{"id":10719,"date":"2021-04-27T17:27:38","date_gmt":"2021-04-27T16:27:38","guid":{"rendered":"https:\/\/www.imprima.com\/?p=8700"},"modified":"2023-09-26T10:17:33","modified_gmt":"2023-09-26T10:17:33","slug":"blog-finding-information-in-contracts","status":"publish","type":"post","link":"https:\/\/www.imprima.com\/blog\/blog-finding-information-in-contracts","title":{"rendered":"Discovering Information: What Does it Cost, What Does it Take?"},"content":{"rendered":"\n
<\/div>\n\n\n\n

By Pieter van der Made<\/b><\/strong> (Executive Chairman) and Jeroen de Knijf<\/b><\/strong> (Senior Data Scientist) at Imprima<\/a><\/b><\/strong><\/p>\n\n\n\n

<\/div>\n\n\n\n

Introduction<\/h2>\n\n\n\n
<\/div>\n\n\n\n

Do you feel you are wasting a lot of time looking for information and documentation? Well, you are not alone. According to McKinsey, \u2019employees spend 1.8 hours every day \u2013 9.3 hours per week, on average \u2013 searching and gathering information\u2019 [1]<\/a>. IDC data shows that \u2018the knowledge worker spends about 2.5 hours per day, or roughly 30% of the workday, searching for information\u2018<\/b><\/strong>.<\/p>\n\n\n\n

These numbers are quite shocking. But it is indeed reality, even more so in M&A Due Diligence<\/b><\/strong><\/a>, where the objective is, first and foremost, just to find information. And that can be challenging. For example, finding red flags in a large number of agreements is often like looking for a needle in a haystack. Imagine how much productivity would improve if the information could be found more easily. In addition, again in particular in Due Diligence, time is limited, so therefore the likelihood of missing certain information is high, reducing the quality of Due Diligence.<\/p>\n\n\n\n

So how do we search for information in documentation? First of all, we must find the relevant documents. Only then can we find the information IN the documents.<\/b><\/strong>
<\/p>\n\n\n\n

<\/div>\n\n\n\n

Traditional Search Methods<\/h2>\n\n\n\n
<\/div>\n\n\n\n

How do we traditionally search for information?<\/p>\n\n\n\n

In the case of a Due Diligence project, the data is normally stored in a Virtual Data Room<\/b><\/strong><\/a>, where the documents are organised in a hierarchical folder structure. Due Diligence professionals (lawyers at the advising law firm or analysts at investment banks) will try to find documents on the basis of this structure (In non-Due Diligence situations the problem is not much different: in most organisations, documents are stored in hierarchical structures). The problem is that the location where documents are saved in this structure is ambiguous and subjective.<\/b><\/strong> Different people make different choices as to where to save documents. Unless the information you are looking for maps 1:1 of this structure, it is often challenging to find the relevant documentation, and hence the information in it.<\/p>\n\n\n\n

An alternative is to tag documents with certain keywords so that a search can be done on the keywords of the documents. However, again, the way documents are tagged is ambiguous and subjective.<\/b> <\/strong>You are very dependent on how others have tagged the documents, and whether the tags are relevant for your query, and whether they have been tagged at all.<\/p>\n\n\n\n

Yet another alternative is to search the whole set of documents, and all their content, with a \u2018full-text search\u2019 (and some logic captured in so-called \u2018regular expressions\u2019): trying to find the documents that you need by looking for those documents that contain certain terms.<\/p>\n\n\n\n

Though this is very flexible, full-text search is not very intuitive and does not capture the semantics of what you are looking for. To use an everyday example: if you google \u201cvegetable\u201d \u201cgarden\u201d and \u201crabbits\u201d, you are probably not looking for documents that contain the words \u201cvegetable\u201d \u201cgarden\u201d and \u201crabbits\u201d: you are more likely to instead be looking for documents or tips that give advice how to keep rabbits out of your vegetable garden (for instance). But a full-text search will not accomplish the latter.<\/p>\n\n\n\n

As a result, a full-text<\/b><\/strong> search is not very accurate, and often even not very useful. It is likely to return any documents that contain the keyword you entered that are actually not relevant for you (in other words, low \u2018Recall\u2019 (for more information on \u2018Recall\u2019 and why it matters, see our previous whitepaper \u2018<\/a><\/b><\/strong>AI that really works\u2018<\/b><\/strong><\/a>). That means you will still spend a lot of time ploughing through the documents only to determine if they are relevant at all. At the same time, what such a search returns is likely not to include all documentation you are looking for (in other words, low \u2018Precision\u2019, again, see \u2018AI that really works\u2018<\/a><\/b><\/strong>).<\/p>\n\n\n\n

\"\"
Download your free copy of the whitepaper<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n
<\/div>\n\n\n\n
<\/div>\n\n\n\n

Machine Learning-Based Search<\/strong><\/h2>\n\n\n\n
<\/div>\n\n\n\n

What we need is something that is:<\/p>\n\n\n\n