Frequently Asked Questions

How does Text IQ fit into the privilege review production workflow?

This answer will vary from firm to firm depending on the workflow option in use. Our solution is designed to plug into any workflow our customer requires, with no training or integration requirements.

The most common approach is to have us run in parallel to the responsive process (either TAR or review) so that the privilege coding will be ready as soon as the TAR model is stabilized.

An alternative approach is to run Text IQ on only the responsive set after that process is complete.

For Additional Information:
AI for Document Review by Stefan John, GC of BASF
Litigation Game Changer by Laura Kibbe, AGC at IQVIA
Litigation Expenses Slashed by Automating Doc Review
How Artificial Intelligence Works and Its New Role in Litigation

Relationship with Relativity
Does Text IQ software work with Relativity?
Yes. Text IQ does run completely independently from Relativity but works well in conjunction with it. If you have already processed the data in Relativity, we have data bridges that can facilitate data transfer to Text IQ’s environment. The inputs that our AI needs often already exist within Relativity.
We currently have a Relativity plugin for version 9.7 and above. This allows for continuous, active updates as you continually code the docs we've flagged as PP.
Screen Shot 2021-04-09 at 5.31.47 PM
Email Threads
How does our platform handle email threads and attachments?
Text IQ can handle both email threads and attachments. For email threads, we’re able to recognize emails in the same thread belonging to different custodians since we can parse out individual email nodes from an email chain. Attachments are processed along with the parent email so that the context of the attachment is not lost.
Privilege Review
How is Text IQ different from Predictive Coding?
Predictive Coding (PC) is a marketing term for document classification. We differ in two major ways:
  1. With PC, you need a human (Lawyer SME) to label a large sample doc by doc. This process can be extremely costly, inherently biased, and take up a lot of valuable SME time doing low-value work
  2. The supervised machine learning, PC looks at the context on one document. We take a socio-linguistic approach and incorporate context from the entire data set, looking beyond the four corners of the document. This allows us to cast a tighter net.
You can see Text IQ AI for Legal - Privilege Review in action here.
How do you deal with over-inclusive search terms?
Text IQ’s AI has been designed specifically to cast a tighter, more accurate, and consistent net around potentially privileged documents and communications—establishing the connections and context at scale that human reviewers often miss.

One practical way that Text IQ avoids over inclusive search terms is by excluding email footer text that includes terms like legal. But where our approach significantly reduces overinclusion is by context—not if the term “legal” shows up, but where it shows up and the relationship between the individuals in the communication.

search term priv review

Search terms narrow the scope of document review somewhat while still missing many sensitive documents. Text IQ’s AI casts a wide enough net to fully encompass all sensitive documents, but yet small enough to not get overwhelmed by false positives. 

How do you score documents that the machine surfaces as privileged?

The score is on a scale of 1 to 0. As the score approaches 1, more legalese is used within the document and there’s a greater likelihood that it is privileged.

Additionally, Text IQ’s AI creates Priv Reasons using natural language processing (NLP) describing why a document earned a specific score. For example, a document with a high score may have a Priv Reason like, “Email communication sent from X to Y regarding or reflecting the advice of in-house counsel Z, as noted in the email body.”

Other data fields included in Text IQ’s output include categories, such as “POTENTIALLY_PRIVILEGED Law Firm/Lawyer Name in parent email,” and Total Recipients, which reflects the percentage of attorneys on the communication in relation to the total number of recipients.
privilege log

If we have a list of law firms, but not the attorney names, can you identify the attorney? e.g. client knows they used JonesDay, but not the attorney name?

Yes, unlike predictive coding, our AI does not use the law firm lists and attorney lists input as search terms. It’s looking deeper into the context to understand how the people on the lists are communicating. While additional info input is helpful, providing just the law firm is enough for this process.

Can your tech help us identify when a lawyer is acting in a business role vs. a legal role?

Yes, our AI solution can successfully tease out an individual's role based on the context of the dataset. It is precise enough to identify the nuance in this dual role issue. We recently illustrated this with a large pharma client.privilege log

For Additional Information: 
POC Gains Buy-In For AI Investment  

Privilege Log
How is the language for priv reasoning generated?
The easy-to-understand Priv Reasons automatically generated by Text IQ AI allow for easy QC of potentially privileged documents. These Priv Reasons are created using natural language processing (NLP) to describe why a document earned a specific score.

Priv Reason example: “Email communication sent from X to Y regarding or reflecting the advice of in-house counsel Z, as noted in the email body.”

Other data fields included in Text IQ’s output include categories, such as “POTENTIALLY_PRIVILEGED Law Firm/Lawyer Name in parent email,” and Total Recipients, which reflects the percentage of attorneys on the communication in relation to the total number of recipients.
priv reason
How does the priv reasoning become the priv log?
Because the Text IQ machine is already generating reasons for scoring a document or communication as potentially privileged, customers can leverage the output with supplemental definitions of privilege categories and labeling protocols to automatically generate a priv log (and integrate with review platforms). Text IQ also provides the ability to redact PII in the priv log and other produced documents.
QC & Validation of Results
Can reviewers perform QC on the documents identified as potentially privileged?
Text IQ provides a full coding panel, as well as a QC workflow. We consider it critical to the effectiveness of the technology that reviewers can annotate or correct where our machine has made a mistake. Reviewer annotations and QC output are also fed into our AI Brain for continuous learning and to refine our model on an ongoing basis. Typically, our clients/firms will already have a defensible QC process in place. We will complete the first-level priv review faster, letting you do more of the same defensible QC as before.
How can I validate the output and build a case for defensibility?
There are a number of ways to validate the output. Reviewers can compare and evaluate Text IQ output against priv screen results or tagged search terms. Another method is to perform statistical sampling of lower score documents (such as 0.2 and below) to assess whether privilege has been missed.
Privacy Redaction
How are redactions applied?
Redaction is a form of editing where information is permanently removed from a document. This is typically used to remove sensitive information, such as personal information (PI) or personal health information (PHI). Redaction can be performed at any step during the document review process or on its own for information governance purposes.

Text IQ’s redaction capabilities remove (not just cover or obscure) content within a region of a native document. For documents such as PDFs, redaction involves blacking out or cutting out areas of the page—permanently removing the underlying text or image. For structured files, such as Excel files, the underlying content is removed and a redaction label is fitted to replace the omitted text.
email_3 redact
How secure are your redactions?
To apply redactions, Text IQ creates copies of the native documents and converts them into either PDF or Excel files. After conversion, redactions are applied to these documents for a consistent redaction experience. Text IQ’s redaction completely removes the underlying text or image so that the underlying information cannot be recovered from the file on which it is applied.
Can you redact metadata?
Yes, Text IQ supports metadata redaction and even provides an experience within the User Interface for this process. This allows review teams to simultaneously redact metadata during the course of their standard review process without isolating the affected documents for a separate workflow.
What is Text IQ's pricing structure?
We charge on a $/document basis that flexes based on the size of the project—larger projects have a significantly lower $/document price.

For answers to specific pricing or billing questions, please contact sales@textiq.com.
Are there any additional fees from the cost of the software?
Our pricing includes all services costs to run our software. We charge a flat per-month hosting fee and a per-hour software training fee. Optional fees include processing fees if the data is not yet processed, speech-to-text transcription, reviewers*, and other miscellaneous fees, if applicable to your needs.

*We do not have in-house reviewers, but can provide reviewers, if desired, through one of our LSP partners.

For answers to specific pricing or billing questions, please contact sales@textiq.com.

If you have additional questions please contact us at info@textiq.com.

Become a Partner in Innovation