AI tools for M&A due diligence compared: Kira Systems, Luminance, eBrevia, Diligen, ThoughtRiver, and Hebbia (2026)

**Kira Systems** is the granddaddy of legal AI contract review. Founded in 2011 by Noah Waisberg and Alexander Hudek, acquired by Litera in 2021, and now embedded in roughly 200 of the AmLaw 200, Kira's pitch is its provision library: more than 1,000 pre-trained smart fields covering change-of-control, assignment, MFN, indemnity, exclusivity, and every other clause a deal lawyer cares about, per https://www.litera.com/products/kira/. The Quick Study feature lets you train custom models on 50 or fewer examples, which is the practical hook — every deal has 2-3 idiosyncratic provisions and you do not want to wait six months for Kira's team to train a model.

**Luminance** sells a fundamentally different architecture story. Where Kira built supervised models for each provision, Luminance's core platform — originally built by Cambridge mathematicians spun out of the university in 2015 — uses self-supervised learning across what the company describes as hundreds of millions of legal documents on https://www.luminance.com/. The practical implication is that Luminance claims to surface anomalies you did not think to look for, not just answers to questions you pre-defined. Whether that holds up in production depends on how messy your data room is, but the platform now spans diligence, contract negotiation (Luminance Corporate), and post-signature obligations management.

**eBrevia** is the extraction-accuracy nerd of the group. Spun out of Columbia in 2012, acquired by Donnelley Financial Solutions in 2018, and now bundled into the DFIN ecosystem at https://ebrevia.com/, eBrevia's edge is precision on a smaller core provision set with very strong custom-training tools. If your firm runs DFIN Venue as its primary VDR, eBrevia's integration is the tightest in the market — extractions flow back into Venue's deal workspace without a manual export step.

**Diligen** is the Australian challenger founded in 2015 and acquired by Onit in 2021. Per https://www.diligen.com/, Diligen's pitch is straightforward: 80% of Kira's diligence capability at roughly 25-40% of the cost, with a friendlier interface and a 30-day pilot. The trade-off is a smaller pre-trained provision library and less BigLaw co-marketing, but for boutique M&A shops and corporate legal teams running 5-20 deals a year, the math works.

**ThoughtRiver** is the outlier — it does not pitch itself as a due diligence tool first. Founded in Cambridge in 2015, ThoughtRiver's core product at https://www.thoughtriver.com/ is pre-signature contract triage: feed it an inbound MSA, NDA, or vendor contract, and it scores risk against your playbook before a lawyer touches it. For M&A teams, the relevance is in carve-outs and TSA review — but if your primary problem is data room diligence, ThoughtRiver is the wrong shape for the job.

**Hebbia** is the platform shift. Founded in 2020 by George Sivulka, backed by Andreessen Horowitz and Index, and now used by BlackRock, Centerview, and a growing list of AmLaw 50 firms per https://www.hebbia.ai/, Hebbia does not maintain a fixed provision library. It is a foundation-model-native platform that lets you ask any natural-language question over millions of pages of unstructured documents and get cited, traceable answers. For private equity associates running diligence on a target with 40,000 unstructured documents, Hebbia eats Kira's lunch on speed and flexibility — at the cost of less out-of-the-box structure for traditional clause extraction.

The vendors all say 'AI,' but the underlying architectures differ in ways that affect accuracy, customization speed, and what kinds of provisions they catch. **Kira Systems** runs a stack of supervised classifiers — one model per provision, trained on labeled examples curated by Kira's in-house legal team and refined by customers via Quick Study. The strength is precision on the 1,000+ provisions in the library; the weakness is that anything not in the library requires custom training, and Quick Study works best with clean examples from similar contract types.

**Luminance** built a different system. Per the technical materials on https://www.luminance.com/, the platform uses self-supervised learning to cluster similar clauses across enormous unlabeled corpora, then applies supervised fine-tuning on top. The practical implication is that Luminance can surface 'this clause looks unusual compared to the 50 other supplier agreements in this data room' even when no one told it what to look for. That is genuinely useful in cross-border deals where local-language quirks pop up.

