Review & redline contracts — playbooks manage, Release review tasks
A review workflow for hundreds of engagement letters — algorithmically exact, question for question, with sign-off workflow and playbook-Editor.
In production
Work. ERLEDIGT.
One of Germany's leading independent commercial law firms.
SKW Schwarz is one of the established German commercial law firms. With over 120 lawyers across four locations — Munich, Berlin, Frankfurt and Hamburg — the firm advises companies of every size across all areas of national and international commercial law. As a member of the TerraLex network, SKW Schwarz is also globally positioned.
What sets SKW Schwarz apart in the German law-firm market: in 2018, with the founding of SKW Schwarz @ Tech GmbH, the firm created its own legal-tech unit. Digitalisation here is not a future project but lived practice. It was in exactly this context that the collaboration with UNOY started — not out of curiosity, but from a concrete operational need.
Hundreds of engagement letters. One playbook. No way to scale.
At a law firm with over 120 lawyers, engagement letters keep coming in — engagement agreements that must be reviewed before the mandate is accepted. The review is based on internal playbooks: structured rulebooks that define which clauses are acceptable, which changes are permitted and where the red lines run. The problem: the volume. Reviewing every engagement letter individually and manually against a playbook ties up qualified resources for hours — across hundreds of contracts, that becomes a bottleneck.
Pure AI solutions — for instance a chatbot you hand a playbook and a contract — produce summaries, not reliable review results. The context is too broad, the results too vague, the auditability too weak. A firm like SKW Schwarz doesn't need rough impressions, it needs solid answers to every single review question.
On top of that comes the organisational side: who issued which review task? Who signs off on the result? And when a playbook changes — how do we make sure the new version works correctly before it goes live? Without a system that brings review, administration and playbook maintenance together, scaling stays a wish.
Before. Intervention. After.
per engagement letter · manual playbook review
Over 120 lawyers, hundreds of engagement letters. Each one had to be reviewed manually against an internal playbook — clause by clause, red line by red line. Pure AI solutions produced summaries instead of solid review results.
UNOY Legal Engine
- →Playbook modelled as machine-readable rule logic
- →Version management for playbook maintenance built in
- →Review as a structured run, not a free-form chat
- →Review tasks, sign-offs and history in the system
Model · Partner
per engagement letter · every clause with a playbook reference
No chat output — a solid review result. Playbook changes are tested before they go live. The bottleneck is gone, control stays inside the firm.
Source: internal measurement SKW Schwarz · engagement-letter review
Work in — result out
Input
- description engagement letter (PDF or Word)
- menu_book Assigned playbook with review questions
- assignment Review task with metadata (client, case handler, priority)
UNOY works
- → AI reads the contract and structures the content
- → Review runs question by question, algorithmically in a loop
- → Tight context per question = more accurate results
- → Algorithmic evaluation of every answer against the playbook rules
- → The review result is assembled and submitted for sign-off
Result
- ✓ Structured review result per engagement letter
- ✓ Every playbook question answered and assessed individually
- ✓ Deviations and red lines clearly flagged
- ✓ Sign-off status and control workflow documented
- ✓ mass processing: hundreds of contracts in batch
minutes
instead of hours per review
100 %
algorithmic precision
Batch
Mass processing supported
3 modules
Review, administration, playbook editor
The decisive difference to a chatbot: our review workflow gives us algorithmically exact results — question by question, traceable and scalable. Exactly what we need given the volume of engagement letters we review.
Supervision model
Self
You build your own digital twin that delivers results. Every rule defined by you, every result under your control.
Your digital expert — built by you.
Partner
UNOY's expertise is included. You work together with the system — we set it up, you keep control.
Expert knowledge included — we set it up, you keep control.
Supervised
The entire work is delivered as a finished result — with legal accountability through Skribe Attorneys.
You hand off the case — we deliver the finished result.
Why a workflow beats a chatbot.
The advantage lies in the architecture of the review process. A chatbot takes a playbook and a contract as input and delivers an overall assessment. That sounds efficient but is unreliable: the context is too broad, the AI summarises instead of reviewing, and the result can't be traced question by question. For a firm that needs solid reviews, that doesn't cut it.
The UNOY workflow takes a different route: every review question from the playbook is asked individually against the contract — in a loop, driven algorithmically. The AI is used only to extract the relevant contract passages, not to assess them. The assessment happens algorithmically against the playbook rules. Tighter context, more accurate results, complete traceability.
This separation of AI extraction and algorithmic assessment is why the workflow scales. Hundreds of engagement letters can be processed in batch without the quality of the result dropping. Every review result is structured, every deviation flagged, every sign-off documented. On top of that, playbooks can be edited and tested directly in the system — before they go into production.
What we learned from the project.
First: AI alone isn't enough for solid contract review. That was the central insight from this project. AI is excellent at extracting and structuring contract content. But the assessment — whether a clause matches the playbook rules — has to happen algorithmically, not via a language model. Only that produces results a firm can use as a basis for decisions.
Second: tight context is decisive. If you put an entire playbook and an entire contract into a language model, the context is diluted. The question-by-question loop — one review question, one contract section — keeps the context tight and the results precise. That isn't just technically cleaner, it's what makes mass processing possible in the first place.
Third: a review workflow without administration is only half the solution. SKW Schwarz needed not just the review, but also the orchestration: who issues the engagement, who approves it, who signs off. And the playbooks themselves had to be editable and testable — not a static document, but a living rule set. Only all three workflows together yield the complete solution.
Technical deep-dive: how we built it with UNOY
expand_more
Review workflow with algorithmic loop
Every review question from the playbook is checked individually against the engagement letter. The loop iterates through every question and collects the results in a structured form.
AI-assisted contract extraction
A language model reads the engagement letter and extracts the passages relevant to each review question. The AI structures, but does not assess.
Algorithmic assessment against playbook rules
The assessment of each answer happens algorithmically — not via AI judgement. That produces solid, reproducible results.
Mass processing in batch mode
Hundreds of engagement letters can be submitted as a batch and processed in parallel. The question-by-question architecture scales linearly.
Administration workflow for review mandates
A dedicated workflow for creating, allocating and tracking review mandates. Every engagement has a defined lifecycle.
Sign-off and control workflow
Review results run through a defined sign-off process. Four-eyes principle, state transitions and complete documentation of every decision.
Playbook editor with test function
Playbooks can be edited directly in the system: add questions, adjust rules, set thresholds. Every change can be tested against sample contracts before it goes live.
Context control via question-by-question architecture
Instead of matching the entire contract against the entire playbook, the context is kept deliberately tight per iteration. That reduces hallucinations and improves accuracy.
Audit trail and result documentation
Every review, every sign-off, every playbook change is logged. Complete traceability for compliance and internal audit.
More customer cases
Puttinger.Vogl
3,000 claim filings — structured and verified
Juridicum Vienna
Tax law meets technology — UNOY in KU TaxTech at the Institute for Tax Law
SKW Schwarz — cease-and-desist letters
Create cease-and-desist letters with COIA — the lawyer steers, skills define the rules
Review contracts, redline, sign off. In minutes instead of days.
30 minutes. Your contract review. No slide deck.