It’s easy to assume that if you want to reduce the total cost of review, then you should try to reduce the various line item costs that go into eDiscovery and Review. However, the biggest economic driver in reducing the total cost of review is finding technology-based approaches to reduce the total number of documents reviewed. This approach generates significantly more cost savings than trying to negotiate line item discounts.
I wanted to step back and illustrate how those economics work. Various pieces of the process fit together to significantly impact the total budget. I’m going to take a standard, illustrative scenario and outline the biggest sources of cost savings from more strategic thinking at the outset.
The Baseline BudgetIn a traditional linear review, up to 70% of the money is spent on the labor of reviewing the documents. In this traditional approach, documents are collected and culled based on date ranges and search term filters. Then, every document in the culled data set is manually reviewed for responsiveness.
Of that $100k budget, 70% of the budget spent on review, a further 80% of the effort is spent eliminating non-responsive documents. To put that in context, for a $100k case, $56k is spent doing work that adds little to no value to the merits of the case.
This is a common, widely-recognized problem, and industry experts are trying to take steps to mitigate this “waste”. A common solution is attempting to reduce hourly review rates by offshoring the work to reduce the total cost of review.
However, there are limitation to that approach, as it does not address the underlying issue. Quite simply, reducing the “cost per document reviewed” is not as effective as reducing the “total documents reviewed” by using more strategic planning with an experienced advisor in the beginning of the review process.
In fact, the biggest economic driver in reducing the total cost of review, is finding technology-based approaches to reduce the total number of non-responsive documents reviewed.
Let me walk you through the economics, with some examples.
Example 1: Save $24.5k By Using Artificial Intelligence to Further Narrow the Document Set Before Review Begins
The fundamental challenge with managing the cost of eDiscovery reviews is the low responsive rate of documents. In our scenario, only 4% of all the documents collected are responsive. That means that out of 400,000 documents collected, only 20,000 are relevant.
A common approach is to use search terms to narrow the document set. In our scenario, that means that of those 400,000 documents collected, 100,000 still need to be reviewed. At a fully burdened cost of $0.70 per document, that still means that $70k is spent on review.
AI, combined with professional research services, can be used at the beginning of the matter to further refine search terms, eliminate data sources and narrow custodian lists. When that service is used, the reviewable document set can be reduced by 35%. So, instead of reviewing 100,000 documents, only 65,000 documents are being reviewed. This approach results in a net savings of $24.5k.
Example 2: Save $9.5k By Using AI to Reduce the Cost of Poor Quality During the Review
In a typical review, there are first pass reviewers, and second pass reviewers. The first pass reviewers look at the documents in bulk. While the second pass reviewers are more sophisticated reviewers, and randomly sample the first pass reviewers’ work to ensure accuracy. The first pass reviewers are typically billed out at lower rates, for example $0.50 per document. While the second pass reviewers are typically billed out at higher rates, for example $1.00 per document.
Take the example of the 100k documents that are being reviewed offshore. $20k of the labor cost is due to quality control issues. With more strategic use of the technology, that number can be reduced to $11.5k.
This is due to a couple factors. First, about 10% of those documents are going to need to be reworked, because the first-pass reviewer won’t meet quality standards. Simply, an inattentive reviewer might code every document the same or, perhaps, missed instructions provided by the review manager. So, 10k of those documents need to be reviewed twice, costing $5k in rework. In addition, 15% of the documents need to be sampled by a subject matter expert to ensure that the review is being completed accurately, costing $15k in quality control measures.
Using AI to guide the review can drive significant efficiencies in the review operations. First, “bad reviewers” are identified more quickly, so instead of potentially having $5k in rework, you could only have $2k. Second, because the AI is actively monitoring and alerting review managers to the front-line reviewer issues, the sample rate of second pass reviewers can be reduced to 7% instead of 15%. This reduces second pass reviewers’ cost by $8k.
Successfully identifying problems early on lets you remedy them immediately, saving expensive rework and reducing overall QC costs by $10k. These features typically come at no additional cost, so net savings are $10k.
Concluding Thoughts – Reduction in the Total Cost of Review Using AI Is $31.6K (32%)Using the two techniques outlined above, proper utilization of AI can collectively reduce the total cost of review by 32% This is substantial savings that can either be reallocated towards more valuable legal advisory work or can be used as a cost reduction measure in managing litigation expenses. While this methodology does not completely eliminate the need to review some non-responsive documents, it greatly reduces that portion of the review and the associated costs.
Acorn’s been a leader in AI and review process design for 7 years. A side benefit to using all this AI is that it requires smaller review teams, which makes staffing easier. The goal isn’t to work more cheaply with lower quality service; it is to work smarter, which in turn, reduces the total cost of review.
AI Trains Your Model to Be Even More Accurate over Time – Savings Tomorrow, AND TodayIf your firm or corporation works with one or two litigation models—maybe you do all IP or all medical malpractice litigation—you probably end up going through documents, data and even fact patterns that look similar, even as your players and exact circumstances change, or you may, in fact, have the same custodians on multiple matters.
The right AI eDiscovery system lets you reuse your existing work product on future cases, without comingling data. Once your AI model is properly trained, it can be applied over and over again to different cases, strengthening your model and improving its results each time, without the costs of starting over on every new case. Instead of retraining your team every few months, you can rely on the AI model to keep building on its own training.
Be Sure to Follow Me for the Latest Content and Subscribe For the Latest Acorn Insights!