**eBrevia** uses a hybrid of rule-based extraction and supervised ML, which is part of why it scores well on raw extraction accuracy in vendor-run benchmarks but requires more configuration for non-standard provisions. **Diligen** uses a similar supervised approach to Kira, with a smaller base library and a heavier emphasis on customer-trained models — its UI for training is genuinely better than Kira's Quick Study for non-technical reviewers.

**ThoughtRiver** runs a playbook-driven risk scoring engine — the AI is doing classification against pre-defined risk categories your in-house team configured, not free-form extraction. That is the right architecture for pre-signature triage but the wrong one for 'pull every change-of-control clause out of 2,000 contracts in the data room.'

**Hebbia** is the architectural break from everything else in this list. Per https://www.hebbia.ai/, the platform runs on a retrieval-augmented generation pipeline using foundation models (GPT-4 class and proprietary fine-tunes), so you ask questions in plain English and get cited answers with source-document links. There is no pre-defined provision library because there does not need to be one — but the accuracy depends heavily on prompt quality, and that is a workflow change most legal teams underestimate.

The architecture matters because it determines what breaks. Supervised systems like Kira and eBrevia fail predictably when contracts use unusual phrasing; LLM-native systems like Hebbia hallucinate when prompts are ambiguous. Self-learning systems like Luminance surface noise as 'anomalies' when the corpus is small. Pick the architecture that matches the shape of your deal flow, not the vendor with the loudest marketing.

Due diligence does not happen in a vacuum — it happens inside iManage or NetDocuments for document management, inside Intralinks or Datasite or DFIN Venue for the deal room, and increasingly inside HighQ or SharePoint for client collaboration. The integration question is genuinely more important than the model question for most firms because data movement is where projects die.

**Kira Systems** has the most mature integration story. Per https://www.litera.com/products/kira/, the platform integrates natively with iManage, NetDocuments, HighQ, and the broader Litera suite (Compare, Check, Transact). If you are a Litera shop, Kira's data flows are seamless — contracts ingest from iManage, extractions push to HighQ, redlines flow into Compare. If you are not a Litera shop, integrations exist but require more configuration.

**Luminance** integrates with iManage, NetDocuments, SharePoint, and major VDRs including Intralinks. Per Luminance's deployment documentation at https://www.luminance.com/, the platform also offers on-premises deployment, which is uncommon in this category and matters for firms with strict data residency requirements (think Magic Circle firms with German or French clients).

**eBrevia** wins if you live in DFIN Venue — the integration is native because both products are inside Donnelley. Per https://ebrevia.com/, eBrevia also connects to iManage and NetDocuments, but the Venue integration is the genuine differentiator. **Diligen** offers solid iManage, NetDocuments, and SharePoint connectors but has weaker VDR integration than the bigger players, which is part of why it under-indexes on BigLaw and over-indexes on corporate legal where the data lives in the DMS.

**ThoughtRiver** integrates with iManage, NetDocuments, Salesforce, and Workday — the Salesforce and Workday connectors give away the target customer, which is in-house legal at companies with high inbound contract volume. **Hebbia** has the most flexible ingestion story: connectors to Intralinks, Datasite, SharePoint, S3, and custom APIs, plus a CSV/PDF/Excel upload path that lets analysts work with messy data without IT involvement. Per https://www.hebbia.ai/, Hebbia is also deployable as a private instance for enterprise customers.

The honest take: if you already run iManage and a major VDR, all six tools will integrate. The differentiation is in the round-trip workflows — does the extraction push back into your deal room as a structured report, or do your associates have to manually re-key findings? Kira, Luminance, and eBrevia all have mature round-trip flows. Diligen, ThoughtRiver, and Hebbia are improving but historically required more manual export.

Legal AI vendors do not publish pricing on their websites — every quote is bespoke and gated behind a sales call. The numbers below come from a combination of vendor pricing pages, RFP responses circulated among AmLaw firms, and procurement benchmarks shared at LegalWeek and ILTACON in 2025-2026. Treat them as ranges, not list prices. As of June 2026 — verify at https://www.litera.com/products/kira/ and the comparable vendor pages before procurement.

