AI First: How HaystackID® Rebuilt the Discovery Operating Model Around Earlier Insight

HaystackID Blog | August 10, 2026
AI First: How HaystackID® Rebuilt the Discovery Operating Model Around Earlier Insight

Editor’s Note: Generative AI (GenAI) is moving to the front of the discovery workflow, and the commercial model is moving with it. HaystackID applies matter-specific AI analysis to a processed population before it is promoted into active review, then delivers processing, hosting, AI capacity, users, and project management as one monthly fee across four capacity tiers. Case Insight™ cites the documents behind its conclusions, so an output can be traced back to the record that produced it. The approach targets total cost rather than unit price, on the reasoning that less data reaching the stages that bill by the gigabyte and the hour can lower what a matter costs overall. For cybersecurity, data privacy, regulatory compliance, and eDiscovery professionals, the timing is worth attention. A federal magistrate judge in California recently applied established reasonableness and proportionality principles to a GenAI review workflow rather than imposing a new standard, which offers an early reference point for what such a workflow may need to demonstrate. The same front-loaded analysis applies to breach notification and regulatory response, where cost is driven by how late personal data gets separated from the larger population. Watch three things through the rest of the year: whether other courts follow that reasoning, how ESI protocols evolve to address AI validation, and whether providers will define AI entitlements and exception handling in writing.

AI First: How HaystackID® Rebuilt the Discovery Operating Model Around Earlier Insight

Disclosure: This company-authored article describes HaystackID products and services. Certain program terms and engagement results are based on internal company information and have not been independently audited.

One of the most consequential cost decisions in electronic discovery is often made before counsel has a substantive understanding of the data. By the time active review begins, the data has already been collected, processed, and staged, and the significant costs may already be tied to a data population that has not yet been evaluated against the issues in the matter.

That sequence is so familiar it rarely gets questioned. Analysis waits until the end, which means every earlier stage runs at full volume, and insight arrives after the spend rather than before it. HaystackID’s enterprise operating model changes that order of operations: applying matter-specific AI analysis to a processed document population before that population is promoted into active review. This gives counsel earlier insight into what the data contains and what the downstream process actually needs to carry. HaystackID then delivers the entire program through one monthly fee offered at four capacity tiers, aligning the commercial model with that changed sequence.

The distinction matters because unit price and total cost are different problems, and only one of them has been getting attention. Legal departments have grown skilled at managing the costs they can see. In the midyear pulse of Norton Rose Fulbright’s 2026 Annual Litigation Trends Survey, a law firm-sponsored poll of 135 U.S. corporate counsel in the energy, financial institutions, healthcare and technology sectors released June 10, 2026, 71% said managing outside counsel costs had either stayed the same or become easier, while 55% reported improvement in managing internal legal budgets. What has proven harder to control is the volume moving through the workflow. Rate discipline does little when the amount of data reaching the meter is set by a process that defers analysis until after the meter starts.

AI First, Not Last

Moving the analysis earlier is the whole idea. Case Insight™ evaluates, classifies, and prioritizes a document population after processing but before it is promoted into an active Relativity review workspace, so the decisions that set downstream cost get made with knowledge of what the data contains rather than in advance of it.

That change in sequence may reduce the population promoted to active review, depending on data characteristics, matter objectives, validation requirements, and client-directed workflows. Where the population is reduced, the effects carry downstream: tier capacity stretches across more matters before an overage applies, and counsel gains earlier access to issue-focused information on scope, strategy, settlement posture, and production. The approach is also designed to reduce repetitive search effort, because the population in front of the team has already been organized around the issues in play.

One completed engagement illustrates the pattern. According to HaystackID’s internal engagement data, Case Insight prioritized approximately 5,600 of roughly 156,000 processed documents for downstream attention in an anonymized matter, a reduction of about 96% in the population carried forward to review.*

*Illustrative only. This figure reflects a single anonymized engagement and has not been independently audited. Results vary by data set, issues in play, workflow design and validation requirements, and HaystackID makes no representation that any client’s matters will produce comparable figures.

Technology HaystackID Controls

A workflow built on someone else’s software is only as adaptable as that vendor’s roadmap allows, which is why ownership of the application layer is a strategic question rather than a procurement detail. Through its acquisition of eDiscovery AI®, HaystackID owns the company that develops Case Insight and CaseBot™ and controls the products’ application and workflow development. For clients, that means direct access to the team shaping the roadmap rather than feedback routed through a reseller into a third party’s support queue.

HaystackID announced Feb. 26, 2026, that it had acquired eDiscovery AI, a legal technology company focused on GenAI document analysis. Terms were not disclosed.

“Our clients are asking for easy-to-deploy GenAI capabilities that deliver deep insights, defensible results, and adaptability to changing use cases,” Chad Pinson, Chief Executive Officer of HaystackID, said in the announcement. Pinson was named Chief Executive Officer Jan. 20, 2026, when Hal Brooks moved to Executive Chairman.

