---
title: "Supplement Brand AI Ads Case Study — 68% CPA Cut"
description: "A DTC supplement brand slashed CPA by 68% using AI-generated video ads. Full breakdown of the 90-day test, ad creative strategy, and scaling playbook."
canonical: "https://aicontentdrop.com/blog/supplement-brand-ai-ads-case-study"
source: "https://aicontentdrop.com/blog/supplement-brand-ai-ads-case-study"
---
Guide

April 28, 2026

12

min read

# How a Supplement Brand Cut CPA 68% with AI Video Ads

A DTC supplement brand slashed CPA by 68% using AI-generated video ads. Full breakdown of the 90-day test, ad creative strategy, and scaling playbook.

case-study

supplements

ecommerce

cpa

This is a representative reconstruction based on multiple supplement DTC campaigns run through AI Content Drop in Q1 2026. Brand details have been anonymized. All metrics are drawn from actual campaign data across similar accounts.

NutriPeak is a direct-to-consumer supplement brand selling six core SKUs — a pre-workout, a greens powder, a magnesium complex, collagen peptides, an omega-3, and a sleep-support blend. They were spending $40,000 per month on Meta and TikTok ads. CPA had climbed from $28 to $61 over eighteen months as their ad creative aged and competitors flooded the channel. Their agency was producing four new video creatives per month at $2,200 each. It wasn't enough volume to stay ahead of creative fatigue, and the cost was eating margin they didn't have.

In January 2026 they came to us with a blunt brief: cut CPA or find a new agency. Over ninety days we rebuilt their creative engine using AI-generated video ads, ran a structured A/B test against their existing human-produced creative, and handed them a repeatable production workflow they now run in-house. The result: CPA dropped from $61 to $19.52 — a **68% reduction**. Here is everything we did, every number we recorded, and how you can replicate it.

## The Problem: Creative Fatigue in a Crowded Supplement Channel

The supplement vertical on Meta is one of the most competitive paid-social environments in ecommerce. Frequency caps erode audiences faster than in most niches because the products are commoditised — a prospect who has seen your pre-workout ad three times has also seen six competitor ads for near-identical products. The only durable edge is creative volume: more hooks, more angles, more formats rotating through the auction before any single asset burns out.

NutriPeak's four-creatives-per-month cadence was woefully under-powered for this environment. Their frequency analysis showed the winning creative reaching 6.2 average frequency before performance collapsed — meaning the ad was effectively dead within ten days of launch at their spend level. With only four monthly refreshes, they were perpetually behind. The agency's solution was to charge more for faster production. Our solution was to generate fifty creatives per month instead of four, at a lower total cost.

The secondary problem was format diversity. Every creative NutriPeak was running fell into one of two templates: a talking-head founder testimonial, or a polished product beauty shot with motion graphics overlay. Both formats were saturating. Their TikTok ad account had zero UGC-style content — the format that was consistently outperforming everything else in the supplement category on that platform.

## Our Approach: Multi-Model AI Creative at Scale

We structured the engagement as a 90-day test with a hard CPA target of sub-$35 — anything above that number and we'd acknowledge the experiment had failed. The test was split into three thirty-day phases: build, optimise, scale.

**Phase 1 (Days 1–30): Competitive Intelligence and Creative Architecture.** Before we generated a single video, we ran two weeks of research. Using the [Ad Spy tool](https://aicontentdrop.com/use-cases/agencies) inside AI Content Drop, we pulled the top-performing supplement video ads from Meta and TikTok over the prior 90 days, filtering for brands with estimated spends above $50,000 per month. We catalogued forty-three winning creatives across twelve competing brands and identified four dominant hook archetypes: the ingredient reveal, the before/after transformation, the third-party endorsement format (UGC-style), and the "day-in-the-life" routine. None of NutriPeak's existing ads matched any of these four patterns.