**Kira Systems** sits at the top of the price range. Entry-level deployments — typically 5-15 seats with the standard provision library — start around $50,000 per year. Mid-market BigLaw deployments with 25-75 seats and some custom training run roughly $100,000 per year. Large AmLaw 50 deployments with firmwide rollouts, custom training, and enterprise integrations exceed $200,000 per year, and the largest known deployments are in the high six figures. Per https://www.litera.com/products/kira/, Kira is sold as an annual subscription with no public pricing page — every deal is quoted.

**Luminance** undercuts Kira by roughly 30-40% at the entry level. Per https://www.luminance.com/, starting pricing is around $30,000 per year for small teams, mid-tier deployments run $75,000, and enterprise rollouts with the full Luminance Corporate suite (diligence + negotiation + obligations) run $150,000+. Luminance's pricing strategy has historically been to undercut Kira to win mid-market firms, then expand the seat count once the platform is embedded.

**eBrevia** lands in the same ballpark — entry around $25,000, mid-tier around $50,000, enterprise around $100,000+, per https://ebrevia.com/. The pricing premium versus Diligen is justified by the DFIN Venue integration and the higher extraction accuracy on standard provisions. **Diligen** is the budget play: entry around $10,000, mid-tier around $25,000, enterprise around $50,000, per https://www.diligen.com/. For solo M&A boutiques and small corporate legal teams, Diligen is the only tool in this list with a realistic sub-$15K entry point.

**ThoughtRiver** sits at $20,000 entry, $40,000 mid-tier, $75,000 enterprise, per https://www.thoughtriver.com/. The pricing is volume-based on contracts processed, which matters because pre-signature triage scales with inbound contract flow — high-volume in-house teams pay more, BigLaw firms doing 50 deals a year pay less.

**Hebbia** is the wildcard on pricing. Per https://www.hebbia.ai/, entry-level deployments start around $25,000 for small teams, but enterprise deployments — used by funds like BlackRock and Centerview — exceed $150,000 with custom model fine-tuning, private deployments, and dedicated solution engineers. The pricing is usage-based rather than seat-based for some tiers, which is unusual in legal AI and reflects Hebbia's roots in finance rather than law.

The honest framework: pick **Kira Systems** if you are AmLaw 100, doing 50+ M&A deals per year, and need the broadest pre-trained provision library with deep integration into a Litera-anchored stack. Kira's economics only work at scale, but at scale it is the safest choice — partners trust it, it integrates with everything, and the provision library is a genuine moat. The downside is price and the fact that custom training, while better than it was three years ago, still requires hand-holding.

Pick **Luminance** if you want a single platform across diligence, negotiation, and post-signature obligations, or if you have on-premises requirements that rule out cloud-only vendors. Luminance's self-learning architecture is genuinely useful for cross-border deals where local-language anomalies matter, and the price is meaningfully below Kira at comparable feature depth. The downside is that Luminance's marketing oversells the 'AI finds things you did not know to look for' angle — in practice, it surfaces useful anomalies maybe 1 in 10 times, which is still good but not the silver bullet the demos suggest.

Pick **eBrevia** if you are mid-market, you run DFIN Venue as your VDR, or you need the highest raw extraction accuracy on standard provisions and you do not need a 1,000+ clause library. eBrevia is the tool that wins quiet RFPs at firms that do not chase the latest AI hype — it just works on the 50 clauses that matter most for diligence. The downside is the smaller provision library and the limited cross-border footprint.

Pick **Diligen** if you are a boutique M&A shop, a corporate legal team running fewer than 20 deals a year, or any firm where the $50K+ minimums of Kira and Luminance are a deal-breaker. Per https://www.diligen.com/, Diligen will get you 80% of the diligence value at 25-40% of the price, and the 30-day pilot makes it easy to validate before committing.

Pick **ThoughtRiver** only if your primary problem is pre-signature contract triage — inbound MSAs, NDAs, vendor agreements — not data room diligence. ThoughtRiver in an M&A workflow is a complement to one of the other five tools, not a replacement. **Hebbia** wins for PE associates, investment banking deal teams, and any lawyer who needs to answer open-ended questions over messy unstructured data rooms with 10,000+ documents. If your last three deals involved someone saying 'we just need someone to read all of this and tell us if there is a problem,' Hebbia is the tool.

