Why intake and the front door is the most important AI investment for in-house legal teams today

(This article was published in the Artificial Intelligence category of Legalverse Media in April 2026)

Introduction

When I started my Masters at Carnegie Mellon, I chose to focus my thesis on legal AI due to my past experience in the field. The goal felt clear: make AI perform better on legal text. There were multiple challenges unique to legal AI that didn’t exist when working with general English text: sentences were long, vocabulary was specialized, documents were enormous, and analytical reasoning was needed to draw inferences across hundreds of pages.

As the field progressed, a lot of those problems started to get addressed. General advances in AI lifted the floor for everyone. Models got larger, context windows expanded, and reasoning capabilities improved dramatically. Legal-specific research from companies and academic groups pushed the boundaries further still. For a moment, it felt like the hard part was nearly behind us.

But the more time I spent with in-house legal teams, the more a different problem came into focus. The bottleneck was often not the model’s capability. It was the completeness of the request being made and the quality of context and precedent available for the AI to draw on.

In most legal departments today, both are in poor shape. Requests arrive stripped of the context that would make them actionable, and past work is scattered across inboxes and shared drives, unstructured and largely unsearchable. Everyone is racing to sharpen the knife, but nobody is asking whether the ingredients on the cutting board are actually good enough to make the dish.

How work actually lands today

In the early days of customer interviews, one question we always asked was: walk us through what happens when a request comes in. The answers, almost without exception, followed the same pattern.

A sales rep fires off a message to the legal team by email, Slack, or Teams: “Hey, need an NDA reviewed for a prospect, can someone take a look?” A PDF is attached. Nothing else. The lawyer picks it up and immediately has to go back: What data is being shared? When does the deal need to close? What is the deal value? The sales rep responds a few hours later, but their answers raise other issues: the counterparty is based in the EU, which brings additional compliance questions into the picture. Another exchange.

By the time the lawyer had everything they needed, a day or two had passed without a single minute of substantive work done. The problems we kept seeing were the same:

• Fragmented context: Information was collected through back-and-forth threads, scattered across channels and inboxes

• Time lost upfront: Days were lost before work even began, gathering information that could have been captured at the outset

• No institutional memory: Nothing was recorded in a structured way, so when a similar request came in later, the lawyer was starting from scratch all over again
  
  

We went in looking for friction in the substantive work itself. What we found was that the bigger problem came before any of it had even started.

The investment that compounds

This is where focus shifts from operational to strategic.

When a request comes in through a well-designed intake system, the context is collected upfront, all of it, in one place. The counterparty, the request type, the business unit, the jurisdiction, the urgency, the specific concern. The request is triaged and routed automatically. The work done to resolve it, the approach taken, the outcome, and the time it took are stored in a structured way. Multiply that by hundreds of requests, and what you have is no longer just an organized inbox. It is a dataset. The raw material that AI actually needs to work with.

That dataset starts to change how work gets done. Take a vendor agreement that comes in for review. Because a dozen similar ones have been handled before and stored properly, the system can surface how they were approached, what positions were taken, and what the final redline looked like. The lawyer is not starting from scratch. The AI tools working alongside lawyers keep getting sharper because they have real, structured precedent to draw on rather than filling gaps with guesswork.

The benefits don’t stop at what’s possible today. As AI solutions in the legal space keep evolving, the organizations that will get the most out of them are not necessarily the ones with the biggest budgets or the most sophisticated tools. They will be the ones with the richest institutional knowledge underneath. Every new AI capability that emerges, whether it is a smarter copilot, a better research assistant, or an autonomous drafting tool, will have more to work with, and will perform meaningfully better, because the foundation was built early.

What a good intake system actually looks like

Intake only delivers on its promise when it is built in a way that people actually use. That sounds obvious, but it is where most implementations fall short. A senior legal ops practitioner captured it well at a conference I attended: “We have a front door implemented, but requests still come in through the windows, the chimney, and the roof.” And it does not always stop there. We have heard from legal leaders who, a few months after rolling out an intake tool, quietly told their employees to just go back to emailing the legal department directly because the tool was creating more confusion than it solved.

The system failed not because the technology was wrong, but because it asked too much of the people using it.

A good intake system does the opposite:

• It lives where business users already work. Teams, Slack, email.

• It identifies missing context and collects it automatically. If a request comes in incomplete, the system follows up with the requestor to gather everything that is needed before the request ever reaches a lawyer.

• It gives the requestor visibility. Automatic acknowledgment, clear ownership, and status updates without anyone having to chase.

• It routes requests automatically based on what was captured, so the right person picks it up without anyone manually triaging.

• It surfaces relevant precedents before the lawyer even opens the file.

• It creates structure in the background without the business user ever having to think about structure.
  
  

Investing in the front door is investing in the quality of every single thing that follows. The teams that understand this today will find they are creating something genuinely valuable: a legal department where AI has real context to work with, where institutional knowledge compounds over time rather than getting lost, and where every new AI capability that emerges has a foundation on which to build.

The front door is where that starts.