelve hours.

Govern the AI, with hard no-go zones

AI does the leverage work inside the system: hook matrices, variant trees, localization drafts against locked claims, the claims lint on every asset, and taxonomy tagging for the learning loop. Humans keep the calls that cannot be delegated: final claims sign-off, ambassador likeness, master craft on anything monument-scale, and the read on what the data means. Every asset is watched start to finish by a human before it publishes, at any volume.

The Results

The delivered asset was the operating system itself: the architecture, the claims library, a five-concept production-ready creative slate, the team model, and a working AI production pipeline. Here is what it is built to produce.

Layer	Monthly output
Net-new masters (human craft)	~150
Adaptations (templated, QC’d)	~850
Variants and localizations (AI pipeline, 100% linted)	~4,000
Finished monthly output	~5,000
Markets served	43
Claims coverage	100% linted

A few things stand out:

• Volume that is designed, not hoped for. The 5,000-a-month target is reverse-engineered from real editor capacity, so the number is honest before a single asset is briefed. Plans that skip the capacity math ship late or ship slop.
• Compliance that scales with the volume. Linting 100% of assets against a locked library, rather than trusting 5,000 human review passes, is what lets a supplement brand move at paid-social speed without inviting an FTC problem.
• A creative slate, not just a system. Five net-new concepts for the flagship product, from a lo-fi creator engine to a monument-scale out-of-home film to a skeptic-converting proof ad, each carrying a real claims-lint pass, so the architecture is shown working, not just described.
• A team model built on lead time. Plans lock a week ahead, briefs land 48 hours before production, and credit for winning cuts is visible in the performance readout, because connecting craft to revenue is the strongest retention tool there is.

Key Takeaways

1. Volume is a multiplication problem. Five thousand assets a month cannot be edited into existence; about 150 masters become 5,000 finished assets through versioning, templates, and AI, or the plan does not survive contact with a calendar.
2. Claims are inventory, not improvisation. A locked claims library plus automated linting on 100% of assets is what lets a regulated brand produce fast without betting the company on asset 3,412 behaving.
3. Fight fatigue with concepts, not variants. Rotate the argument, budget the near-duplicates per audience, and diagnose whether the fatigue is really the delivery pattern before briefing anything new.
4. Delegation needs guardrails, not trust. Claims lint, published exemplar reels, written kill criteria, and decision memos with defaults are what let volume move without one person becoming the bottleneck.

The category is regulated and the volume is extreme; the discipline is not exotic. Start from the capacity math, make compliance part of the architecture, run creative as a learning loop, and let systems rather than heroics hold quality at scale. That is the same systems-over-heroics approach I bring to organic and AI search visibility now, at a different scale and in a different market, but on the same principles.