AI Law, Risk & Governance

AI in Legal and Compliance: Use Cases, Approval, Accountability

Marin Niehues18 August 202510 min read

Woman handing over a document for signature

AI relieves legal and compliance departments where large volumes of text need review, provided a named person approves every result against a fixed review standard. Lawyers decide when predefined triggers occur. For executives, two decisions matter: which tasks suit AI, and how the workflow is designed so that responsibility and quality are settled for every case. The following sections set out the use cases and describe the organization that secures their use. This article provides an overview and does not constitute legal advice.

“AI and compliance” has two meanings. This article covers AI as a tool for legal and compliance work. The obligations a company must meet when it uses AI itself are described in the article on the EU AI Act.

AI creates value wherever a lot of text must be read, compared and classified before the legal assessment begins. The department receives contracts for review and queries from the business units every day. On top of that come reports from employees and internal policies that need maintaining. Every case starts with reading and classifying, and only then does the legal assessment begin. This is where AI comes in. Language models process text in seconds. They summarize documents, compare clauses against templates and sort cases by topic.

Five use cases lend themselves to a start:

  • Contract review: comparing clauses against the company's standard positions and flagging deviations
  • Document review: pre-sorting large data rooms by topic, for example in transactions
  • Regulatory monitoring: analyzing new publications from authorities and assigning them to the affected functions
  • Policy management: comparing internal policies with one another and with external changes
  • Case handling: pre-sorting and summarizing incoming queries. Whistleblower reports and client-related documents are subject to confidentiality and professional secrecy rules that determine the choice of system.

What these fields share is the division of roles. The AI prepares material. The lawyer assesses and decides. Working time shifts from searching to assessing, and this shift is the economic core of every business case in this area. Tasks whose result takes direct external effect are different, such as a binding statement to a customer or a submission to an authority. Here the AI provides a draft, and full responsibility for the text lies with the person who signs it.

Legal departments can draw on four tool categories: general AI assistants, specialized legal AI assistants, contract review tools and platforms for automating legal workflows. The choice depends on the use case and on how confidential the content is.

  • General AI assistants: language models under an enterprise license, often built into office software. They suit summaries and drafts without confidential client content.
  • Specialized legal AI assistants: applications tailored to legal work with research, drafting and document analysis. Internationally, Harvey is among the best-known providers in this category.
  • Contract review: tools that check contracts against a company's own playbook and flag deviations. Legartis is an example from the German-speaking market.
  • Legal workflow automation: platforms on which legal departments map requests, approvals and standard documents without programming. Bryter is a well-known example.

For selection, feature lists matter less than three tests with your own documents. Does the tool give the source location for every statement? Can your own playbook be stored? And where is the data processed? A trial with real cases over two to four weeks replaces any product demonstration.

An evaluation framework for applications: value, error impact, verifiability

The selection of the first applications follows three criteria. Value measures how much processing time a case type ties up today and how often it occurs. Error impact describes what happens when an error enters the workflow undetected: does it remain internal and without consequence, or does it reach contractual partners and authorities? Verifiability indicates how quickly a person can verify a result at the source.

The three criteria produce a sequence. Applications with high value and high verifiability, where an error stays internal, come first, such as pre-sorting documents and comparing contracts against standard clauses. Next come applications with medium error impact that require a four-eyes principle, for example preparing submissions. Applications with high error impact and low verifiability initially remain with the lawyers. They are added once measurement and approval are defined for them. Whistleblower reports do not belong in the first wave because of the confidentiality obligations.

The general approach to prioritization is described in KI-Use-Cases priorisieren. For legal and compliance, error impact and verifiability carry more weight than in other functions.

AI results are approved in three logged stages: an AI proposal with source reference, a check at the source by a named person, and a lawyer's decision when predefined triggers occur. This requires four roles to be filled:

  • Functional accountability: the head of legal or compliance decides on fields of use and approval thresholds.
  • Process accountability: a product owner prioritizes requirements and steers the further development of the application.
  • Quality assurance: lawyers review samples and document error types.
  • Operations: IT ensures the availability and access rights of the application.

