How Balancing Automation and Talent Reshapes the Flow of FOIA and Public Records Requests
A Conversation with Michael Sarich, former Director of FOIA at the Department of Veterans Affairs.
Public records and FOIA requests are undergoing a profound transformation in response to the combination of a downsized workforce and the seemingly unstoppable rise in information being requested.
To answer this challenge, the agencies responsible are exploring emerging technologies and how best to integrate them into their workflows. By blending human expertise with purpose-built AI tools, forward-thinking organizations are setting new performance benchmarks and transforming how records requests are answered.
In this first of a two-part series, Michael Sarich, former Director of FOIA at the Department of Veterans Affairs and a 2025 Fed100 IT Winner, offers guidance on how to navigate this evolving landscape.
What have been the most significant changes to the public records requests process that you’ve seen in recent years?
Three things are shifting in tandem, and the first shift is a shrinking workforce. The Deferred Resignation Program pulled out a layer of senior practitioners who carried institutional knowledge that doesn’t transfer well.
The second shift is that, while request volume keeps rising, tools keep advancing. AI capabilities that were demos in 2024 are now FedRAMP-authorized, in production, and changing what’s possible inside the agencies that know how to deploy them.
The third shift is the one reshaping the field. Two years ago, the conversation about AI in FOIA was forward-looking. Today the conversation is divergent, it’s a growing distance between the agencies operationalizing the new capabilities and the ones still treating them as a future-state problem.
The fast movers are setting a new performance floor for cycle time, defensibility, and requester experience. The agencies that haven’t engaged are not just standing still, they’re falling further behind every quarter the commercial AI market advances. That divergence is the most consequential change in federal FOIA since I started in this work. It’s also the most fixable, because the tools are now in procurement reach for any agency willing to make the call.
What are the blockers holding agencies back from adopting these new technologies? What’s stopping them from making that call?
There’s a lot of AI anxiety in this space. You’ve got to understand that most leadership has made it through a 15- or 35-year career in their agency. They’ve seen a way that has worked for that agency for a long time, they’ve built on that system, and this new technology is disruptive.
It’s not dissimilar to horses and automobiles. They aren’t complementary, no one is hitching a horse to the front of a car. For some folks that’s challenging, it can be frightening to be on the bleeding edge.
I think one of the challenges with AI is that it’s moving so quickly that it’s hard for people who aren’t in it every day to figure it out. You know, bison are famous for marching into the storm, because they know it’s moving past them. A lot of workers in the public sector are not pushing forward, they’re staying put.
Looking back, can you speak more on the ways you have seen LLMs affect the handling of sensitive data?
The 2024 prediction held, but it held faster than I expected. Closed-environment LLMs are now operational in federal FOIA workflows, not theoretical. Search and retrieval that took hours now takes minutes. Summarization of large document corpora is reliable enough to be the default starting point on big productions. Response letter drafting is largely automated for routine work. Deduplication and clustering surface patterns that no human reviewer is going to find by hand in a hundred-thousand-document response.
For an agency processing at federal scale, that’s the step change the community has been talking about for a decade. It just arrived.
Where the work is still maturing is at the defensibility frontier. Privacy Act and PII redaction remains human-in-the-loop, but the loop is tightening faster than most agency counsels appreciate.
Confidence-scored redaction with reviewer oversight is closer to operational than it was in 2024, and the agencies piloting it are getting comfortable with workflows that would have been unthinkable two years ago. The next two years will determine how far that envelope expands.
My bet is further than the conservative consensus expects.
The shift I did not anticipate clearly enough in 2024 was the upstream effect. Federal employees are now using LLMs to create the records that will eventually be FOIA-ed. That changes the records management problem before it ever reaches the FOIA shop, and it makes the case for modern, AI-aware platforms across the records lifecycle, not just at the FOIA tail end.
The agencies thinking about this as an integrated lifecycle problem are positioning themselves correctly. The ones still treating FOIA as a downstream cleanup function are missing where the real leverage is.
You mentioned the upstream effect of AI use in the FOIA and public records request process. What other unexpected areas do you think AI might, or is already having, an effect?
The downstream effects are around downstream governance, making sure it’s being used ethically, and that its use is directly supported by the law. Defensibility is so important, being able to trace back to the actual core of why you made a decision, you have to be conscious of it as a practitioner.
Because the public has an inherent right to government transparency, agencies should be prepared to justify the decision to use AI at any level of work and address whether the tools used have any downsides, like an algorithmic bias for example.
That’s why companies like Everlaw are so important, because the technology isn’t new to them. Everlaw understands the long tail of AI-powered technology, and candidly speaking, they understand it unlike anyone in this space.
For smaller agencies catching up with new technologies, where should they begin integrating automation?
Three steps, in order. First, map your intake. Where do requests come from, in what volume, by what category, and where do they get stuck? The agencies that have done this groundwork are already deploying tech effectively, because they know what problem they are buying a solution to.
