It can be stated that nowadays almost all documents are electronically stored information (“ESI”), even if they have been at some point executed in a non-electronic form [1]. Clients send documents to their lawyers via email, data rooms, or utilizing secure file transfer protocols. Arbitrations are taking place through electronic submissions, which also include the document production phase. The review of documents for document production is becoming increasingly automated, with parties using a range of tools with different precision and recall. No one mails by post documents identified in course of document production review.
With that in mind, there is little to no difference between document production and what has been referred to in the past as e-document production [2]. Document production is e-document production.
If that is the case, model document production schedules and model procedural orders should start reflecting this. However, there is no model schedule that would address this reality.
How ESI document production works
Typically, the first step is counsel requesting their client to share all documents that are connected to the dispute. Usually, counsel would not want the client to conduct the document review for purpose of production on their own, nor decide on the relevance of specific documents. As such, counsel may expect thousands of files, including not only text documents, but also data models, emails, images.
The next step is the review of the documents for their relevance to the requests filed by the opposing party. Depending on the tools used by counsel, the process will be different, and will require different levels of human involvement.
While depending on the method used by the search engine of the tool, the specific queries will differ, as illustrated below in Example 1, the main search terms used for the process will in most instances be the same.
Example 1
| Request: The draft(s) of the Plan prepared and/or reviewed by John Smith and/or John Doe and/or Jane Roe prior to July 2010 (date of the Plan submitted by Claimant) [3]. | |
| Method of the tool | Possible Initial Query |
| Boolean Search | “Plan” AND (“John Smith” OR “John Doe” OR “Jane Roe”) AND date:<07/01/2010 |
| Keyword Search | Plan John Smith John Doe Jane Roe before:07/01/2010 |
| Concept Search (AI-Driven) | Find all drafts of the Plan prepared or reviewed by John Smith, John Doe, or Jane Roe before July 2010 |
| Predictive Coding (TAR) (assuming it uses Boolean Search method in part) | “Plan” AND (“John Smith” OR “John Doe” OR “Jane Roe”) AND date:<07/01/2010 |
| Metadata & Structured Data Search | document.title: “Plan” AND document.author: (“John Smith” OR “John Doe” OR “Jane Roe”) AND document.date:<07/01/2010 |
The steps following the initial query will differ, as will the level of human involvement necessary to perform each step of the process. Example 2 illustrates the potential differences in the next steps in the document review process, depending on the method used by the tool.
Example 2
| Method used by the tool | Next Steps |
| Boolean Search | 1. Manually review search results for false positives/negatives. 2. Manually refine search terms if needed (e.g., add date ranges, exclude irrelevant terms). 3. Manually apply filters (e.g., document type, custodian, email sender). 4. Manually review and tag relevant documents for production. |
| Keyword Search | 1. Manually validate retrieved documents for accuracy. 2. Manually expand or refine keyword list based on results, including proximity searches (e.g., NEAR/x) to improve relevance. 4. Manually apply metadata filters (e.g., author, file type) to narrow down the results. 5. Manually perform final review before production. |
| Concept Search (AI-Driven) | 1. Manually examine concept clusters to understand document relationships. 2. Manually review AI-suggested concept groupings for refinement. 3. Manually re-rank or prioritize relevant concept clusters. 4. Validate with manual sampling and human tagging. 5. Generate final production set based on AI-prioritized documents. |
| Predictive Coding (TAR) | 1. Manually review and tag an initial training set of documents for relevance. 2. Let AI learn from tagged examples and predict relevance for remaining documents. 3. Manually conduct quality control by reviewing AI-classified documents. 4. Iterate training cycle until relevance confidence reaches an acceptable threshold. 5. Finalize and produce documents. |
| Metadata & Structured Data Search | 1. Manually apply structured filters (e.g., document creation date, author, last modified date). 2. Extract and validate structured reports from databases. 3. Manually ensure chain of custody by verifying metadata integrity. 4. Convert structured data into usable formats (CSV, PDF, native files). 5. Manually review extracted data for completeness before production. |
In the final reviews the counsel would also address issues of privilege and confidentiality—either manually or in a semi-automated manner.
Main issues arising in production of ESI
In the context of ESI production, the following issues may arise:
- Preservation of documents in case of a dispute, in particular, if parties’ retention policies may result in automatic deletion of relevant ESI.
- Volume of document production, if parties use tools with different level of advancement in the process of identifying the documents.
- Scope of production, including search terms used by the parties in the process of identifying the documents, type of documents to be produced (including deleted files) or production of data which is maintained only in the database and the extraction of which would require creating a new document [4].