Michael Sarlo, Chief Innovation Officer and President of Global Investigations and Cyber Incident Response at HaystackID, described where clients are putting the technology to work. “Clients are no longer evaluating GenAI in theory; they are operationalizing it across early case assessment, investigations, regulatory response, and review quality controls,” Sarlo said.
The practical result is a single accountable party for the application layer and the service, with fewer integration seams and a defined path for expanding capability over the life of a relationship.

The Same Approach in Incident Response and Regulatory Work

Sarlo’s list points to something worth drawing out: the logic travels beyond litigation. Data-breach notification work has a similar shape to discovery, with a large corpus, a smaller subset that carries obligations, and a cost curve driven by how late in the process anyone separates the two.

The exposure is current. Half the respondents in the Norton Rose Fulbright survey, 51%, named data and cybersecurity breaches as a class action trigger, and 46% reported rising federal dispute exposure tied to AI. HaystackID brings the same capability set to that work, including digital forensics and remote custodian collections, sensitive data identification through Protect Analytics AI®, managed review through ReviewRight® Global Managed Review, and production and export in required formats, across litigation, investigation, and regulatory matters in the U.S. and EMEA.

One Monthly Fee, With Line-Item Transparency

Changing the sequence only pays off if the commercial model reflects it. A workflow that reduces data upstream saves nothing if the contract still meters every stage as though it did not. HaystackID has operated subscription-based eDiscovery programs for over a decade, and each program is configured around the client rather than sold from a shelf. The model consolidates data processing, repository hosting, Case Insight, CaseBot, GenAI consulting, RelativityOne review hosting, cold storage, platform users, project management, and CoreFlex™ into a single bundled monthly fee.

Bundled does not have to mean opaque, and the program is built so it is not. Each tier states the processing, hosting, AI, user, consulting, and project management capacity it includes; consumption is tracked against those allowances inside CoreFlex and visible to the client; burst rates are set in advance for every component, so an unusually heavy month is a known cost rather than a negotiation; and services outside the bundle, such as forensic collections, managed review, and productions, are available at pre-agreed rates. Traditional per-gigabyte and hourly rates can be supplied alongside the subscription so procurement can benchmark one against the other.

AI capacity is designed not to scale down with the entry tier. Case Insight capacity, CaseBot capacity, and GenAI consulting hours are set at the same level across tiers, so the smallest tier is not intended as a reduced-capability version of the program, and a client sizing to a modest caseload is not sizing away from the analysis that makes the sequence work.

Control When You Want It, Support When You Need It

Predictable pricing is only half of what a legal team needs from an operating model. The other half is being able to change how work gets delivered without renegotiating anything. CoreFlex, which HaystackID launched in 2025 as a cloud-native interface built on Microsoft Azure, is the operating and reporting layer that makes delivery mode a choice rather than a commitment. Teams can self-serve where it is efficient and transition to full-service HaystackID support as a matter becomes larger, more urgent, or more sensitive. Because both modes are designed to run in the same workspace on the same platform, changing mode is intended to be a staffing and access change rather than a technical one, without a separate provider to engage.

The transition is designed to run in both directions. Project management, forensic support, analytics, GenAI expertise, managed review, or production support can be added to an in-flight matter and stepped back out when the matter no longer needs it, while the operating model, reporting structure, and technology environment stay constant. That constancy is what keeps the portfolio view coherent, with reporting at both the matter and program level covering data processed and hosted, technology usage against allowances, project management utilization, and spend tracked against the selected tier, with variance visible early. LegalTech Breakthrough named CoreFlex Overall eDiscovery Solution of the Year on Nov. 13, 2025, in the program’s sixth annual awards.

Defensibility by Design

None of this matters if the work cannot be defended, and that is the question counsel asks first about any AI-assisted workflow. Courts have begun to place AI-assisted review inside familiar doctrine rather than treating it as an experiment.

In Schulte v. LinkedIn Corp., No. 22-cv-00237-HSG (LB), U.S. Magistrate Judge Laurel Beeler of the Northern District of California signed a discovery order June 30, 2026, filed the following day, addressing LinkedIn’s use of Relativity aiR. The plaintiffs asked the court to prohibit LinkedIn from using search strings to pre-cull documents before aiR review, to compel LinkedIn to run aiR across all custodial files, and to compel disclosure of additional metrics. Beeler denied the requests, finding that using search terms to pre-cull documents before providing them to technology-review platforms satisfies the reasonableness and proportionality standards, and declining to compel disclosure of “elusion estimates, the document error rate, and the number of human reviewers being used to validate Relativity aiR’s predictions.” The plaintiffs had not shown that LinkedIn’s 25 search strings were deficient, and the court ordered the parties to meet and confer within 21 days about the search strings and whether any adjustments were warranted. The order describes Relativity aiR as a form of technology-assisted review.

The order offers an early example of a federal court applying established TAR, reasonableness and proportionality principles to GenAI review. Its reach is limited to the record, the parties’ ESI protocol and the challenges presented. It involved a different party and a different tool, and it is not an endorsement of any provider. Lawyers at WilmerHale described it in a July 20, 2026, client alert as applying traditional TAR principles to GenAI discovery, and Arnold & Porter reached a similar reading.