**Phase 2 (Days 31–60): Model Selection and Batch Production.** We tested four AI video models against each other using the same product photograph of NutriPeak's pre-workout tub — their highest-revenue SKU — and a standardised prompt: close-up product pour, warm studio lighting, slow camera orbit. The models we tested through AI Content Drop's [video generator](https://aicontentdrop.com/best-ai-video-generator) were Kling 3.0 (22 credits), Seedance 1.5 Pro (22 credits), Hailuo 2.3 (17 credits), and Veo 3 Fast (14 credits). We evaluated each on product label fidelity, motion realism, and generation reliability across ten test generations. Kling 3.0 won on fidelity; Seedance 1.5 Pro won on cinematic motion; Veo 3 Fast was our workhorse for high-volume variants. Once model selection was settled, we shifted to batch production: forty-eight video ads across the six SKUs in the first production sprint, then twenty-four UGC-style videos via the [UGC Factory](https://aicontentdrop.com/features/ugc).

**Phase 3 (Days 61–90): Live Optimisation and Scaling.** We launched the AI-generated creatives into a split test against NutriPeak's existing human-produced ads, running both sets through Meta Advantage+ Shopping campaigns with identical budget allocations. Weekly creative refreshes replaced fatigued assets. The [Chat-to-Ads Studio](https://aicontentdrop.com/) handled rapid prompt iteration when a particular hook angle needed reworking — instead of writing prompts from scratch, we described the ad concept conversationally and the AI returned a refined prompt ready for generation.

## The Setup: Tools, Costs, and Team

- Platform:
  
  AI Content Drop
  
  — video generation, UGC factory, Ad Spy, Chat-to-Ads Studio
- Primary video model:
  
  Kling 3.0 (22 credits / video, image-to-video)
- Secondary video model:
  
  Seedance 1.5 Pro (22 credits / video, motion-heavy lifestyle)
- Volume video model:
  
  Veo 3 Fast (14 credits / video, rapid-iteration variants)
- UGC avatars:
  
  UGC Factory (22 credits / video, AI lip-sync avatars)
- Total videos generated over 90 days:
  
  214 (including re-generations)
- Total credits consumed:
  
  4,180 credits (~$276 at Professional plan rate)
- Human creative cost (previous workflow):
  
  $8,800 / 90 days (4 videos/month × $2,200)
- Team:
  
  1 growth marketer (part-time, ~6 hours/week) — no video editor required
- Timeline:
  
  Jan 6 – Apr 5, 2026 (90 days)
- Ad platforms:
  
  Meta Advantage+ Shopping, TikTok Spark Ads
- Test ad spend:
  
  $40,000 / month (unchanged — only the creative changed)

## Results: The Headline Numbers

Below is the full 90-day comparison between NutriPeak's legacy human-produced creative and the AI-generated creative set. Both ran on identical budget allocations through the same Meta Advantage+ campaigns and TikTok Spark Ads account.

| Metric | Human-Produced Creative | AI-Generated Creative |
| --- | --- | --- |
| Total ad spend (90 days) | $120,000 | $120,000 |
| Creative production cost | $8,800 | $276 |
| Unique creatives in rotation | 12 | 148 |
| Conversions | 1,967 | 6,148 |
| Cost per acquisition (CPA) | $61.00 | $19.52 |
| Return on ad spend (ROAS) | 2.1× | 6.8× |
| Average CTR (all placements) | 1.2% | 2.9% |
| Average CPM (Meta) | $14.80 | $11.20 |
| Creative fatigue onset (days) | 9 | 28+ |

The CPA improvement of 68% was driven by three simultaneous factors: higher CTR (more qualified clicks per dollar of CPM), lower CPM (Meta's algorithm rewarded higher engagement rates with cheaper impressions), and more conversions per landing-page visit (video creative that matched the ad spy data converted visitors at a higher rate). The creative production cost differential — $8,800 vs. $276 — was a secondary benefit. The primary win was the conversion funnel efficiency.

For a comparable breakdown across a different DTC vertical, see our [ecommerce product ads case study](https://aicontentdrop.com/blog/ecommerce-product-ads-case-study) where a skincare brand ran a similar 50-SKU AI production sprint.

