Is TAR right for your next case?

28
Oct
2013

David Carns

Technology Assisted Review is a relatively new methodology for reviewing documents in the context of the eDiscovery workflow. It is touted to be faster, more accurate and less expensive than traditional review, which are difficult claims to ignore. With the promise of reduced data sets, increased accuracy and lower costs, TAR seems like a highly desirable way to manage the volume of data involved in discovery. How, though, does one know if a case is the right fit for this new technology? Let's explore a short checklist that can help you determine whether to use TAR in your next case.

1. What is the right number of documents?

TAR is effective at taking a sizeable number of documents and pushing much of the document review and analysis process to a machine learning algorithm. Human review is still needed, however, in order to train the machine learning algorithm and prepare it for automated review. As a result, there is a point where the time involved in training the system is sizeable enough that it would be just as effective to review all of the documents rather than using TAR. A good rule of thumb is that if you have fewer than 10,000 documents, you might be better off simply performing a manual review, since with 10,000 documents you can, using fairly conservative estimates, perform a manual review with a small team in just one week. The TAR process on any volume of documents will take 2-3 days to ensure that you go through all of the proper steps and QC. At that rate, TAR may not actually provide the time benefits you are seeking.

2. Are the documents text-searchable?

TAR is based on a machine's ability to understand a document's text. If you do not have text, you do not have a TAR process. If your dataset contains non-text documents or if you are unable to extract or create OCR text from a document, then TAR will not be effective. For example, if you have a large number of blueprints or schematic diagrams in your document collection, you will likely not see any benefit from using a TAR methodology.

3. Do the documents contain mostly numeric data?

TAR uses text to find patterns of information that can be used to automate review. When documents contain a large amount of numeric information (spreadsheets, for example), then TAR methodology is often less effective. If you do have a fair number of spreadsheets in your document collection, you may be better off using a TAR system that will ignore the numeric values and only analyze the text in each spreadsheet.

4. Have you already run basic filters on the document collection?

Almost all document collections contain explicit, non-responsive data. Documents may be out of the litigation's date range, may simply be spam email, or documents may be part of a threaded conversation regarding non-responsive topics. It is advisable to run basic and reasonable document filters on all sets of data before running TAR. You may find that after conducting such filtering that the corpus is under 10,000 documents. If you do have a sizeable corpus remaining, at least your TAR training set will be based on a more cohesive set of documents than if you let the explicitly non-responsive documents remain in the corpus for review.

5. Can you group documents into two categories?

TAR is able to categorize documents into two categories. If you are using technology to determine which documents are likely to be relevant or irrelevant, then TAR is a good option. However, if you need to determine whether a document belongs to one of 20 different categories, then TAR will not be a good option. When you train a TAR system you must choose whether a document belongs in category A or category B. It uses that training to make determinations of other documents in the corpus. TAR is well suited to making black and white determinations, but not those in shades of gray.

6. Are there many mixed-language documents and large families?

TAR is based on identifying patterns of information within the text of a document. If you have a variety of languages throughout your corpus, then it may be more difficult for the machine learning algorithms to detect useful patterns.

TAR decisions are made at the document level. You may find in your set of data that some TAR decisions create conflicts at the family level; this is especially true for datasets with numerous family groups. You will have the opportunity to resolve any family conflicts during the QC phase of the TAR workflow.

7. Do you have access to a highly knowledgeable attorney for two to three days?

TAR requires intensive review of documents for at least two to three days from an experienced attorney who is highly knowledgeable about the facts of a matter, but such a resource is sometimes difficult to find. TAR review demands a close scrutiny of documents in order to train the computer's machine learning algorithms. Although it is possible to distribute the initial training process across a number of knowledgeable attorneys, it is considered best practice to only have one “voice” during the TAR training process and, therefore, multiple attorneys are not advised.

8. Are you ready to save your clients a lot of time and money on the document review process?

If you are able to address most all of the items on the TAR checklist above, then all that remains is to ask yourself whether you are ready to save your clients a lot of time and money on the document review process, because TAR, when done right, can be an incredible technology. You can expect to save about 50 percent of the costs over traditional document review when using TAR, along with an approximate 80 percent savings in total review time, so keeping close watch on the checklist is important.