In this technology-driven world, the use of mobile phones (for everything really!) continues to skyrocket and it is not limited to our personal life only but due to the global pandemic where workplaces became virtual (or hybrid), usage of mobile phones for business working also increased significantly. Checking emails, virtual meeting via Zoom or GoToMeeting, reviewing or editing business documents, and business communication (messaging and phone calls) all have been done by employees via their mobile phones.
That is the reason data collected from such mobile devices is increasing in current litigations. Mobile data has plenty of multimedia files (i.e. photos, audio, and video) and chat/text files as well. Therefore, specialized tools and techniques should be in place to collect data from mobile devices in a forensically sound manner. Chat/text files and native formats can easily be ingested, processed, and reviewed, as these are searchable files. One can easily cull down data by using relevant terms but photos or images are not generally word searchable in the traditional sense. Therefore, litigation teams can’t always use them as easily or effectively, even as photos become crucial evidence or game-changer in some cases eg IP Litigation.
Secondly, even if we talk about non-mobile data collected for general eDiscovery, it also has images or photo-based evidences especially in Infrastructure, IPR, or Civil disputes. Product design or blueprints, BIM designs, and drawings are in image formats most of the time. All these image-based data are in GB or TB volumes in some cases. Manually reviewing this data could be time-consuming and costly. Here technology can be your valuable resource slashing down the attorney review time and ultimately cost.
Let’s imagine a scenario
An energy regulator notifies an organization for breaching environmental regulations during a drilling operation. There are 15 custodians in scope whose computer and phones data are being collected as ESI. The data collected, of around 180 GB, consists of email, chat backup, documents, and photos. Out of the total volume, 35 to 40 percent of data were multimedia files, the majority of them are photos collected from individuals’ phones on the drill site. These photos are of everything but the regulator is seeking photos in which engineers can be seen working on the pipes and drilling.
Manually reviewing the image dataset is time-consuming and expensive as well. For example, if we consider 400 photos per GB then reviewers need to go over a total of 28,800 photos. Let’s assume a speed of 2 photos per minute for manually reviewing and labeling those photos and marking them responsive/non-responsive. It might take at least 240 human hours (approx. a whole week for a team of 6 attorneys.)
On the flip side, if we leverage machine learning-enabled Image Analytics in this scenario it only takes a few hours. Firstly, Image Analytics identifies similar or duplicate images and trims down the data set accordingly. Thereafter ML-powered algorithm uses object detection techniques to recognize objects, scenes in the images/photos (in this case drills, pipes, and water). Based on their analysis, the Image Analytics engine adds a label to each image. There can also be multiple labels added to the images. For example, if there is a photo in which four engineers are working together on the drill site. It adds labels like hardhat, drill, pipe, etc.
Once we have labels ready on each image, we can easily filter them out using relevant search terms. It also shows an accuracy level with each label added by the system. From our example above… let us now assume there are around 8,640 images (30% of total volume) that hit/qualify from our search terms (eg hardhat, drill, pipe). We can further trim down the volume by executing advanced searches (i.e. images that have two labels named “drill” and “water”). Even if we push all 8,640 filtered images for a manual review then it will still cut down review time to 72 hours because the images have been culled and only need to be tagged responsive or non-responsive, no need to add a label because it’s already done by ML-powered Image analytics.
Watch this video to learn how Knovos eDiscovery Image Analytics works
Isn’t it worth considering? Especially when the pressure of cutting-down eDiscovery costs and improving service delivery is higher than ever for law firms. PS it’s not just images, even copyright infringement type of issues can also be investigated with Image Analytics by searching usage of logos and designs within a variety of documents and files.