Data accuracy is our top priority. We’ve optimized our AI-sales processing algorithms to identify statistical anomalies and outliers across our entire data set. New algorithms are applied app-wide to label (and in some exceptions adjust) every single data point we track, so that we can report our own confidence in data transparently.

To that end, each data point is assigned a Confidence Score, from 0 to 10, 10 being the highest. The confidence is bucketed into 3 tiers:

1. High: the data had little to no statistical anomalies (outlined below)
2. Moderate: the data tracked had some statistical anomalies
3. Low: the data tracked had many statistical anomalies

We then aggregate the scores for every product across a company to score a company’s general confidence.

A product with no penalties will receive a confidence score of 10.

How does the model work?

We use a combination of machine learning algorithms and fundamental statistics to help us identify outliers in sales volume and inventory data.

Sales data is delivered in a time series format, that we calculate rolling averages and historical means, medians, and standard deviations from. While calculating, we flag anomalies like:

• large inventory drops
• restocks of inventory
• dramatic increases in sales volume without any other changes to a product

When we identify a data point with 1 or multiple anomalies, we dock the confidence score as a result.

Example:

An example of this could be over 7 days with the following data points:

Jan 1: 50
Jan 2: 45 
Jan 3: 52 
Jan 4: 60 
Jan 5: 70 
Jan 6: 39 
Jan 7: 43 

Then the next day shows a calculated sales volume of 1,300:

Jan 8: 1300 -> a large spike above the historical mean