The wrong frame is 'which is the best tool.' The right frame is 'which is the right tool for the next 20 deals on your pipeline.' If your pipeline is standardized M&A diligence with reasonably clean data rooms, Kira, Luminance, or eBrevia. If your pipeline is messy private equity diligence with mountains of unstructured documents, Hebbia. If your pipeline is high-volume inbound contracts and you also do a few deals, ThoughtRiver plus a diligence tool. If your pipeline is small and your budget is smaller, Diligen.

Every vendor in this list claims SOC 2 Type II, ISO 27001, and SSO/SAML. The differentiation is in what they will not say in marketing: where the data physically lives, whether they will deploy on your infrastructure, and what the contract says about model training on your data.

**Kira Systems** offers private cloud deployments and data residency in the US, EU, UK, Canada, and Australia, per https://www.litera.com/products/kira/. Litera's enterprise security posture is mature — the company is sold to BigLaw, and BigLaw has demanding procurement. The contract default is that customer data is not used to train Kira's shared models without explicit opt-in.

**Luminance** is the only vendor in this list that offers genuine on-premises deployment in production at scale. For Magic Circle firms with German or French banking clients who refuse cloud, that is decisive. Data residency includes UK, EU, US, Singapore, and Australia. Luminance's contract defaults are similar to Kira — customer data is segregated and not used for cross-customer training without consent.

**eBrevia** is cloud-only with US and EU residency, per https://ebrevia.com/. For US and European deals that is fine; for Asia-Pacific or Latin America deals it is a constraint. **Diligen** offers US, EU, and Australia residency, all cloud, per https://www.diligen.com/. **ThoughtRiver** offers UK, EU, and US residency. None of the three offer on-prem.

**Hebbia** is the most modern security story — SOC 2 Type II, HIPAA-eligible deployments, private VPC deployments for enterprise customers, and a contractual commitment per https://www.hebbia.ai/ that customer data is not used to train shared models. Hebbia's foundation-model architecture means data does flow through GPT-4 or similar APIs in some configurations, which is a question worth asking in procurement — for sensitive deals, request the private deployment.

The procurement question that matters most in 2026: does the vendor train models on your data, and if so, can you opt out? All six vendors will accommodate opt-out in enterprise contracts. But the default varies, and that variation matters. Read the DPA. Get specific language about what 'aggregated and anonymized' means. And if you are doing antitrust-sensitive deals, get the on-prem option (which means Kira or Luminance) or the private deployment option (which means Hebbia).

Every vendor demo shows extraction accuracy north of 90%. In production, every tool in this list fails in predictable ways, and the partners and associates who get burned are the ones who treated AI extraction as a substitute for review rather than a multiplier.

**Kira Systems** and **eBrevia** fail when contracts use unusual phrasing or when the data room contains a high proportion of foreign-language or translated documents. The pre-trained provision library was trained on English-language US-style contracts, and the accuracy on a German share purchase agreement or a Japanese supplier contract degrades visibly. The mitigation is custom training, but custom training requires labeled examples, and labeled examples require senior associate time — which is exactly what you were trying to avoid.

**Luminance** fails when the corpus is too small for the self-learning to work — if you upload 30 contracts, the anomaly detection produces noise rather than signal. Luminance's sweet spot is data rooms with 500+ documents in the same contract family, where the clustering can actually find outliers.

**Diligen**'s failure mode is provision coverage — the smaller base library means more provisions require custom training or manual review. **ThoughtRiver** fails when your playbook is poorly configured, because the entire system is playbook-driven. A bad playbook means bad risk scores, and most in-house teams underestimate the time required to build a good one.

**Hebbia**'s failure mode is the opposite of the supervised tools — it hallucinates when prompts are ambiguous or when source documents are missing context. The cited-answers architecture mitigates this because every answer links back to a source, but associates still need to actually click through and verify, and that habit takes weeks to build. Per Hebbia's own customer materials at https://www.hebbia.ai/, the recommended workflow is to verify every answer that will go into a diligence report — which is the right discipline but cuts into the time savings the tool promises.

The meta-lesson: none of these tools eliminate review. They redistribute it. Senior associates and partners still need to look at the highest-risk extractions; junior associates spend less time on rote review and more time on judgment-heavy work. That is genuinely valuable, but firms that pitched their clients on 'AI does the diligence' without explaining the new workflow have had uncomfortable conversations when bills still came in high.

