Veritone and the Cold Case Foundation got together to discuss Using Veritone Assess to Analyze Decades-Old Evidence
Cold case investigations often stall not for lack of leads, but because those leads are buried across thousands of pages of unstructured case files such as decades of interview transcripts, crime scene reports, and suspect information with no chronological order.
The Cold Case Foundation, a nonprofit that partners with law enforcement agencies nationwide to help solve unsolved homicides and disappearances, is using Veritone Assess, an AI-powered data analysis tool, to speed up that process. Assess rapidly extracts actionable leads, builds chronological timelines, and identifies critical intelligence gaps from decades of disorganized case files— work that previously took investigators months to do by hand.
We spoke with Dana Boss, Crime Analysis Supervisor at the Cold Case Foundation, and Kelly Inabnett, a Senior Solutions Architect at Veritone and former law enforcement analyst, about how AI is changing the way investigators work cold cases. This conversation includes how the technology surfaces new leads, how human analysts verify AI-generated findings, and why accuracy checks remain essential to responsible AI use in criminal investigations.
Q: Dana, tell us about your background and how you got involved with the Cold Case Foundation.
Dana Boss: I’ve been a crime analyst since 2012. I started volunteering with the Cold Case Foundation in 2020 after seeing them featured in a documentary on Netflix. I was drawn to their mission, sent in an application, and that’s how I got started. I’ve been with them for almost six years now, and my role is Crime Analysis Supervisor. I oversee a group of analysts who review all the cold cases submitted to the foundation.
Q: How is the Cold Case Foundation using Veritone Assess in its casework?
Dana Boss: We’re using Assess on several cases right now. One of the big ones — I’ve actually been working on this case for about five years, and I’d already completed an initial analysis on it that ended up being roughly 47 pages. So we put the case into Assess and ran through the exact same analysis again using the platform, to see what it would surface.
The way we approach it is through what we call the ’10 filters of profiling.’ I go through each of those 10 filters and pull out all the information relevant to it. Then I go back through everything Assess surfaces and verify that it’s accurate before it goes anywhere.
Once we’ve reviewed it internally, we review it with the agency, and then they’re able to keep moving forward with their investigation. The case I’m referencing specifically is about 42 years old. The agency has been actively investigating it that entire time, but this gave them some new leads to go out and follow up on.
Q: What did the manual process look like before, and how has that changed?
Dana Boss: Before, everything went through manually. I’d be working through it all on a computer, and the files don’t always match up chronologically the way they should. Nothing comes to us in order.
Putting the case into Veritone Assess, though, the platform is able to make those connections and put everything into a logical order — especially across all the different filters we use. Those filters include things like victimology, which covers everything pertaining to the victim; three different filters related to the crime scene itself; modus operandi; offender signature, meaning whether the offender is organized or disorganized; and general suspect information.
Before Assess, I’d have to go through every individual report myself and manually extract that information — and none of it comes pre-categorized. I’d be taking notes as I went: on this date and time, this detective or officer interviewed this person, or identified this person as a suspect. Then I’d have to categorize all of that myself afterward.
With Assess, I can simply ask it to give me all the information about the victim, and it goes through every single report and pulls that information out for me automatically. All I have to do at that point is go back through and verify that the information is accurate.
Kelly Inabnett: That really resonates with my own background. I spent my law enforcement career working human trafficking, crimes against children, and sexual assault cases — a lot of manual review of reports, social media, and digital evidence, which leads to a lot of burnout.
Now at Veritone, I bring that experience into how we apply AI to law enforcement work, thinking through what would have made my own job easier. Dana’s example is a good one: normally you’re working through 20 or 30 reports with a legal pad, writing down everything related to victimology. By the time you hit report 19, you vaguely remember something similar in report 3 or 4 and have to flip back to check. AI doesn’t have that lag — it remembers exactly what’s in report 3 or 4 and can flag when two things match or contradict each other.
Q: How did Veritone and the Cold Case Foundation start working together, and what does that collaboration look like day to day?
