Posted by Michael Eichsteadt, vice-president of engineering at Legal Futures Associate iManage
Eichsteadt: Simple data retrieval is no longer sufficient
Search has long been understood as data retrieval – the ability to call back information and check a box on finding something.
Legal professionals today need more of a 360-degree view on a matter, including who’s involved, how new information impacts strategy, relevant institutional expertise, and other key contextual information that could affect outcomes.
This need is driving a shift toward extracting context from data, making it efficiently available to AI systems to turn raw information into new knowledge – and vendors increasingly need to deliver against this value shift.
The search stakes
Mechanical ‘search’ was always built on a flawed premise: that users know what they’re looking for.
In legal work discovery, cross-referencing and information interaction are what leads to the strategy on a matter. A lawyer preparing for litigation doesn’t just want the document they remember – they want everything relevant that they might have forgotten, overlooked or never known existed, and compared within the context of the new matter and circumstances.
For lawyers, a false negative is far riskier than a false positive. They would rather wade through 25 results knowing nothing was missed than receive three clean results and wonder what wasn’t returned.
This is the fundamental tension at the heart of search in the legal world: incomplete results aren’t merely inconvenient – they can have massive consequences for the quality of service that legal professionals are able to provide.
AI brings another layer of complexity
With AI in the picture, the stakes get higher. In our new, AI-enabled landscape, search is no longer just an intermediate step for legal professionals on the way to performing some other task in their workflow. AI systems are now the consumers of retrieved information.
It used to be that a legal professional would find a document, read it and then act on what it contained. AI is now offering to pick up that load – to synthesise, analyse and act on information in a semi-autonomous way.
This means the demands on retrieval have intensified: users still need search to be right, and on top of that, they want AI to do something meaningful with the results.
This is where the distinction between data and knowledge becomes critical.
Data is what sits in the document management system (DMS): potentially millions of documents and emails combined with firm precedent, patterns of working and institutional expertise, extraneous information, discarded material, revisions and drafts.
Knowledge is what you can actually extract from that data, and the value is how you put it to use.
The challenge of context extraction
Context extraction – the process of pulling the genuinely relevant material from a vast sea of documents, emails, and other files – is harder than it sounds because there’s an objective and subjective nature to context.
For example, if you ask AI to summarise a document, you might accept a dozen different versions of that summary. The output is inherently a matter of perspective.
Move into more complex professional territory and it’s a completely different story. The specificity of legal work – who is involved, what precedents apply, what a particular partner’s preferred approach has been, prior outcomes – demands context that is both accurate and tailored.
There is also the sheer infrastructural challenge of context extraction. Asking an AI to sift intelligently through enormous volumes of data – making probabilistic determinations in real time about what matters – is computationally expensive and architecturally demanding.
What vendors must do
For vendors, this need for context demands a response. The question is how to help AI make that data usable in an efficient manner – how to stitch together a quilt, so to speak, from thousands of individual patches to create a ‘context fabric’.
One part of the solution is enrichment: using AI to identify and extract relevant metadata from documents and populate structured fields within the DMS, so that end users or agents performing intelligent action can access important information without having to scan deeply into every file.
(Deep document scanning is expensive and time-consuming – and structured, enriched metadata makes retrieval faster, cheaper and more reliable).
The harder and more important challenge is that end users still do not always know what is relevant in a large document corpus. Well-governed, semi-autonomous agents – working within appropriate guardrails – can help identify relevant information across a body of documents, surfacing what matters even when no one knew to look for it.
In other words, vendors shouldn’t approach context extraction as a matter of returning what users already know exists.
They should focus on helping professionals discover what they didn’t know they had – the knowledge hidden in years of work product, and the connections no individual would have thought to search for.
The next era belongs to the knowledge-ready
Technology must do more of the work when it comes to unearthing knowledge and the valuable context it can provide because simple data retrieval is no longer sufficient.
The firms and vendors who recognise this shift – and build toward it – are the ones who will be able to drive better business outcomes in the AI era and turn their knowledge into an enduring competitive advantage.
Leave a Comment