Laurent El Ghaoui, and Serge Marquié, sumup.ai
Following Uber and Lyft, it is now Slack’s turn to go public. As with any company filing for a public listing, Slack had to submit a special document, called S1, to the SEC. The S1 document is intended to provide information to market participants, prior to the public listing, about company structure, projected revenue, customer acquisition, business risks, etc.
S1 filings are typically long documents (221 pages in Slack’s case), with lots of financial and legal jargon. Sometimes, based on new events, changes, or on a request for clarifications by the SEC, the company submits one or more amended versions, referred to as S1-A filings. In the case of Slack, an investor is facing today a set of four S1 documents, one original and three subsequent amendments, totaling 889 pages: not your typical breakfast read.
Since Slack generates a lot of interest, there is a plethora of news articles offering their analysis of its S1 documents. This article from Crunchbase is typical: after offering a general presentation of Slack, the author comments on numbers such as projected revenue, as found in the S1 document; this article does not mention the several versions. As such, these analyses, while useful, focus on a tiny part of the S1 documents, that which is contained in the various numerical tables.
As an investor, we may be interested in assessing various risks associated with the company; we would also like to understand what events prompted the company to file amended versions. For this, one must go further, and enter into unstructured data land, and analyze the text itself.
In a previous study on a large number of Amazon and Facebook SEC documents, we have shown how to use large-scale text analytics to quickly uncover insights that are relevant to investors. In this post, I focus on contrast analysis of the different versions of the S1 filing: how are the four versions different? Based on those differences, what are the questions a potential investor should ask? Our focus here will be on contrasting the last filing (S1-A#3) against the first.
Contrast topic analysis is a novel technology developed at sumup.ai that allows one to quickly contrast two (sets of) documents. It automatically extracts a few keywords and corresponding sentences, chosen because they collectively capture the difference between the two sets. The list of “contrast” keywords is called a “contrast topic”. We can extract multiple contrast topics, each revealing a different facet of the difference between the two corpora.
A basic approach to the contrast problem consists in just finding words that are in the new document (S1A#3) and not in the other (the original S1). The list comprises 89 “new words”, which are involved in 329 “new sentences”.
This approach is already informative .The most frequent new words shown above reveal a potentially interesting topic around “revolving”, “covenants” and “borrowing” — more on this later.
Before diving into this aspect, let us note that the simple counting approach has shortcomings. Some words in the list are not informative, such as “amendment” — that word does characterize well the S1-A#3 document against the original S1, but we expect S1-A documents to contain that term, as that is what S1-A documents are: amendments. Another example is “Chicago”, which refers to one office location which was not mentioned in the original S1 filing. Such terms are “false positives” and bring no new information.
In addition, there are words that are common to the two documents, and thus not part of the “new word” list above, such as “credit”; however their context and surrounding language may be very different, hence it should be included in the list of contrast terms. The term is a “false negative”.
Contrast analysis solves both problems of false positive and negatives. In the table below, we present the result of contrasting each one of the versions against the previous ones, leading to a set of “most distinguishing” keywords. The tool also picks a representative sentence for each such keyword, in order to illustrate the context in which it is mentioned.
The first term in the topic is “unaudited”, which points to the presentation, by the company, of new financial statements since the first filing, which have not been yet audited. The term appears 9 times in the original filing, and 38 times in the new. This raises a question: why are there so many unaudited documents?
The second set of terms points to “revolving credit” as a topic of importance in the contrast between S1 and S1-A#3, consistent with our first simple word counting approach. The sentences shown point back at specific pages of the document, namely pages 38 and 87 in S1-A#3, where we learn that:
- “Our revolving credit facility provides our lenders with a first priority lien against substantially all of our assets, and contains financial covenants and other restrictions on our actions that may limit our operational flexibility or otherwise adversely affect our results of operations.”
- “In addition, the revolving credit facility contains financial covenants, including a minimum liquidity balance and a minimum revenue amount . We were in compliance with all covenants under the revolving credit facility as of May 30, 2019.”
Revolving credit facility is a fairly constraining type of loan, which raises a few questions for the analyst:
- Why Slack had to enter such a credit facility? Is it because the company needed some kind of bridge loan, perhaps because they had to delay their public listing with respect to an original date?
- Why was this loan not mentioned in these terms in the initial S1?
- As of 30 May 2019, Slack is in compliance with covenants, but how far are they from breaching those covenants?
- The syndicate of financial institutions involved in the loan have a lien on all assets; are proceeds of the direct listing going to repay a fraction of this facility? Are there other covenants that must be satisfied first?
What have we learned from this analysis? Mostly we learned which questions a potential investor should further investigate. Thanks to the contrast analytics technology developed by sumup.ai, one can quickly identify new, contrasting information in different versions of S1 filings, leading one to meaningful and focused questions that have a potential impact on valuation and investment decisions.