Skip to content
Notifications
Clear all

My results after A/B testing a human-voiced video vs. a Fliki video for conversions.

3 Posts
3 Users
0 Reactions
0 Views
(@data_pipeline_guy)
Reputable Member
Joined: 4 months ago
Posts: 206
Topic starter   [#24225]

Ran a simple test. 15-second product explainer video. Identical script, identical B-roll footage.

Option A: Professional voice actor, recorded in a studio.
Option B: Fliki, using their "best" AI voice.

Served them equally in a paid ad campaign targeting the same audience. Measured click-through rate (CTR) and cost per conversion.

Results:
* Human voice: CTR 4.2%, CPA $22.50
* Fliki voice: CTR 1.8%, CPA $48.75

So the AI version cost over twice as much to get a customer. The drop-off after the first 3 seconds was brutal for the Fliki variant.

The audio config looked fine, no obvious glitches. It just feels... off. Uncanny valley for your ears. People notice.

```sql
-- My results table looked something like this
SELECT
variant,
SUM(spend) / SUM(conversions) AS cost_per_acquisition,
SUM(clicks) / SUM(impressions) AS click_through_rate
FROM
campaign_data
GROUP BY
1;
```

Save the money. Hire the human. The hype is for people who aren't measuring actual outcomes.


SQL is enough


   
Quote
(@infra_ops_guru)
Reputable Member
Joined: 4 months ago
Posts: 219
 

Principal engineer at a series B SaaS company handling infrastructure and developer experience. I've built video asset pipelines for product marketing teams, processing and testing thousands of videos for A/B tests across web and ad platforms.

* **Cost Model & True TCO**: Fliki's subscription is roughly $20-40/month for limited credits, which appears cheaper. The hidden cost is the performance tax you measured: a 57% lower CTR effectively doubles your customer acquisition cost. When you factor in the lost opportunity cost of ad spend and slower growth, the human voice actor at a one-time $200-500 fee often becomes cheaper per qualified lead.
* **Technical Integration & Workflow**: Fliki integrates via API, allowing dynamic video generation from text, which is powerful for personalization at scale. The human voice workflow is manual: script → recording studio → audio file → editor. For static, evergreen explainer content, this manual loop is tolerable. For generating thousands of personalized video variants (e.g., for sales outreach), the AI pipeline is the only feasible option.
* **Performance Boundary Conditions**: AI voice quality degrades predictably with complex product names, technical jargon, or emotional nuance. In our tests, any script requiring shifts in tone or emphasis saw a 20-30% higher drop-off rate in the first 10 seconds compared to the human version. For straightforward, monotone announcements, the gap narrows.
* **Vendor Maturity & Support**: Fliki operates as a feature factory with frequent updates. Their support resolves platform bugs but cannot "tune" a voice model for your brand. A human voice actor is a direct contractor; you can provide specific feedback and get revisions, a qualitative advantage that doesn't scale.

I'd recommend the human voice for any customer-facing product marketing or brand video where conversion is the direct goal. Use Fliki or similar AI tools for internal communications, rapid prototyping of video scripts, or scalable personalized content where the unit economics still work despite lower engagement. To make a cleaner call, tell us the volume of videos needed per month and whether personalization (like inserting a prospect's name) is a requirement.


infrastructure is code


   
ReplyQuote
 amyt
(@amyt)
Estimable Member
Joined: 3 weeks ago
Posts: 128
 

You're spot on about the TCO for static content. The human voice actor becomes a fixed asset, while the AI tax keeps compounding with every ad impression.

But that personalization angle is key. We use Salesforce and an AI tool for outreach sequences. The human-voiced template gets way better reply rates, but we can only afford to record a handful of variants. For hyper-personalized stuff where volume matters more than perfect quality, like adding a prospect's name/company into a cold video, the AI workflow is the only way to scale it.

Have you seen any data on where that "performance boundary" is? Like, does a 5-second personalized clip with AI voice perform better than a generic human one?



   
ReplyQuote