Key takeaways for using Okkular AI based Tag-gen solution
- Over 8 weeks of full-time hours saved for tagging over 20k products. This helped in reducing cost and time significantly and go to market faster.
- Produce over 20 product specific Tags/attributes for each apparel, which helped in making the search engine more efficient and robust
- Over 80% accuracy on an average with AI classification and the ability to audit to make it 100% accurate
- Classify multiple colours which would have taken additional time and cost to add for every product manually.
Intro to Dresswhere.com
Dresswhere is an upcoming aesthetic gamified search-engine for fashion. They don’t expect users to know what they want, so instead of asking them a very complex question (‘what are you looking for?’), they ask them a very simple question repeatedly: ‘which of these three apparels is best visually?’ Every time the user chooses an apparel or skips, they’re shown another three apparels that are slightly more specific to their tastes based on their previous choices. The effect is that, over the course of ~3 minutes, the user zooms in on an individually curated collection of apparels. Whenever they scroll down, they see their top results, and can click through to the retailer.
What was the pain point for Dresswhere to make the search engine efficient?
To create the Dresswhere search engine and predict a closer matching apparel, Dresswhere had to Tag every apparel’s image with at least 6-8 attributes/Tags. The catalogue inventory was over 20K products and requiring ongoing 3-4k products added every month. Dresswhere gets the product catalogue feed from multiple Fashion brands and the data received in the feed was inconsistent and was not enough for the search engine to work efficiently. While performing the task of tagging products manually, Dresswhere had two main issues.
- Time taken/cost incurred to tag over 20k images and ongoing 3-4k products every month. Dresswhere estimated this would take approximately 9-10 weeks to tag over 20k products.
- The inconsistency and error rates
The above two issues were two key pain points to overcome for Dresswhere to successfully launch their product.
How did Okkular solve Dresswhere pain points?
Okkular.io has developed an AI based visual analysis solution which helps in automating the product tagging process. The AI tool generates tags from just the product images in under 15secs per product, increasing the speed to over 15 times faster compared with manual tagging. The tagging process also helps in reducing errors, generating consistent data and significantly reducing costs.
As an example, Okkular Tag-gen solution can generate over 12 attributes for a dress including key attributes like neckline, style, length, print and also tags for occasions and avatars. The tags generated can be used in many ways including, titles, description, filters, product specific keywords, and meta tags.
Dresswhere provided us the product feed in a google feed format and the output of Tags were uploaded by connecting via Okkular API to Dresswhere platform
Working closely with Angas (Founder Dresswhere), we were able to develop additional key tags which were very specific to Dresswhere. We also trained and generated over 20 tags instead of 6 to 8 which was an initial requirement for Dresswhere if the task had to be performed manually.
During this project we frequently received feedback from Dresswhere to improve the classification accuracy, user interface and additional features which helped us co-create a solution which has not only helped Dresswhere but many other retailers using our solution currently.
What are the benefits for Dresswhere?
- Over 8 weeks of full-time hours saved. This helped in reducing cost and time significantly and go to market faster.
- Produce over 20 Tags/attributes for each apparel, which helped in making the search engine more efficient and robust
- 80% accuracy on an average with AI classification and the ability to audit to make it 100% accurate
- Classify colours from Apparel which would have taken additional time and cost to add for every product manually.
Okkular is also working with Dresswhere to provide Okkular Visual search solution to help their users find similar looking products from multiple brands once they have identified a dress, they want using the Dresswhere search engine.
If you would like to learn more about Okkular solutions, contact
Mahendra Harish: email@example.com
For more information regarding Dresswhere, Check them out at https://www.dresswhere.com
Contact Angas Tiernan: firstname.lastname@example.org