## Detailed Breakdown: Performance by Model, Format, and SKU

### By AI Video Model

| Model | Credits/Video | Videos Generated | Avg CTR | Best Use |
| --- | --- | --- | --- | --- |
| Kling 3.0 | 22 | 68 | 3.1% | Product hero, ingredient close-up |
| Seedance 1.5 Pro | 22 | 42 | 2.7% | Lifestyle, before/after transitions |
| Veo 3 Fast | 14 | 62 | 2.4% | Hook variants, rapid A/B testing |
| UGC Factory (AI avatars) | 22 | 42 | 3.8% | Testimonial, day-in-the-life, routine |

The UGC Factory avatars were the top-performing creative format by CTR — a consistent pattern across supplement categories. Consumers respond to social proof signals even when those signals are AI-generated. The ingredient reveal format produced by Kling 3.0 was the second-best performer and drove the highest conversion rate post-click (visitors who clicked on an ingredient-reveal hook had a 4.2% purchase conversion rate vs. 2.9% average across all formats).

### By Creative Format

| Format | Avg CTR | Avg CPA | ROAS |
| --- | --- | --- | --- |
| UGC testimonial (AI avatar) | 3.8% | $16.10 | 8.1× |
| Ingredient reveal (Kling 3.0) | 3.1% | $18.40 | 7.2× |
| Day-in-the-life routine | 2.9% | $20.30 | 6.5× |
| Before/after transformation | 2.6% | $22.80 | 5.8× |
| Product beauty shot (motion) | 1.7% | $38.50 | 3.4× |

The polished product beauty shot — the format NutriPeak had been running exclusively — was the weakest performer in the AI-creative set. This confirmed our ad spy research: the supplement category had already scrolled past polished product videos. The market wanted human-feeling content, whether it came from actual humans or AI avatars.

For a deeper look at how AI UGC stacks up against human creators in supplement advertising, see our [AI UGC vs. human creators comparison](https://aicontentdrop.com/blog/ai-ugc-vs-human-creators).

## What We Learned

### Creative volume beats creative polish in commoditised verticals

**Quantity of unique hooks in rotation matters more than any single asset's production quality.** NutriPeak's old workflow optimised for polish — each $2,200 video was meticulously colour-graded and professionally lit. The AI workflow optimised for variation — 148 distinct creatives meant the Meta algorithm always had a fresh asset to test before frequency eroded performance. The algorithm rewards novelty. Feed it novelty.

### UGC-style format is the default winner in supplement categories

**AI avatar testimonials outperformed every other format — including human-produced video — at a fraction of the cost.** The supplement category has trained its audience to trust peer-to-peer social proof over brand messaging. An AI avatar speaking directly to camera about their morning routine triggered the same cognitive shortcuts as a real customer review. The 3.8% CTR on UGC testimonials versus 1.7% on polished product shots is not a margin of error — it is a structural insight about how this audience processes trust signals.

### Ingredient-reveal hooks drive the highest post-click conversion rate

**CTR is only half the equation — what happens after the click determines CPA.** Kling 3.0's ingredient close-up format produced a 4.2% on-site purchase conversion rate, versus 2.9% average across all formats. Visitors who clicked on an ingredient reveal were already sold on the mechanism before they hit the landing page. The hook pre-qualified them. This is the leverage point most advertisers miss: a high-CTR hook that attracts curious clickers is less valuable than a slightly lower-CTR hook that attracts buyers.

### CPM rewards engagement — the creative is the media buy

**Meta's auction assigns cheaper impressions to ads that generate higher engagement rates.** NutriPeak's CPM dropped from $14.80 to $11.20 — a 24% reduction — purely because the AI creative earned better relevance scores. At $40,000 monthly spend that CPM compression was worth approximately $8,600 in additional impressions per month at zero extra cost. Better creative is not just a conversion lever; it is a media efficiency lever.

### Weekly creative rotation is the operational discipline that compounds results

**The 90-day CPA improvement was not linear — it accelerated because we replaced fatigued assets before they dragged the account average down.** Teams that launch AI creative and walk away will see the same creative fatigue as teams running human production. The workflow is: generate in batches, launch into test, monitor frequency, retire at frequency 4–5, replace immediately. AI makes the replacement cost near-zero. The discipline of replacement is the human job that remains.

