Case insights now arrive up front in document review, but human oversight still decides the outcome
The most disruptive thing artificial intelligence has brought to e-discovery is not speed or cost – it is timing. For decades, litigation teams learned what their documents contained only at the end of a review, after the budget was committed and the strategy was set. Generative AI has inverted that sequence, says Tiana Van Dyk, managing director of Epiq Canada in Calgary. Canadian Lawyer spoke to Van Dyk about Epiq’s expansion into data management, analytics, consulting and AI implementation in 2023.
"The biggest shift now is the fact that those insights are available to you at the very beginning, which entirely shifts the way that we can plan around our cases, how we can deliver strategy to our clients," she says.
That reordering – insight first, review second – is what she finds most consequential about a technology the profession is still learning to trust. It also lets teams flag scope creep, a missing custodian or missing data at the outset, not at the end of a review.
Where AI is genuinely delivering
The clearest wins are in document review, where sheer volume has always been the enemy. Generative AI classification "took CAL and TAR to the next level," Van Dyk says, pointing to the technology-assisted review (TAR) and continuous active learning (CAL) methods firms have leaned on for years. In plain terms, technology-assisted review uses software trained on lawyers' own relevance decisions to sort large document sets, and continuous active learning refines that model as review proceeds, steadily surfacing the most relevant material first. Because generative AI’s output can be validated with the same precision, recall, and elusion testing that lawyers already know, adoption has moved faster than she expected after the slow climb it took to make TAR routine. In Everlaw’s “2025 Ediscovery Innovation Report,” generative AI use among e-discovery professionals reached 37 percent, up from 12 percent two years earlier.
Van Dyk has watched the field reinvent itself many times over. "We've turned the tables 10, 15 times in what that reality of e-discovery actually looked like," she says – part of why high-volume litigation now mirrors the kind of firm-wide generative AI adoption reshaping document-heavy practices.
A tool in the kit, not a replacement
Van Dyk is blunt about the hype. "I think that it's more important to think of this as an additional tool in your toolkit than necessarily a replacement," she says. TAR and CAL still suit certain reviews better depending on the responsive rate and the type of data, and generative AI can be layered on top rather than swapped in.
When the technology underdelivers, she argues, the cause often lies with the user. "Your expectations have to be tailored to what the tool is built for … it's not the tool, it's our perception," she says, adding that the trouble starts "when you try to take a purpose-built tool and apply it to something it's not made for." She uses a kitchen metaphor to make her point: "When you are baking something in your kitchen, you cannot only use your oven. You have to have mixed things," she says.
From provider to partner
The way Epiq Canada works with clients has changed as quickly as the tools have. "One of the things that we often talk about is the partnership versus the provider," Van Dyk says. The old norm was transactional: a firm outsourced a discrete task and received a finished product back. "What we have now is much more embedded partnership," she says, with shared goals and far more back-and-forth across the client's team.
That model spans the market. "We absolutely work with single practitioners. We work with the biggest law firms and some of the biggest corporations in Canada," she says, including merger-and-acquisition matters and Canadian Competition Bureau work. Its Epiq Discover platform puts collection tools, AI, TAR and CAL within reach of smaller shops that were once priced out, which is where Van Dyk sees the clearest line to access to justice. "We're most successful when our clients are successful," she says.
Responsible adoption and the overconfidence trap
The harder conversation is about competence. "There's an obligation in the legal field to be proficient with technology," Van Dyk says, and that duty has not changed since the TAR and CAL era – it has only grown more complicated. Where a statistical score once forced lawyers to dig into what a number meant, today's systems explain themselves in fluent prose. "The human-looking nature of it makes people be overconfident," she says.
Her answer is human oversight and honest self-assessment. "The onus is on you to become competent or to be surrounded by people whose opinions you can rely on as experts," she says. Not every lawyer needs to grasp complex retrieval protocols, in her view; they can bring in an expert, as courts have always done – a stance consistent with the profession's broader caution about the limits of AI and where legal judgment still rules.
What clients now expect
Clients, meanwhile, assume their lawyers already use AI, partly to push bills down as the profession drifts "away from the billable hour." That demand can outrun the technology. The request to use it keeps "going up and up and up and up and could potentially push the boundaries of what technology actually does," she says. The job, then, is to set limits. "We have to do it well, and we have to provide the guard rails because the truth is clients don't understand what it's used for," she says.
Epiq is taking part in the Canadian Legal Summit in Toronto in October, where Van Dyk wants practitioners to leave with a plan instead of unease. "AI is not all or none, and it's not 100 percent risk," she says. The risk of asking Copilot to summarize a week's calendar is not the same as that of running a generative AI document review, she notes: "It's completely manageable risks as long as you understand what you're doing," she says.
This article is based on an episode of CL Talk, which can also be found here:
The episode is also available on our CL Talk podcast homepage, which includes links to follow CL Talk on all the major podcast providers.