Kelly Inabnett: We were introduced through the Cold Case Foundation, Butch [Rabiega, AI Program Director at the Cold Case Foundation] was our primary contact initially, and Dana was brought in as someone more immersed in current investigative practice and more hands-on with the technology. I’d go in on the front end to make updates, and Dana would show me the before-and-after results so we could adjust the system to fit what analysts actually need. It’s really a case of making sure AI is a tool, not the driver — being an “AI driver” rather than an “AI passenger.”
Dana Boss: What’s nice about this setup is that I’m the end user giving direct feedback, not an engineer guessing what will work. If something doesn’t work for a case, or I need a specific feature, I can tell Kelly directly.
For example, I have a case that may have been perpetrated by a second suspect, but I don’t want to analyze the two cases together initially — I want to analyze them separately first, then bring them together. Being able to talk through that kind of workflow with someone who understands both the analysis and the platform means the product actually works in practice, not just in theory.
Q: Can you talk more about that specific case — analyzing two potentially related cases separately before combining them?
Dana Boss: We’re still working through that one. I’m analyzing the cases separately for now, but the goal is to eventually bring the analyses together and determine, one way or another, whether they’re connected. I don’t want to force a connection that isn’t there. We’re in the early stages, so I can’t speak yet to how well it works in practice — but in theory, it should work.
Kelly Inabnett: To put the time savings in perspective: Dana’s first case took roughly six months to analyze manually and produced comparable results in the platform within an hour or two. If we apply that same ratio — six months for one case, six months for another, then some additional time to combine them — you’re talking about roughly a year and a half of work using traditional methods.
Using Veritone’s platform, that becomes a matter of hours. Whatever the final numbers are for a specific case, the larger point is that the analysis itself, the “grunt work,” gets compressed dramatically — which means investigators get to the real work, deciding on follow-up actions, much faster.
Q: AI isn’t perfect. How do you handle accuracy and human oversight?
Dana Boss: I have a good example of this. In one case, I asked the platform for a list of suspects, then asked it to prioritize who should be followed up on first. It gave me three names that weren’t listed in the reports at all. At first I thought I’d simply missed them — these were names I hadn’t seen before. When I asked where it found that information, it acknowledged the names weren’t actually in the case file and corrected itself. That’s exactly why human review matters: without a case expert checking the output, you risk sending investigators to chase leads that don’t exist.
Kelly Inabnett: Even if we built an engine that we claimed was 100% accurate, I’d still expect the investigator to verify everything — that’s true of any evidence, AI-generated or not. As the case expert, you’re the one who verifies whether something did or didn’t happen. Mistakes happen, AI isn’t perfect, and neither are people — that’s why the checks and balances matter.
Q: There’s a lot of skepticism about AI in law enforcement. What’s your take?
Dana Boss: There’s a lot of “bad juju” about AI out there, but as long as it’s used professionally and in compliance with agency policies and standards like CJIS [Criminal Justice Information Services], it needs to be part of the job. This is the direction the world is going. Defense attorneys are already starting to use it, and if law enforcement doesn’t adopt it responsibly, we’ll be playing catch-up. That’s historically been a weakness in law enforcement — we’re slow to adopt new technology, and a hard stop on AI just means we get left behind.
Kelly Inabnett: The people committing these crimes are already using AI to plan them, so it has to be part of the response too. There are good drivers and bad drivers when it comes to using AI — the goal is to educate people on how to use it properly, with the right checks and balances, rather than taking output at face value. If people understand that they’re still the ones processing the information and making the final decision, it changes the conversation from “AI is coming for my job” to “AI is helping me do my job better.” Otherwise, we’re going to be drowning in digital evidence without it.
Want to learn more about Veritone Assess? Visit here to read more and watch the demo. Don’t be afraid to reach out directly if you want to start a conversation.
Further Reading:
The Future of Evidence Management: AI Solutions for Law Enforcement
Overcoming the Top Challenges of Digital Evidence Management
AI and Privacy: Balancing Technology and Compliance in Law Enforcement