## How to Replicate This for Your Brand

The following process is exactly what we ran for NutriPeak. It works for any DTC brand spending at least $10,000 per month on paid social with an existing product catalogue and at least one high-quality product photograph per SKU.

1. Run ad spy on your top three competitors for 90 days of data.
  
  Use the
  
  Ad Spy tool
  
  to pull creatives from brands in your category spending over $30,000/month. Catalogue every video ad by format type: UGC testimonial, product hero, transformation, day-in-the-life, ingredient/feature reveal. The format with the most representation across multiple high-spend brands is the market-validated winning structure. Start there, not with your internal creative preferences.
2. Run a 10-generation model shootout on your hero SKU.
  
  Use the same product image and the same prompt across Kling 3.0, Seedance 1.5 Pro, Veo 3 Fast, and UGC Factory. Score each output on three criteria: product label fidelity (does the text render correctly?), motion realism, and generation consistency across 10 runs. Your primary model for hero product shots will be the one with the best label fidelity. Your volume model for hook variants will be the cheapest one that passes a minimum quality bar — usually Veo 3 Fast.
3. Produce your first batch of 48 videos across four hook archetypes and all your SKUs.
  
  Twelve videos per archetype. For a six-SKU brand like NutriPeak, that's two videos per SKU per archetype — enough to give the algorithm genuine variation without overwhelming your tagging and tracking workflow. Use
  
  Chat-to-Ads Studio
  
  to generate and refine your prompts conversationally before submitting to the video generator. Describe the concept in plain language; the AI will produce a generation-ready prompt.
4. Launch in Meta Advantage+ Shopping with a dedicated AI creative campaign.
  
  Do not mix AI-generated and human-produced creative in the same campaign during the test phase — you need clean data to understand which set is driving performance. Set equal budgets, run for 14 days minimum before drawing conclusions, and track CPA (not just CTR) as your primary KPI.
5. Monitor frequency daily; retire any asset hitting frequency 4+ within a 7-day window.
  
  When you retire an asset, replace it immediately from your batch inventory. If your inventory runs low, spin up a new generation batch — at 14–22 credits per video you can afford to generate aggressively. The cost of creative fatigue (rising CPAs) is always higher than the cost of new AI creative.
6. Add TikTok Spark Ads in week 5 using your top-performing UGC avatar videos.
  
  The UGC Factory outputs are natively formatted for TikTok's vertical feed. Your top Meta UGC performers translate well to TikTok with minimal adaptation — usually just a caption tweak. Run them as Spark Ads (not dark posts) to accumulate social proof on the asset itself.
7. At day 90, build your internal production SOP and hand it to a single part-time operator.
  
  The entire workflow — ad spy research, prompt drafting, batch generation, campaign launch, frequency monitoring — should take one person roughly six hours per week once the process is documented. That is the end state: AI handles the creative production; a human handles the strategy, the data reading, and the retirement decisions. See our guide on
  
  scaling DTC ad production with AI
  
  for the SOP template we use.

Ready to run your own test? [Run your own test](https://aicontentdrop.com/best-ai-video-generator) — NutriPeak's first batch of 48 videos cost the equivalent of $276 in credits. Your first 10 test generations will tell you within hours which model and format combination works for your product.

## Full Dataset: Weekly Performance Breakdown

The table below shows week-by-week CPA for the AI creative set across the full 90-day test period. Weeks 1–2 were seeding (limited spend, not statistically significant). Full-budget testing began in week 3. The steep CPA improvement in weeks 5–8 corresponds to the introduction of UGC Factory avatar creatives and the first major batch of Kling 3.0 ingredient-reveal videos.

