What is your methodology?
We follow a machine learning based estimation approach where Your daily sales data from AVC is used to correlate with your daily best seller rank and other signals like reviews count, SOV. This model is used to predict sales of all the SKUs in those categories.
Your method is dependent on availability of enough SKUs in any given category? If you have not worked with anyone in my category, how will this work?
Even 1 customer’s data is sufficient for our models to train and start estimating the whole category as long as there are some min number of SKUs. We don’t need data from any other customer to get started. Our ML models are optimized to work even with sparse data. It won’t work only if you have let’s say less than 10 SKUs in a category of 10,000 SKUs.
What is your accuracy and how do you benchmark it?
There is no gold standard to compare accuracy against and hence it’s difficult to give an accuracy metric. We compared our numbers against ABVP (Amazon Brand View Pro) and found our numbers to be more than 85% accurate across all studies categories. We did an accuracy measurement for SGF and it was around 93% at a brand level.
What do you need from me/us to get started? How much time is required before we see the results?
We need an Excel file containing all your active Amazon ASINs along with respective categories and subcategories. Essentially it’s your taxonomy file based on however you classify your products. A typical row looks like this BADXF4302 || Deadbolt || Electronic Deadbolt. After getting this information, we take about 2-3 weeks to create categories, classify and start estimating sales.
How do you categorize? Do I have flexibility to categorize as per our taxonomy?
We use your taxonomy as a guiding force. Our algorithms find all the similar/substitute SKUs in each of the categories and subcategories you provide to us. We provide the raw data for each category. The classification can be changed based on your feedback. Recategorization is also a self service feature in the UI.
What if I don’t agree with classification of certain SKUs? How can I get that changed?
This can be accomplished directly in the SKU level view in the Market Share report, or by filing a support ticket. If you are dealing with persistent categorization issues, reach out to your customer success manager to review.
How do you determine 1P vs 3P sales?
We gather buy box information from the product detail pages each day which allows us to estimate a 1P vs 3P buy box percentage. Sales are then split accordingly.
Do you provide variant level estimates?
Yes, we estimate the break up of sales by variants based on no. of reviews generated in a time period by variants.
How much historical data can you provide? We track share number changes YoY?
If your categories overlap with an existing model we can produce historical data, otherwise, the data will start from day that you go live.