As for the privacy concerns, we use data to train our machine learning models. The outcome of those models includes categorization by activity (internal meetings, phone calls, legal filings, etc.).

The "shared" component is for training purposes. At no point do we share invoice content. The key to convert a numeric vector back into a partial view of the original text changes every day. Security and confidentiality of customer data is a big area of concern for us. We also provide our customers with a named list of people with access to the content.