Second, get the unglamorous infrastructure right: searchable email archives, a single system of record, defined SOPs that survive turnover. Modern AI tools deliver dramatically more value on top of a clean records environment than on top of a messy one, not because the tools are weak, but because the tools amplify whatever they are pointed at.
Third, build a coalition with IT and senior leadership before procurement. The smaller agencies winning at this are the ones treating FOIA modernization as an enterprise investment, not a side-office project.
For agencies eager to adopt AI, what foundational work should come before the technology?
Four things unlock the full value of AI in FOIA and public records requests. Each one is paired with a capability it makes possible, but skipping any of them doesn’t block AI deployment, it just leaves capability on the table.
Email archives should be complete and searchable, record retention should be enforced rather than aspirational, and there should be a defined system of record for every communication channel including Teams, Slack, and whatever comes next.
Modern semantic search across a clean records environment is the single biggest productivity gain in FOIA processing. Across a fragmented one, the same tool delivers a fraction of the value.
Second, classification consistency. Does your staff apply exemptions the same way? Are decisions documented and reviewable? That consistency is what unlocks predictive coding and exemption-likelihood scoring.
When the underlying decisions are coherent, the AI learns the agency's posture and accelerates it. When the underlying decisions are inconsistent, the AI surfaces the inconsistency, which is actually valuable diagnostic information, but it’s not the production gain agencies are looking for.
Third, governance. Who decides what gets withheld, on what basis, and how does that decision survive personnel turnover? That governance layer is what unlocks defensible automation, the workflows where AI moves work forward with documented human oversight at the decision points that matter.
Without governance, even good AI output can’t be defended in litigation. With it, AI accelerates work that is already defensible by design.
Fourth, the right metrics. Cycle time, age of backlog, defensibility in litigation, requester satisfaction. Volume is a vanity metric. Modern tooling produces telemetry on all of these in real time, but only if the agency knows what to measure. The metrics layer is what lets leadership see what AI is actually doing, which is what convinces them to keep funding it.
Get these four right, and AI delivers a step change in operational performance. Skip them, and AI still delivers, but it delivers compounding errors and unmeasured outcomes. The tools are the leverage point. The foundational work is what turns the leverage into outcomes.
Under your leadership, the VA achieved a 99.4% customer satisfaction rate. What steps did you take to accomplish that endeavor?
We treated requesters as customers rather than adversaries. That sounds obvious, but it’s not how most FOIA shops operate by default. Practically, it meant acknowledging requests within hours, not weeks. Communicating proactively when scope was unclear, before processing time was wasted. Calling back when the response was complex. The single biggest source of FOIA dissatisfaction is not denials, it’s silence.
Most requesters can accept a "no" if they understand why. What they can’t accept is being ignored for nine months. Closing the silence closed most of the satisfaction gap.
Second, we got serious about scope dialogue at intake. If a requester asked for "all records related to" something, we didn’t process that for six months and then dispute scope on the back end. We called, narrowed, documented, and processed what they actually needed. Faster for them, faster for us, and the result was the kind of response that builds trust rather than triggers an appeal.
While at the VA, you also reduced its FOIA backlog from 20.5% to 0.5% while quadrupling annual request volume from 23,000 to 119,000. If you had to start that process from scratch today, what would you do differently?
If I were re-running the VA backlog reduction today, I would compress the timeline meaningfully, not by skipping the relationship and training work, which is still where success is built, but by deploying tools that did the heaviest lifting for me on volume.
I would also invest harder upstream of FOIA, in records management itself. FOIA can’t be better than the records it processes, but the tools to operationalize good records management now exist. Findability by design is no longer a multi-year infrastructure project. It’s a procurement decision and a governance commitment.
The most leveraged dollar at any agency is the one spent on the records environment, and that dollar now buys more capability than it did even three years ago. That’s the lesson I would re-learn faster.
What, or who, benefits the most from automation in public records requests?
The requester benefits first. Faster, more accurate responses are the public's interest in transparency made operational, and they are increasingly the baseline expectation. Every other digital interaction the public has, from banking to healthcare to government benefits, is moving toward responsive, transparent, real-time fulfillment.
FOIA cannot remain the exception. The agencies modernizing now are matching a customer experience the public already expects.
The processors benefit next. Fewer demoralizing repetitive tasks, more time on the judgment calls that actually require expertise. Modern tools change the FOIA job from a clerical role with legal moments into a legal role with operational support. That’s a recruiting and retention story, not just a productivity story.
The agency leadership benefits third, because a defensible posture in litigation is a function of the workflow, not a press release.
The least obvious beneficiary is the agency mission itself. FOIA backlogs erode public trust. Eroded trust costs agencies political support. Lost political support costs agencies budget and authority. Modernizing FOIA is one of the few cost-cutting investments that pays back into the mission, not just into compliance.
This is part one of a two-part interview. In part two, Sarich provides more insight on cultivating leadership and security for the future of FOIA and public records requests.
To learn more about how Everlaw can help government agencies speed up response times for FOIA or Public Records requests, request a demo today.