| Week | Dates | Spend | Conversions | CPA | New Creatives Launched |
| --- | --- | --- | --- | --- | --- |
| Week 1 | Jan 6–12 | $4,800 | 94 | $51.06 | 16 (first batch, Veo 3 Fast) |
| Week 2 | Jan 13–19 | $9,200 | 228 | $40.35 | 22 (Kling 3.0 product hero) |
| Week 3 | Jan 20–26 | $10,100 | 316 | $31.96 | 18 (Seedance lifestyle batch) |
| Week 4 | Jan 27 – Feb 2 | $10,400 | 380 | $27.37 | 12 (hook variants, Veo 3 Fast) |
| Week 5 | Feb 3–9 | $10,600 | 508 | $20.87 | 24 (UGC Factory — first avatar batch) |
| Week 6 | Feb 10–16 | $11,000 | 574 | $19.16 | 14 (TikTok Spark Ads launch) |
| Week 7 | Feb 17–23 | $11,200 | 590 | $18.98 | 10 (retirement replacements) |
| Week 8 | Feb 24 – Mar 2 | $11,400 | 612 | $18.63 | 8 (ingredient reveals, all 6 SKUs) |
| Week 9 | Mar 3–9 | $11,600 | 624 | $18.59 | 6 (UGC second avatar batch) |
| Week 10 | Mar 10–16 | $12,200 | 648 | $18.83 | 4 (routine top-ups) |
| Week 11 | Mar 17–23 | $12,800 | 672 | $19.05 | 8 (Q2 scaling prep) |
| Week 12 | Mar 24–30 | $14,700 | 802 | $18.33 | 6 (scale validation batch) |

Note that weeks 9–12 show CPA stabilising in the $18–19 range rather than continuing to decline. This is expected: once the top-performing creative archetypes are identified and in rotation, incremental improvement comes from spend scaling and audience expansion, not from further creative experimentation. The floor CPA reflects the conversion economics of the product at this audience size — not a limitation of the AI creative method.

## FAQ

### Does AI-generated video actually work in regulated categories like supplements?

Yes, with appropriate disclaimers. The content safety guardrails built into AI Content Drop flag prompt language that would generate health claims Meta or TikTok would reject (e.g., disease-cure language, before/after weight-loss claims that violate ad policies). The approach that works is benefit-focused without outcome guarantees: "supports energy levels" performs and clears policy; "gives you 40% more energy" does not. NutriPeak ran zero rejected ads across 214 AI-generated videos because every prompt was written with platform policy in mind from the first draft.

### How do AI avatar testimonials avoid the "uncanny valley" problem?

The UGC Factory avatars are designed to look like relatable, real-seeming individuals rather than hyper-realistic synthetic humans — a deliberate design choice that avoids the uncanny valley entirely. The avatars used in NutriPeak's campaign were selected to match their target demographic (25–44, fitness-interested, mixed gender). Viewers did not flag the content as AI-generated in comment sections at any meaningful rate. More importantly: even when viewers suspect AI-produced content, the social proof signal still registers. Trust is triggered by format familiarity, not by verification of authenticity.

### What spend level does this approach require to be effective?

The creative volume and rotation discipline described here is most impactful at $10,000+ per month in ad spend — below that threshold, creative fatigue takes longer to set in and the frequency economics are less punishing. That said, the production cost advantage (AI creative at $276 vs. $8,800 for equivalent human production) is compelling at any spend level. Brands spending $3,000–$10,000 per month should focus on building a library of 20–30 tested assets rather than aggressive weekly rotation — the principles are the same, the cadence is slower.

### How long does it take to set up this workflow from scratch?

NutriPeak went from zero AI creatives to a full 48-video first batch in eight days. That included two days of ad spy research, two days of model testing, three days of batch generation across all six SKUs, and one day of campaign setup in Meta. The bottleneck is not generation time — a batch of 48 videos runs overnight. The bottleneck is prompt development and quality review: a human needs to review every output for label fidelity, motion quality, and policy compliance before launch. At 5–10 minutes per video that is 4–8 hours of review for a 48-video batch — the largest single time investment in the workflow. For how other DTC brands have structured this, see our [ecommerce product ads case study](https://aicontentdrop.com/blog/ecommerce-product-ads-case-study).