Introduction
It has been more than 2 years since the inception of ChatGPT, which sparked the AI boom, at least from the investment side. You might have tried chatting with these models and already used them for tasks like generating copy for your product pages, updating your SEO tags, or creating blog posts. However, today, I want to show you a more advanced use case that involves analyzing your product reviews to understand better how to improve your product rating.
How product rating improvement can impact your sales
Here is a study done by Mcnsey how product rating improvements affect sales across different categories:
Depending on the initial score and the category, the improved rating can lift sales from 5-10% per basis point improvement. So, for example, if you have a product with a score of 3.5 and you improve it to 4, with an average lift per 0.1 of 7%, the total expected improvement in sales in the next 24 months can be around 35%. Now imagine doing this across all your products, what would be the impact on your bottom line?
The best thing is that this doesn’t involve increasing your ad spend or hiring more people to run new marketing campaigns. All you need is to have enough review data for your products that could allow you to understand why some products are underperforming.
How many reviews are enough? Around 100 reviews per product is a good starting point, the more, the better. If you want to collect more reviews, then have a look at this article instead.
If you have enough product reviews, let's see how to leverage them to increase your rating.
How to use AI to analyze your product reviews
I have made a quick video explaining this part if you prefer to watch it, check out here:
The first step to analyzing your reviews is identifying which topics/categories are driving the negative & positive sentiments of your product ratings. You might already know these categories if you have dealt with multiple customers, but it's easy to find out if you don't. Just export the reviews for one or more products and past them together with a prompt like the following:
Can you extract similar topics for positve and negative sentiment of the following reviews. To make sure its accurate group each topic by the number of reviews attributed.
Exclude any topics not related to the product (e.g. delivery, customer support).
Here are the reviews:
```
Cranberry Sorbet
Then some flowers arrived broken and the bouquet looked withered. Also the bouquet was no where close to the fullness / size of the original picture and when I sent them the images to compare as I spent 45 pounds , and they replied saying "its exactly the same". Which it was clearly not.
City of Angels
Poor quality and quantity and no response to my complaint
```
If you’re using Trustpilot, you can get reviews by clicking on the export button:
Now, once you have the topics/categories that you see are driving your ratings, it’s time to analyze each product independently.
For this, you will need to use another prompt, here is an example:
What we do is provide instructions for the model to find a list of categories in the review and determine if they are positive or negative. Then we take these scores as JSON files and combine them together into a single JSON file, which we can open within a spreadsheet or, even better, load it to a cloud datawarehouse like BigQuery. Once loaded, you can analyze the data using SQL queries and create reports.
Alternatively, if you are analyzing a limited number of reviews, 100-1000s, then you can ask the model to aggregate topics for those reviews and provide a summary of them in the output. The downside is that you have to do it manually for each product, and it can take a long time and cost more to process as you run large prompts on every question. For example, on average, if you are analyzing 10000 reviews, that would be half a million tokens. So, depending on the model, it could cost from $0.05 to $1.2 per request for non-reasoning models.
Creating a product performance comparison report
Now, let's take a look at how to evaluate the AI category data for your product reviews. As an example, we will be using the StackTome product performance report, but you could use the same process for any other tool, like BigQuery if you have loaded your data there.
When you load the data, you should be able to see all categories scored per product like here:
Next, you want to filter products that have a sufficient number of reviews per category, and we want to look at products that have scored lower than 4 for any of the 3 categories:
Finally, we can inspect each matched product's individual reviews and see what recurring problems could be addressed to improve the rating:
Once the product is improved, you can re-run the analysis after a few months to see if new reviews don’t contain the same problems and track if the rating has started to improve. With improved ratings, you can expect product sales to increase as well.
If you want to formalize the process, then here is a recommendation from McKinsey on how to use it in your company:
Summary
That’s all you need to know how to get more insights on improving your product rating and lifting sales. With new tools powered by AI, it is easier than ever to understand your product issues and use them to your advantage in growing your eCommerce brand.
If you want to fast-track this process and get your insights right now, then consider signing up for the StackTome standard plan - https://www.stacktome.com/pricing - and our team will help you set up the report that you can use to start improving your products today.