- Format of production, in particular, what format should the documents be produced in.
- Scope of human verification of the output of the relevancy review and/or the redaction of privileged documents.
- Concerns about authenticity of documents.
- Legal concerns, pertaining to data privacy and protection regulations, or retrieval of data held by third parties (outsourced).
Existing guidelines
Standards applied widely in international arbitration do not provide solutions addressing these issues. While the existence of some of these matters has been explicitly acknowledged in the IBA Rule on Taking of Evidence (“IBA Rules”) [5] (for example, the use of search terms to identify the relevant documents [6] or cybersecurity and data protection concerns [7]), the IBA Rules provide no specific answers in addressing them. And many arbitrators might not be fluent enough in the fast-changing world of technology to identify and address these issues on their own.
In the past, some institutions issued guidelines aimed at aiding arbitrators in dealing with the challenges arising from production of ESI. Among such institutions are ICC [8] and CIArb [9]. While they address some of the issues that may arise in production of ESI, they are not comprehensive, do not provide model document production schedules accounting for ESI production and do not account for the changing technology.
Document production schedules—existing models
Parties involved in arbitration will be familiar with standardize forms that were created to facilitate the document production process, namely the Redfern Schedule [10] and, inspired by it, Armesto Schedule [11]. Both these tools facilitate document production, as they give structure to parties’ arguments and actions in this phase of arbitration.
Redfern Schedule
The Redfern Schedule—the older of these tools—has a simpler structure. It structures parties’ arguments in the document production phase by introducing a table. Each of the columns of the table has to be completed with specified content:
- The requesting party presents the description of the document requested and the justification for the request in the first two columns;
- The responding party presents its reasons for refusal (if any) in the third column.
- The tribunal’s decision is rendered in the fourth column.
Armesto Schedule
The Armesto Schedule, proposed in 2020 [12], is more complex, with parties filling out a separate table for each document request [13]. Each table includes several subcategories, based on the IBA Rules, which the parties must fill out in the specified rows and columns (when relevant). Each subsection has a specified word limit. The table is additionally color-coded, to make it easier for its users to identify which part they are meant to fill out. The structure of the table is as follows:
- The requesting party must identify the document(s), justify their relevance to the case and demonstrate that the document(s) are not it its possession or custody is specified cells of the table.
- The responding party may reply to each of the three elements of the requesting party’s case, as well as present one or more of specified six types of objections in specified cells of the table.
- The requesting party may then reply to the objections presented by the responding party in specified cells of the table.
- The tribunal’s decision is rendered in the last column of each table.
The Armesto Schedule comes with additional documents, including the Procedural Order Model [14] and affidavits, which establish the framework for the document production process.
However, even the more comprehensive Armesto Schedule does not addresses issues that arise in the context of production of ESI.
Is it time for an ESI production schedule?
Without a model approach, tribunals must actively engage in the document production process to prevent inefficiencies, technological imbalances, and disputes over the conduct of the process.
A standardized ESI production schedule would help address these challenges by ensuring consistency and certainty from the outset of the proceedings. Using such model documents would manage expectations of the parties from the outset, potentially limiting the disputes in later phases of the proceedings over issues such search methodologies, scope of human verification, or format of production. It would also help tribunals manage issues of custody of electronic documents and data privacy concerns.
Part two of this article will propose a detailed structure for an ESI-oriented document production schedule.
[1] In almost all cases, documents executed in paper format are first produced using a computer program (making them an ESI) and afterwards they are scanned (which turns them to ESI again).
[2] See Int’l Chamber of Com., ICC Arbitration & ADR Comm’n, Report on Managing E-Document Production in Arbitration, at 12 (July 2016).
[3] This example is based on actual request filed by a party in an investment arbitration case.
[4] See Chiann Bao & Michael Hwang, 2020 Revision of the IBA Rules on the Taking of Evidence in International Arbitration, Kluwer Arb. Blog (Mar. 28, 2021).
[5] Int’l Bar Ass’n, 2020 IBA Rules on the Taking of Evidence in International Arbitration (2020).
[6] Id., art. 3.3(a)(ii).
[7] Id., art. 2.2(e).
[8] Int’l Chamber of Com., ICC Arbitration & ADR Comm’n, Report on Managing E-Document Production in Arbitration, at 12 (July 2016).
[9] Chartered Inst. of Arbitrators, CIArb Framework Guideline on the Use of Technology in International Arbitration (2020).
[11] José Antonio Moreno Rodríguez, The Armesto Schedule: A Step Further to a More Efficient Document Production, Kluwer Arb. Blog (Apr. 4, 2020).
[12] Id.
Leave a Reply