3 ms·

Show HN: I built an AI that predicts which images will sell more products

3 points by apentecoste 2y ago

Hey everyone, I'm a developer who has spent the last 6 months building an algorithm that predicts which images will sell. Results are great, AI predictions land within 10% of human ranking in 93% of cases. It’s been an exciting journey, but now I'm stepping into the world of marketing, which is pretty new to me. I’ve created Focalstudio's landing page to showcase what this AI can do, but I’m not sure if it’s up to the mark. I would really appreciate some honest feedback on it. What am I doing wrong? What can I improve? I’m here to learn and make it better. Arthur

apentecoste 2y ago
Hey HN, Just wanted to add a bit of context here. I built this algo because my business owners friends did not know which image to use to sell more products. They don't have enough user traffic to do proper A/B testing and can't afford to pay for survey or focus groups. I’d love to hear your thoughts on the landing page and the concept itself. Is it clear what the product does? Does the page communicate the value effectively? Any feedback would be super helpful. Looking forward to your thoughts!
XCSme 2y ago
Your landing page looks cool and product use-case is clear. My main concern is that predicting sales is very hard, and I need to be convinced of how can the AI even tell which one would sell better. From what I understand, most of the "sell better" part is simply having a higher quality image, not necessarily with what it's shown in the picture. So, I could show a high quality picture of a shampoo while selling a phone, and the AI will tell me that it will sell better than a low-res photo of the phone. So, overall, I don't trust AI capabilities enough to decide which photo will do better. I would think a basic A/B test will be more accurate than using AI. Maybe not focus on the "AI" part, as lately people are less likely to buy something that is AI-focused.
- apentecoste 2y ago
  Thanks a lot for the feedback. It actually resonates a lot. I've been playing with a few ideas to increase confidence in the output. My two best ideas right now are: 1. Enable users to connect their sales data (e.g. shopify integration) so that impact can be quantitatively measured. 2. Augment the predictions with more context from the entire product page. e.g. check that there are at least 6-8 pictures, with different styles, different angles, check that the product page layout is easy to read, price and add to cart button should come out... Curious to hear which one you think would most efficiently address your point

3 points by apentecoste 2y ago

apentecoste 2y ago

Hey HN, Just wanted to add a bit of context here. I built this algo because my business owners friends did not know which image to use to sell more products. They don't have enough user traffic to do proper A/B testing and can't afford to pay for survey or focus groups. I’d love to hear your thoughts on the landing page and the concept itself. Is it clear what the product does? Does the page communicate the value effectively? Any feedback would be super helpful. Looking forward to your thoughts!

XCSme 2y ago

Your landing page looks cool and product use-case is clear. My main concern is that predicting sales is very hard, and I need to be convinced of how can the AI even tell which one would sell better. From what I understand, most of the "sell better" part is simply having a higher quality image, not necessarily with what it's shown in the picture. So, I could show a high quality picture of a shampoo while selling a phone, and the AI will tell me that it will sell better than a low-res photo of the phone. So, overall, I don't trust AI capabilities enough to decide which photo will do better. I would think a basic A/B test will be more accurate than using AI. Maybe not focus on the "AI" part, as lately people are less likely to buy something that is AI-focused.

apentecoste 2y ago
Thanks a lot for the feedback. It actually resonates a lot. I've been playing with a few ideas to increase confidence in the output. My two best ideas right now are: 1. Enable users to connect their sales data (e.g. shopify integration) so that impact can be quantitatively measured. 2. Augment the predictions with more context from the entire product page. e.g. check that there are at least 6-8 pictures, with different styles, different angles, check that the product page layout is easy to read, price and add to cart button should come out... Curious to hear which one you think would most efficiently address your point