Kira Built-in Provision Models
Kira’s built-in machine learning models cover due diligence, general commercial, corporate organization, real estate, and compliance, with more being added on a regular basis. They are created by our team of lawyers, who have decades of combined experience at premier law firms.
Built-in Provision Groups
Our team of experts—who practiced at large firms such as Skadden, Weil, Fried Frank, and Reed Smith, and graduated from law schools such as Harvard, McGill, and NYU—is continually working to add more models. Currently released provision groups include the following areas.
- Due Diligence: Covers over 20 provisions such as change of control, assignment, exclusivity, license grants, and notice.
- General Commercial: Includes over 25 provisions such as liquidated damages, termination for convenience, automatic renewal, alternative dispute resolution, and force majeure.
- Compliance: Export control, FCPA compliance and more.
- Corporate Organization: Covers common provisions from shareholders’ agreements such as board/manager selection, veto/approval rights, and rights of first offer.
- Real Estate: Covers over 30 commercial lease provisions such as rent, notice, sublet, description of premises, common area maintenance, parking, signage, and utilities.
- ISDA Schedules: Includes over 20 provisions such as netting of payments, credit event merger, termination currency, and multibranch clause.
Kira is the world’s most powerful and accurate machine learning contract analysis system, thanks to a unique combination of superior algorithms, developed by our in-house R&D team, and professionally trained provision models.
Just how accurate is Kira? Our standards require that virtually every built-in provision achieves at least 90% “recall.” This means that our software will find 90% or more of the instances of the provision. Our “precision” (a measure of false positives) differs provision by provision but is consistently manageable.
How good is 90% recall? Multiple studies in the field of information retrieval have found that the theoretical maximum of human review without technology assistance is around 65% recall, and with typical technology assistance in eDiscovery context, maxes out around 75-80%. See, e.g., Maura R. Grossman & Gordon V. Cormack, Technology-Assisted Review in E-Discovery Can Be More Effective and More Efficient Than Exhaustive Manual Review, XVII RICH. J.L. & TECH. 11 (2011) at 23-24.
In other words, 90% recall is probably a whole lot better than the typical junior associate at a law firm. And with Kira, 90% can be your team’s starting point, allowing you to focus your time on the remaining 10%.
All Artificial Intelligences are not Equal
Kira is built on advanced state-of-the-art machine learning technology and delivers accurate results even on new and unknown agreements. Alternative technologies include “rules-based” A.I.s, and comparison-based approaches (some of which leverage a very different form of “machine learning”), both of which have been around for several decades. These legacy approaches require a human to predict the variability of the documents and write guidelines that help the machine identify particular clauses. They can work reasonably well if you are reviewing highly similar documents or for simple provisions (e.g., provisions like governing law). However, these older technologies have proven incapable of delivering consistent results over a highly varied set of documents, and seldom come anywhere close to the precision and recall standards set by Kira’s machine learning algorithms.