AI has dominated the eDiscovery conversation for two years running. But underneath it, a quieter question is getting answered one matter at a time, across every kind of legal team. Where should the work actually live?
For a while, it was easy to believe the deployment conversation in eDiscovery was over. Cloud had won. AI was the story. Where your data sat and who was operating the environment had been settled (or so the industry narrative went), and everything interesting was happening upstream in generative AI models, review acceleration, and analytics.
That framing was always incomplete. It’s now openly wrong.
The deployment question is back on every legal-tech agenda, and with good reason. Vendor roadmaps are compressing. AI capabilities are being repriced. Regulators on both sides of the Atlantic are tightening what “acceptable cloud” means for different data types. Corporate legal departments are pulling more work in-house, which means they’re taking on infrastructure questions their outside counsel used to answer for them. And practitioners who thought they had a five-year architecture plan are being asked to defend it in eighteen months.
You won’t see much of this in the trade-press headlines. It shows up somewhere else: in procurement meetings, in matter-planning conversations, and in the quiet realization that a system chosen four years ago no longer fits the matters coming through the door.
The Landscape Isn’t Binary
Part of what makes this conversation so muddled is that the industry keeps framing it as cloud versus on-premises. That framing has been outdated for years. There are at least four distinct models in active use across the profession today, and each one carries a different economic profile, a different risk profile, and a different operational footprint.
| Deployment Model | What It Is | Where It Tends to Fit |
| Public Cloud (SaaS) | Multi-tenant environment operated by the platform vendor on shared public cloud infrastructure. The dominant deployment model for new eDiscovery adoption. | Teams optimizing for speed of deployment, elastic scale, access to the latest AI and analytics features, and minimal internal IT burden. |
| Private Cloud | Single-tenant, dedicated environment hosted and managed by a provider. Cloud economics with tenant isolation. | Regulated industries and risk-averse enterprises that want modern cloud capabilities without multi-tenant exposure, or without ceding platform control. |
| On-Premises | Infrastructure the organization owns, hosts, and operates behind its own firewall. | Matters governed by strict chain-of-custody, sovereignty, air-gap, or classified-data requirements. Narrower than it used to be, but not extinct. |
| Hybrid | A deliberate combination, routed by matter type, data sensitivity, client mandate, or cost profile. | Firms and legal departments are balancing routine matters against sensitive or client-restricted work. |
That last row is where a surprising number of organizations actually live, often without meaning to. They built on-prem years ago, added cloud for a specific matter or client, kept both running, and never made a formal call on which environment was supposed to be the target state. Accidental hybrid is one of the most expensive postures in the industry. More on that in a minute.
Why the Question Is Urgent Again
A few forces are converging right now to push deployment decisions off the back burner and into current planning cycles.
Vendor roadmap pressure is real. Major platforms have signaled, some explicitly and some through where they’re putting their product investment, that on-premises versions are being sunset for new matters within the next few years. Whether or not the platform you run today is on that list, the market direction isn’t ambiguous. Practitioners who assumed they had another architecture refresh cycle before the question got urgent are finding they don’t.
AI has been repriced. Through late 2025 and into 2026, generative AI features that used to be sold as premium add-ons (first-pass review, privilege analysis, case strategy assistants) started getting folded into base subscription tiers across a wave of platforms. That has three consequences. Access to modern AI is no longer gated by a separate procurement decision. The economics of legacy on-prem deployments look worse because those capabilities are harder to deliver outside the cloud. And the pricing conversation the industry has been having for a decade, with per-gigabyte processing, hosting, and review layered with feature surcharges, is being rewritten in favor of flatter, more predictable structures.
Regulatory pressure keeps expanding. Cross-border matters, sector-specific rules in financial services and healthcare, government contracting requirements, and evolving obligations around AI systems in Europe and the United States are all putting pressure on where data resides, who has access to it, and how automated decisions get documented. For some organizations, that pressure closes doors: public cloud simply isn’t an option for certain matters. For others, it opens them. The operational rigor of a mature cloud provider is easier to defend than a homegrown on-prem environment that hasn’t been audited in years.
Corporate legal is bringing work in-house. In-house departments are pulling more of the eDiscovery lifecycle back from outside counsel and service providers. That shifts who’s actually making deployment decisions, often from a firm’s IT and litigation-support leads to a corporate legal operations function with its own governance requirements, its own budget mechanics, and its own view of what “control” means. The answer to the deployment question often changes with the questioner.
The Trade-Offs Practitioners Are Actually Weighing
Behind every deployment decision sits a set of trade-offs that no single vendor or model resolves cleanly. The practitioners doing this work well tend to walk through them explicitly, rather than defaulting to whatever environment they inherited.
The first is control versus capability. Public cloud environments give you the fastest access to new features, the deepest bench of security engineering, and the elasticity to handle unpredictable data volumes. What they don’t give you is granular control over configuration, upgrade cadence, or the physical location of every byte of data. That trade-off is acceptable, even attractive, for most matter types. For a small but important subset, it isn’t.