What the order does illustrate is the kind of process that withstands challenge: reasonable, proportional, and demonstrable, supported by what the parties negotiated into their ESI protocol. Demonstrable is the operative word, and it is where the design of the tool matters as much as the design of the workflow.

Case Insight is built to show its work, especially since AI-generated assessments may contain inaccuracies and are subject to validation and human review. The assessment memo cites the documents behind its conclusions, so a reviewer can move from a stated theme, issue, or priority back to the specific records that produced it rather than accepting a ranking as a black box. Traceability of that kind is what turns an AI output into something a team can examine, test, and explain.

Validation runs alongside it. HaystackID reported recall rates exceeding 90% for Case Insight when it introduced the technology in March 2025. That figure is a starting point rather than the validation itself, because the program does not rely on a single published benchmark to stand in for matter-level proof. The validation record for a given matter includes sampling of the population not carried forward, human review at defined checkpoints, documented criteria for what counts as relevant, and a record of how the workflow was configured. Those criteria and the escalation path are agreed with the client before the work begins, not reconstructed after a challenge.

Validation answers whether the output is right. The security controls answer a different question: who can access the data to produce it. Matter-specific access controls determine who can query what; outputs are auditable, and client data remains subject to the security, confidentiality, and data-governance requirements established for the engagement.

One practical consequence deserves attention from counsel: negotiate the ESI protocol with the AI workflow in mind, because the protocol, not the marketing material, is what a court will read. It is also worth remembering that Schulte is one discovery order from one magistrate judge. Analysis published in LegalTech News on Jan. 6, 2026, predicted a first wave of defensibility opinions during the year and cautioned that courts were unlikely to fully endorse AI for responsiveness determinations. Programs built for disciplined implementation and human-in-the-loop checkpoints are the ones positioned to absorb whatever the next order says. Clients and their counsel remain responsible for legal strategy, responsiveness determinations, privilege decisions, production decisions, and other substantive legal judgments.

Questions Worth Asking Any Provider

Those are the answers HaystackID gives about its own program. The more useful exercise is putting the same questions to everyone, because a bundled fee should simplify without concealing, and the discipline buyers apply to rate cards should carry over.

Ask for the entitlement behind every component, stated in units, so processing, hosting, AI capacity, consulting hours, and project management each carry a defined allowance rather than a shared assurance of sufficiency. Ask for overage rates set in advance for each of those components. Ask whether capacity can be adjusted downward as well as upward once actual consumption is known, since the risk in a multiyear subscription is committing to a tier sized on assumption.

Ask how GenAI exceptions are handled. A file a model cannot process, a population that exceeds a context limit, a re-run after a change in prompt or criteria: each may affect capacity or cost depending on the commercial arrangement, and each is worth defining in the agreement rather than discovering on an invoice.
Ask who owns what, because “who owns the model” is really five questions. Who provides the underlying foundation model, who develops the application layer, who controls the prompts and workflow logic, who owns the outputs generated on client data, and which of those can be changed at a client’s request. A provider that develops its own application layer has latitude that a reseller does not, but that is a different claim from owning the foundation model, and the distinction is worth drawing out in writing.

Ask what validation documentation the provider produces by default, and whether the AI output can be traced back to source records. And ask what the reporting actually shows, because consumption tracking that reports volume without connecting it to spend and outcome is a dashboard, not a program.

The Bottom Line

The industry has spent years negotiating the price of steps it never questioned the order of. Moving analysis earlier changes what the rest of the process has to carry, and that single change is what connects the pieces described here: a smaller population entering review, a fee structure that stays legible as it does, delivery that flexes without a migration, and outputs a team can trace back to the record.

Which raises the question worth putting to your own program this quarter: if AI can characterize the data before the full population moves into active review, how much of the downstream cost and effort can be avoided or redirected?

To discuss how an AI-first operating model would fit your portfolio, contact HaystackID for a briefing.

News Sources

HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI® technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.

Assisted by GAI and LLM technologies.

SOURCE: HaystackID

Advisory Note: As organizations adopt GenAI across the discovery lifecycle, the ability to understand a matter early is becoming a strategic advantage rather than a luxury. HaystackID’s Core Intelligence AI Case Insight, delivered in partnership with eDiscovery AI, is a GenAI-powered matter intelligence engine that transforms large, complex data collections into actionable case intelligence within days. By identifying key documents, mapping relationships, surfacing emerging themes, and highlighting potential risks at the outset of a matter, Case Insight enables legal teams to make better-informed decisions before investing in full-scale review. Combined with expert legal oversight and rigorous validation, the platform delivers rapid insight while maintaining the defensibility today’s matters demand. Whether informing early case assessment, refining review strategy, or accelerating fact development, Case Insight helps organizations move from reactive document review to proactive, intelligence-driven discovery.