Discovery Isn’t Sequential. So Why Is Your Workflow?
Our profession is redrawing its map as we speak. EDRM 2.0, the first real update to the reference model in over a decade, is in public comment through the end of July, and it is thoughtful work by people who care deeply about this field. But a reference model has one job: documenting the consensus about how the work flows. It can’t tell you why the work flows that way. It isn’t supposed to.
Two decades of process thinking have trained all of us to describe discovery by its stages. Ask instead what discovery is for, and the stages disappear. Strip it down and you get three answers. You have to find the facts that decide the matter: what happened, who knew, when they knew it. You have to produce responsive material to the other side, because the federal rules say so. And you have to protect the material that must not go along for the ride: privileged communications, personal information, the things that get redacted rather than produced. Three jobs. Every workflow ever built in this industry serves some mix of them.
One of those jobs wins cases. The other two keep you out of trouble. And the incumbent workflow is a conveyor belt that runs the trouble-avoidance jobs first, every time. The facts arrive at the end: collection, processing, endless rounds of search term negotiation, predictive coding, review, privilege screens, and then, at the bottom of all that, the tiny folder of documents that actually matter. The attorneys who need that folder typically meet it six months in, with depositions bearing down and most of the budget already spent.
If you’re reading this, you probably don’t need to be sold on AI. You’ve watched it analyze a corpus and you know what it can do. So I’ll skip the sermon about capability and make the ask directly: change the order of operations.
The high cost of “what happened?”
Discovery as a whole eats somewhere between a fifth and half of total litigation spend in complex matters. Now consider what all that money buys: knowledge of what your own client’s documents say. The most expensive line item in modern litigation is, functionally, a very slow reading comprehension exercise.
Every innovation this industry has produced took a swing at that line item, and every one of them swung in the same direction: cheaper. All of it made the pipeline less expensive to run, and none of it moved when a legal team understands its case, because all of it lives inside the pipeline, downstream of collection, processing, and hosting. Thirty years of optimizing the price of finding out late.
Agentic AI is the first technology that moves the when. An agent doesn’t need your data collected, processed, loaded, and indexed before it can start reading, which means where the reading happens is now a design question rather than a fact of the pipeline. Maybe it happens behind the client’s firewall, before collection is even scoped. Maybe it happens the moment data lands, before anyone has said the words “search terms” out loud. Maybe the trouble-avoidance work rides along at ingestion, and privilege becomes a metadata field populated on the way in rather than a review call made at the end. Every one of those is on the table today, and every one pulls the same work forward: who talked to whom, when the tone shifted, where the paper trail goes quiet at exactly the wrong moment. The understanding that used to arrive in month six can start forming in month zero, and it doesn’t ask the pipeline’s permission.
Compliance has a deadline. Understanding doesn’t.
The traditional edrm sequence survives because of a simple asymmetry: two of discovery’s three jobs are enforced and one is not. Production comes with court-ordered deadlines and regulatory response windows. Protection comes with waiver, and with the special misery of clawing back a privileged document from opposing counsel. Miss those obligations and something happens to you. But no court has ever ordered a legal team to understand its own case by a date certain. No statute demands a timeline of key events. Nobody in the history of this profession has been sanctioned for figuring out the story late.
So every calendar in the matter belongs to the other two jobs. The ESI protocol, the meet and confer, the production schedule, the vendor statement of work: all of it is scaffolding for enforced obligations. Fact development doesn’t get a workstream or a budget line or even a date. It gets whatever is left over: senior attorneys at the end of the pipeline, reading what the process surfaced, under time pressure the process itself created. The job with the highest stakes in the matter runs on scraps, and it runs last.
The matter, reordered
Run fact development first and the matter changes shape from the opening weeks.
Counsel’s advice is grounded in what the corpus actually contains instead of what everyone assumes it will eventually yield. Anyone who has sat through an early case assessment built on custodian interviews and an org chart knows the difference between “we believe the exposure is limited” and “we have read the story, and the exposure is limited.” One of those is a guess wearing a suit. Scope flows from the same knowledge: custodian fights and proportionality arguments calibrated against known exposure instead of fear of the dark, and deposition prep that starts from a story rather than a document dump.
Then there’s the decision that matters most to a legal department, which is rarely trial. The substantial majority of civil matters settle, and settlement is a business decision that currently gets made with the best information arriving last, which means the burn rate often makes the decision before the facts do. Put the narrative in hand at month zero and settle-or-fight becomes a choice made on facts, on your timeline, at your number.
And the trouble-avoidance jobs come out ahead in the bargain. When the corpus is understood early, privilege stops being a late-stage panic and starts being a property of the data: flagged on the way in, carried as metadata, validated instead of hunted, and transferrable to the next matter. The same goes for the personal information you’re obligated to protect. Names, account numbers, health data. Finding PII at the end of review is remediation; finding it at ingestion is hygiene. The jobs with deadlines get easier when the job without one goes first.
Sequence is a decision
Changing the order of operations is not a procurement exercise. No tool reorders a matter on its own, and anyone claiming otherwise is selling a faster conveyor belt. This is a decision, made matter by matter, to run the case-winning job first and let the deadline jobs draft behind it. Corporate legal departments own the data, the budgets, and the outside counsel guidelines. The pipeline has always been shaped by whoever holds the enforcement power, and on this question, that power sits with the client.
For twenty years, the answer to “when do we find out what happened?” was “when the pipeline finishes.” That was never a design principle. It was a workaround for a capability that didn’t exist, and the workaround has outlived its reason. This was never really a question of which tools to deploy. It is a question of when the most important work in a matter gets done, and for the first time in the history of this practice, that is a choice.

Joe Pirrotta
Joe is a recognized expert in the Legal and Compliance industry, with over fifteen years of experience in process design, product management and long-term strategy. He began his career as one of the founding members of Integreon’s Onshore Managed Review service line, growing the business across four continents and managing bet-the-company litigations and investigations for Fortune 100 companies and AmLaw 200 firms. He then went on to launch a series of new product offerings and lead multidisciplinary teams as the head of the legal business units of two outsourcing organizations. Joe brings his deep domain expertise and experience in productizing legal services to ProSearch to reimagine the practice of review for a transformed era. Joe earned his J.D. from Rutgers University School of Law and is admitted to practice law in good standing in New Jersey.