The second is predictability versus flexibility. Subscription pricing looks attractive at contract signing and can get volatile at scale, especially as data volumes grow or matters run longer than expected. Organizations with repeat-litigation profiles or predictable investigative workloads are running the multi-year math and finding that flat-fee, capped, or all-inclusive structures (whether cloud, private cloud, or on-prem) beat pure consumption pricing when volume is guaranteed.
The third is internal capacity. On-prem deployments carry real human costs that almost never get charged against the eDiscovery budget. IT staff dedicated part-time to environment maintenance, scaling cycles every three to four years, the specialized expertise required to keep these systems running: all of it shows up on other ledgers. That skill set doesn’t translate cleanly to cloud administration either, which means a migration doesn’t just move the workload. It also obsoletes the expertise the team spent years building.
The fourth is client expectations. Corporate clients are writing deployment requirements into engagement letters more often than they used to. Law firms handling regulated-industry work are finding that their architecture decisions have downstream implications for which clients they can serve at all. The days of choosing an environment purely on internal preference are ending.
The fifth is the easiest one to underestimate: the exit ramp. Whatever you choose today, how reversible is that choice? What does it cost to move a single matter, or an entire book of matters, if the current environment stops fitting? Vendors don’t compete on this question, which means buyers have to ask it themselves.
The Cost of Accidental Hybrid
The sharpest point in this whole conversation, and the one most likely to get missed, is that running two eDiscovery environments by inertia is a lot more expensive than running either one by design.
Doubled security surface. Doubled monitoring. Two skill sets on the IT team. Operational confusion for the business-development staff trying to route matters to the right environment. Inconsistent client disclosures about where data actually lives. Duplicate license and support contracts. And, quietly, a governance posture that’s harder to defend under scrutiny, because it grew organically instead of by design.
Hybrid can be exactly the right answer when it’s chosen. When it’s the result of not having chosen, it’s a hidden tax on the practice.
Where This Leaves Us
There’s no universal right answer to the deployment question. But there is a right process for arriving at one that will hold up. It starts with being honest about the matter mix. What does the work coming through the door actually look like? It requires being honest about internal capacity. What infrastructure and expertise does the organization genuinely have, and want to keep? It requires reading client mandates and regulatory obligations with the assumption that they’ll get stricter, not looser. And it requires running the numbers on a multi-year horizon, not a single procurement cycle.
We’re surveying the ACEDS community because nobody, not the vendors, not the trade press, not the analysts, has a clear picture of how practitioners are actually navigating these trade-offs right now. The industry conversation has been dominated by product announcements. What we don’t have is a good read on what’s happening on the ground.
Early Signals from the Survey
Initial responses from the ACEDS community are already sharpening the picture. We’re not going to publish specific numbers until the sample is closer to complete, but a few directional signals are worth flagging now, because they reframe the conversation in ways the industry narrative hasn’t caught up to.
The “cloud won” narrative is overstated. A majority of respondents so far do operate entirely in cloud/SaaS environments. But a substantial share of the community still runs some form of on-prem or privately-deployed platform, either exclusively or alongside cloud. Deployment isn’t a settled question for the profession as a whole.
The market looks bimodal on private-deployment need. Most respondents report that only a small fraction of their matters require private deployment for residency, sovereignty, or security reasons. But a striking minority reports the opposite. For them, the majority of matters carry those requirements. Very few sit in the middle. That looks less like a gradient and more like two distinct market segments answering fundamentally different questions.
The most-cited decision factors pull against each other. The two factors carrying the most weight in respondents’ current deployment decisions are access to the latest AI and analytics capabilities on one side, and predictability of total cost over a three-to-five-year horizon on the other. Reconciling those two is the single hardest problem in eDiscovery architecture right now, and the early data suggests practitioners are feeling it more acutely than the industry conversation has acknowledged.
AI parity in private deployments isn’t a niche concern. Among respondents whose matters require on-prem or private deployment, the appetite for feature parity with cloud AI is high, and higher than the industry narrative would suggest. The assumption that private deployment means giving up modern AI is being rejected by the very practitioners who need private deployment most.
These are early signals, not conclusions. The picture sharpens with every response.
If you touch eDiscovery in any capacity, whether you’re in corporate legal, at a law firm, on the service provider side, in government, or consulting, we’d genuinely value your perspective on how you’re thinking about deployment in 2026.
Take the ACEDS eDiscovery deployment survey: https://www.surveymonkey.com/r/QFQFH2Q
It takes just a few minutes. We’ll publish the results back to the ACEDS community in a follow-up piece, so everyone can see how the profession as a whole is thinking through the trade-offs: the models winning ground, the concerns cited most often, and the questions still open. The more responses we get, the more representative and useful that picture becomes.
