Any legal team using AI right now is navigating the same moment: the demos were compelling, the contract is signed, the tool is live, and now someone in leadership is asking a question that nobody has a clean answer to: What is the AI actually doing for us?
Not "how many documents did it touch?" or "how many hours did we save?" But rather, what finished, defensible work came out the other side, and how do we know it was good? That question is harder to answer than it should be.
The industry has spent years reporting AI value in metrics that sound meaningful but only measure activity, such as hours saved, user counts, documents processed, questions answered, and analysis types applied. The metrics confirm AI was engaged, but fail to demonstrate what it produced, or the value it delivered. It's a bit like judging a law firm by how many hours were billed rather than by whether the client is happy.
The legal industry isn't alone here. Gartner reported that only one in five AI investments generates any measurable return, and just one in 50 delivers transformational value. McKinsey's 2026 State of AI report found that ~39 percent of companies report any enterprise-level financial impact from AI. The gap isn't in the technology; most organizations struggle to measure the impact.
It's time for a better unit of measure. At Relativity, we're calling it a high-value result, or HVR.
What is a High-Value Result?
An HVR is a finished piece of professional work that a qualified human confirmed and used, and that we can trace back to the workflow that made it.
High: it requires legal or professional skills to do the work
Value: a human accepts the result based on the significance it provided
Result: the output is something auditable and usable
Here's the simplest way to picture it. For data breach response, often hundreds of thousands of records need to be screened to identify who had their personal data exposed. The AI processes the data and surfaces the likely matches. A professional confirms the findings, and they go into the final notification report. Each confirmed individual in that report is an HVR: a finished, human-verified determination that meets a regulatory standard and can be defended if challenged.
You can apply the same logic to a privilege review. A document is flagged as potentially privileged. The AI generates a prediction with supporting reasoning. A lawyer reviews it and accepts the result. That's an HVR: one defensible privilege decision, traced back to the AI reasoning that supported it and the lawyer who approved it.
The specific job changes, but the rules stay the same: AI helps complete important professional work, a qualified human confirms it, and something finished comes out the other side.
The Three Criteria of a High-Value Result
Not everything AI touches is an HVR. To count, an output must:
- Meet a professional standard. The output has to be good enough that a qualified professional (e.g., a lawyer, a FOIA officer, or another legal professional, etc.) would stand behind it as meeting their requirements.
- Be accepted by a human. Someone with the right qualifications should confirm, correct, or approve the output before it becomes an HVR. A human verifies the AI result through sampling, validation, citation checks, and QC. The human check is what turns raw AI output into something you can rely on.
- Be auditable. Teams have to be able to show the workflow the result came from, the AI reasoning behind it, and the human review that confirmed it. If it can’t be proven and defended, it doesn't count.
These criteria exist because the legal profession's exposure to sanctions, waiver risk, and regulatory scrutiny means "the AI said so" is never enough. HVRs are designed around how legal work gets defended, not just how it gets done. That’s why speed, volume, and cost savings, which are indicators of operational efficiency, not underlying value, do not make the list. Those are byproducts of value, not the definition of it.
Verify, Not Duplicate
Courts, regulators, and clients are asking not just whether you used AI, but how you supervised it. An HVR framework is your evidence of a process. Every HVR comes with a human verification event and a traceable workflow: the documentation you need to defend a privilege log, a breach notification, or a production set when the process is scrutinized.
However, verification isn’t the same as redoing the work. A human should not re-review every single AI result, as doing the work twice destroys the value created. The smart approach is to verify, not duplicate. A qualified person validates the AI results using the tools our profession already trusts, such as checking citations, measuring accuracy with statistical random sampling, and running quality control on a representative set to confirm the results hold up across the whole population. This is the same discipline that made technology-assisted review defensible in court for over a decade, and it is how you get both trust and speed.
When a human does choose to change the AI result, the AI hands them everything they need to adjust the output far faster than before. That’s what keeps the value intact.
The Insights Are Valuable Too
When Relativity aiR does its work, it shows the reasoning, citations, and considerations behind every result. While these insights are not HVRs, they do carry real value.
First, they make verification fast. When a lawyer can see exactly which passage the AI relied on and why, checking the work takes a moment instead of an afternoon. Transparency is not just an important part of AI’s work; it accelerates the verification process.
Second, they keep giving value long after the first decision. A good rationale or summary becomes a quick reference the team returns to again and again. When you revisit a document weeks later, the AI insight reminds you why it mattered and points you straight to the key facts. It is a record of the analysis that keeps paying off. So while the HVR is the headline, the surrounding insights ensure the whole process is trustworthy and reusable.
What the Numbers Tell Us at Scale
Relativity’s products, across various Legal Data Intelligence use-cases (e.g., document review, data breach response, FOIA, etc.), are producing millions of HVRs today. This year as of September 9, Relativity has produced over 620 million generative AI HVRs and we’re still counting. .
That’s millions of finished, professional-approved, traceable pieces of legal work, produced with AI. Not hours, or users, or documents processed, but real value delivered.
That number matters for what it represents, and for the direction it points. When we started tracking this metric, the central question in legal AI was still whether it could be trusted. That question is settling. The question now is how fast your practice moves from "AI is being used" to "AI-confirmed work is the new baseline."
The HVR count, and the rate at which it has grown, is the clearest measure of that shift we've found. Without a defensible unit of measure, transformation is just a word. With one, you can show exactly how far you've come, and what the trajectory looks like from here.
The Next Question: What Is a Result Worth?
Let’s be honest about what HVRs solve and what they don't … yet. Counting confirmed results answers "what did the AI produce, and was it good?" It doesn't answer "what was it worth?" A breach notification fulfills a statutory obligation. A fact chronology reshapes case strategy. Each is one HVR, and each delivers a different kind of value at a different magnitude.
That's the next dimension of this work, and we're actively working on how to quantify it, so the conversation can move from how many to how much. Even then, some value will be challenging to measure: a single AI-surfaced document can shift the entire outcome of a case, and no methodology can fully capture that impact.
We believe defining what a result is worth is a conversation the industry needs to have together. We recognize the value depends on the matter, the client, and the risk, and the professionals doing the work, like you, are best positioned to weigh it. So we are starting the conversation imperfectly, rather than waiting for a perfect answer.
What This Means for You
The legal industry is moving from “are we using AI?” to “what did AI actually produce, and was it good?" HVRs answer the second question with the rigor this profession demands.
The numbers only matter if they mean something. When they represent confirmed work, human oversight, and traceability, they become one of the most powerful things a team can bring into a room.
Customers in our early pilots immediately saw HVR data as a way to demonstrate AI value to their own clients and firm leadership. We are working toward putting more of this data directly in our customers' hands.
Every verified HVR is how we earn the right to be the AI platform for legal data intelligence. We'll be sharing more in the coming months on how HVRs are tracked across practice areas and what teams are learning as they begin to use them. We'd love to hear where you are in this conversation.
So: What's the question your leadership is asking about AI that you don't yet have a clean answer to? That might be exactly where HVRs come in.
Graphics for this article were created by Sarah Vachlon.
