From ROI 0.5 to 1.4: How a Battery Brand Scaled Ad Returns with AI Video

Ad budget is flat. The bid strategy hasn't changed. But Qianchuan video ad ROI hasn't moved in months.

This is the situation the performance marketing lead at a battery brand was facing before bringing in an AI-powered creative production workflow. Multiple core SKUs running simultaneously across Douyin, JD.com, and Taobao — yet the content team could only produce around 100 videos per week, nowhere near enough to maintain the testing velocity required to improve returns. Qianchuan video ad ROI was stuck at 0.4-0.6, consistently below industry benchmarks.

The math was clear: the problem wasn't budget. It was creative throughput.

The Challenge: The "Creative Drain" Problem in E-Commerce Advertising

Batteries are a functional product category. Purchase decisions are fast, and users are highly sensitive to product claims around battery life, use cases, and value. But this category brings its own specific content production challenges in paid advertising:

Platform audiences want completely different content

Douyin users respond to story-driven, scenario-based content. JD.com and Taobao feed users want direct product specs, comparison data, and price anchors. Every script needed to be adapted from scratch for each platform — and as the number of test angles grew, so did the manual workload.

Repetitive creatives get deprioritized fast

Early in the campaign, the team ran the same video assets across multiple ad accounts. Platform algorithms quickly identified the content similarity and suppressed distribution. The lifespan of each creative shortened — but headcount didn't increase to compensate.

Low data density makes optimization nearly impossible

When you can only ship 100 videos per week, the number of winning creatives is small. There isn't enough statistical signal to identify patterns — which content angles perform better on which platform, which hooks drive higher completion rates for this product category. Gut-feel content decisions meant low testing efficiency.

Cross-platform, multi-SKU content matrix costs are multiplicative

Three platforms × multiple SKUs theoretically requires an enormous volume of content. Manual scripting and editing simply couldn't scale to cover it. The result: the team constantly had to narrow scope, leaving potential ad placements uncovered.

Old Way vs. New Way

	Manual Production	AI Batch Production (Clipo)
Script creation	Editor writes individually: 3-5 scripts per person per day	AI generates multi-angle variations from structured inputs
Platform adaptation	Manual rewrite for each platform, redundant work	Same product inputs, one-click platform-specific variants
Weekly output	~100 videos/week	200+ videos/week
AB testing capacity	Limited angles per week, insufficient sample size	Multiple angles live simultaneously, fast signal
Footage retrieval	Memory-based folder browsing, 10-20 min per clip	Semantic search, 30 seconds to locate
Data feedback loop	Data and production disconnected, patterns hard to extract	Performance data feeds directly into next batch direction

The Solution: Step by Step

Step 1: Asset-ize the Footage Library

The brand had accumulated substantial footage across multiple product shoots — but it was organized by batch date and product model, with no semantic structure. Finding a specific scene meant relying on memory and manually scanning through directories. Ten to twenty minutes per search was normal.

After uploading the full library to Clipo, AI automatically analyzed and annotated every clip in natural language:

• "Product placed on outdoor grass, front-facing packaging visible"
• "Model using flashlight outdoors, close-up on battery compartment"
• "Unboxing sequence, charge indicator LED highlighted"

From that point on, footage retrieval became a search query — type a semantic description, get matching clips back in 30 seconds. This is the prerequisite for everything else. Without a structured asset library, batch production has nothing to work with.

Step 2: Multi-Platform, Multi-Angle Parallel Testing

With a searchable asset library in place, the next step was generating multiple creative angles simultaneously and testing them across platforms.

The team structured each SKU's core value propositions — model specs, key use cases, runtime performance, target audience — and fed them into Clipo alongside reference high-ROI videos. The AI decoded the structural logic of top performers and generated multiple copy-angle variations:

• Scenario angle: "Three days camping. This battery didn't quit."
• Comparison angle: "Same price range, runtime comparison — the competitor didn't like these numbers."
• Problem-solving angle: "When did you last change your batteries? This one will make you forget that question."
• Review angle: "I tested 6 battery brands this month. Only one made me reorder."

The same footage set, four distinct angles, each slightly adapted for Douyin vs. JD.com vs. Taobao content preferences, batch-rendered into finished videos.

Step 3: Build the Production-Distribution-Data-Optimization Loop

Once the batch-produced videos went live, performance data began flowing back: which platform, which angle, which hook type was driving higher CTR? Which SKU had the lowest cost-per-conversion? Which opening structure produced better completion rates?

This data stopped being something "someone looked at" and became direct input into the next production batch. High-performing script structures were logged as templates and prioritized for replication in the following week. Underperforming angles were retired or retested with different footage combinations.