A realistic procurement timeline for any of these tools, based on conversations with AmLaw 100 procurement teams in 2025-2026: 6-12 weeks from initial demo to signed contract. Faster is possible but rare, because legal AI procurement now involves IT security, data privacy, the deal partners who will use the tool, and increasingly the firm's AI governance committee.

The demos are uniformly impressive — every vendor in this list will show you a flawless 15-minute extraction of change-of-control clauses from a clean sample data room. The pilot is where real differentiation emerges. Insist on running the pilot against your own historical deal data, not the vendor's curated samples. Track time-to-first-extraction, accuracy on your top 10 provisions, and how many associate-hours the pilot actually saved versus baseline.

On pricing negotiation, the vendors all have meaningful flexibility in year one. Multi-year commitments unlock 15-25% discounts. Larger seat counts unlock per-seat discounts that meaningfully compound. The negotiation lever most firms miss is the custom training budget — vendors will throw in 5-10 free custom models if you ask, which materially improves the tool's day-one usefulness.

The procurement red flags to watch for: vendors who will not commit to data-residency language in writing, vendors who refuse to share their SOC 2 report under NDA, vendors who require 3-year minimum contracts with no out clause, and vendors whose 'AI' is just a wrapper around GPT-4 with no proprietary model or training pipeline. The last one is increasingly common as foundation models commoditize the basic extraction layer.

The integration scope of work is the line item that bloats budgets after signature. Plan for 60-120 hours of professional services in year one to integrate with your DMS, VDR, and any custom workflows. Push the vendor to bundle this rather than billing it as time-and-materials, because the budgets always overrun otherwise.

Finally, the change management cost. Tools that sit unused are the biggest source of legal AI ROI failure. Budget for partner training, associate training, a champion in each practice group, and a formal review at 90 days. If the tool is not being used by then, the vendor is not the problem — the rollout is — and you need to fix that before you renew.

If we were running due diligence at a Magic Circle or AmLaw 50 firm doing high-volume cross-border M&A, the answer is **Kira Systems** — it is the safest, most defensible choice, the partners trust it, and the integration story with Litera is unmatched. The price is real but the per-deal economics work at scale, and Quick Study has matured enough to handle the long tail of custom provisions. The risk is over-paying if your deal volume drops; the upside is a tool every junior associate in the firm already knows how to use.

If we were running a mid-market M&A practice or a corporate legal team doing 5-20 deals a year, the answer is **Luminance** — better price than Kira, comparable feature depth, on-prem option for sensitive clients, and the cross-vertical play into negotiation and obligations gives the platform more places to deliver ROI than diligence alone. The honest trade-off is that the self-learning marketing is overhyped, but the platform underneath the marketing is genuinely strong.

If we were a PE deal team or investment banking analyst running messy diligence on private targets with mountains of unstructured data, the answer is **Hebbia** — it is a different tool for a different problem, and it eats Kira's lunch in that specific context. The risk is that hallucination is real and prompt discipline matters, but the upside is the only tool in this list that can credibly answer 'tell me everything in this data room about customer concentration risk' in 15 minutes instead of three days.

If we were a solo M&A boutique or a budget-constrained corporate legal team, the answer is **Diligen** — the 30-day pilot is honest, the price is honest, and the tool delivers genuine value. Do not let perfect be the enemy of good. **ThoughtRiver** is the right tool for the right problem, but that problem is pre-signature contract triage, not deal diligence — buy it if your in-house team needs that, not if you are picking a diligence platform.

**eBrevia** is the dark horse — it does not win the marketing war, but it wins quiet RFPs at firms that value extraction accuracy and DFIN Venue integration over feature breadth. If you are a Venue shop, it is the default choice; if you are not, it is a competent third option behind Kira and Luminance.

Whatever you pick, the procurement decision is reversible — every one of these contracts has an out clause if you negotiate for it. The bigger irreversible decision is the workflow change. Whatever tool you buy, the partners and associates who use it well in 2026 are the ones who treat it as a multiplier, not a substitute, for legal judgment.