This is something the manual workflow couldn't achieve: when weekly output is only 100 videos, the data sample is too thin to surface reliable patterns. At 200 videos per week, the signal density increases significantly — and patterns become actionable.

Step 4: Systematically Scale What Works

After several testing cycles, the team started identifying structural templates that consistently delivered strong performance on specific platforms for specific SKUs — for example, "on Douyin, for battery products, a scenario-entry + spec-showcase combination outperforms other angles by 40%+ on ROI."

These high-value structures were systematically replicated: swap the footage, adjust copy details, increase variant count, while keeping the proven core structure intact. This is what the creative formula looks like in practice: Creative Output = (Accumulated Learning + New Footage) × Feedback Quality × Iteration Volume. The deeper the historical learning, the more efficient each iteration becomes.

Results

After implementing the full workflow and running it consistently:

• Creative output doubled: Weekly production increased from 100 to 200+ videos, with significantly lower per-creative production cost
• Qianchuan ROI improved substantially: Douyin channel ROI moved from 0.4-0.6 to 1.3-1.4, rising from below-benchmark to standard industry performance
• Full three-platform coverage: Simultaneous paid distribution across Douyin, JD.com, and Taobao — no longer sacrificing platform coverage due to creative production constraints
• Testing velocity accelerated: The number of content angles testable per week increased dramatically; time to identify high-converting directions shrank from weeks to days

ROI going from 0.5 to 1.4 has a simple underlying logic: in creative-consumption-driven paid channels, testing volume is the core lever for ROI. Double the creative output means double the hypotheses tested in the same window of time. The probability of finding high-converting angles increases proportionally.

Key Takeaways: A Repeatable Creative Scaling Playbook for E-Commerce Ad Teams

1. Asset organization is the prerequisite for scaling production

The production bottleneck is usually upstream of editing — it's finding the right footage. If you can't retrieve it, it effectively doesn't exist. Structuring your historical library is the step that unlocks everything else.

2. Platform adaptation should be "same-source variants," not separate productions

The same product value proposition can be reformatted for different platforms at near-zero marginal cost with AI. Treating each platform as a completely separate creative effort is the slowest and most expensive approach.

3. The essence of Qianchuan ROI improvement is testing density

Improving Qianchuan video ad ROI isn't about finding one perfect creative — it's about testing enough hypotheses per unit of time. At 200 weekly variants, the statistical chance of identifying high-converting angles is dramatically higher than at 50.

4. The testing flywheel compounds more value than any single winning creative

A viral video is a one-time event. A workflow that consistently ships 200+ variants per week, tracks their performance, and feeds findings back into the next production cycle is a compounding asset. Build the system, not just the content.

5. Category characteristics should shape angle prioritization

Battery products differ from impulse-purchase categories like beauty or fashion. User decisions are more rational, and practical information — runtime specs, use case scenarios, value comparisons — carries more weight. AI-generated content angles need to be filtered and prioritized according to category-specific decision drivers.

Frequently Asked Questions

If Qianchuan ROI isn't improving, is it a creative problem or a bidding strategy problem?

Both are possible, but with insufficient creative volume, it's hard to reliably distinguish them. When you're only launching a few dozen creatives per week, the sample is too small for statistically meaningful diagnosis. The recommended approach: increase creative output to 100+ per week first, then use the resulting data to identify whether the issue is content angle, audience targeting, or bid strategy. Data density is the prerequisite for accurate ROI diagnosis.

Will platforms detect and penalize AI-generated video ad creatives?

Platforms penalize repetitive creative — not AI-assisted production. Clipo's batch output is genuinely differentiated: each video draws on different footage segments and uses a distinct copy angle. From the platform algorithm's perspective, these are separate creatives. The key is that differentiation must be real — not just changing a subtitle while leaving the video frame unchanged.

Can we start AI batch production without a large existing footage library?

Yes, but the efficiency gains scale with library size. With footage from at least 2-3 product shoots, you have enough material to start. In the early phase, AI tools can still meaningfully reduce script writing time and generate copy variations even with a limited library — while you build up the asset base over time.

What type of e-commerce ad teams benefit most from this workflow?

Teams running simultaneous campaigns across multiple performance channels — Qianchuan, Xiaohongshu native ads, JD marketing, Taobao Zhitongche — with a continuous need for differentiated creative supply. Particularly: teams managing multiple SKUs in parallel, teams with major sales event deadlines requiring creative scale-up, or teams where current creative testing volume is insufficient to diagnose and improve ROI. The smaller the team relative to the creative demand, the higher the per-person leverage from this approach.