On this basis, approval works in three stages:

  1. Proposal: the AI produces a proposal and gives the location in the document for every statement.
  2. Check: a named case handler checks the proposal at the source against a defined review standard.
  3. Decision: a lawyer decides as soon as predefined triggers occur, for example a deviation from standard clauses or a contract value above a set threshold.

Every stage is logged. The log records input and result as well as the reviewer and their decision. It remains available later for queries and internal audit.

Legal departments handle particularly confidential content. Before first use, the head of legal and IT security define which document classes may enter which systems. This classification is an organizational decision and belongs in the company's AI governance. The legal assessment of individual data flows in a specific case is a matter for legal counsel.

The costs consist of licenses per user, set-up with the company's own playbook, integration with contract management and document storage, and ongoing quality assurance. Specialized providers usually quote prices only on request, tiered by number of users and scope of features. The calculation therefore becomes robust only with an offer based on your own volumes.

For the business case, a simple calculation applies: reading time saved per case, multiplied by cases per year and the internal hourly rate, weighted by the share of cases that actually run through the application. This benefit is set against total costs over the term. Experience shows that introduction and operation cost at least as much as set-up. Where external support is added, it is typically billed in the German consulting market at around 1,000 to 2,500 euros per consultant day, depending on seniority.

How Niehues Consulting delivers such initiatives

Niehues Consulting brings AI into operations and builds organizations that deliver. Founder Marin Niehues is a management consultant with projects ranging from VC-funded scale-ups to DAX 40 corporations and assignments in 20+ countries on six continents. For initiatives in legal and compliance, the work follows the Niehues Delivery Model with its phases Understand, Build and Introduce:

  • Understand: business analysis of the review workflows, requirements from the legal department, business case and the selection and sequence of applications. Guiding question: who decides, who delivers, who uses?
  • Build: prototypes on real documents in the real workflow. Guiding question: does the prototype prove its value in the real workflow with real data, confirmed by the lawyers?
  • Introduce: transfer into operation with internal or external teams, along with clear workflows and user enablement. Guiding question: who owns operation from day one?

As delivery manager, project manager or product owner, Niehues Consulting takes over steering until the application runs in regular operation. How an AI application moves from prototype to daily use overall is set out in the article From pilot to scale. The entry point is the free 30-minute initial consultation. You can book an appointment here.

Frequently asked questions

Can AI replace lawyers in companies?

AI does not replace legal judgment. It takes over preparatory tasks such as review and comparison, and the lawyer is accountable for the result. For the executive team this means the department's capacity grows, and the role shifts from searching to assessing and negotiating.

Which compliance tasks are suitable for getting started?

The starting point is tasks with high verifiability and internal error impact: comparing policies, analyzing new publications from authorities and pre-sorting general queries from the business units. Whistleblower reports follow only once a confidentiality concept is in place. Here the value shows early, and the lawyers gain experience with the review workflow.

What is the difference between legal tech and AI?

Legal tech is the umbrella term for software in legal work, and AI is one technology within it. Legal tech also includes contract management, rule-based document automation and matter management. AI extends these tools with the ability to understand and draft text.

How accurate is AI in contract review?

Accuracy depends on your own playbook and your own contracts and can only be measured reliably there. A test set of 50 to 100 contracts that have already been reviewed has proven itself for measuring hit rate and missed deviations before launch. In operation, the reviewers' correction rate shows whether accuracy holds.

How is the quality of AI results assured in the legal department?

Quality is assured in the workflow. Every AI statement points to its source location, and lawyers regularly review samples. In addition, the team continuously measures how often reviewers correct a result. This rate determines whether a field of use is expanded.

Is the AI literacy obligation of the AI Act relevant for legal departments?

Yes, the AI Act requires companies that deploy AI systems to ensure their staff have sufficient AI literacy, and this applies to the legal department too. For lawyers, this mainly means knowing the limits and typical errors of the tools in use. At the same time, the legal department is often the function that helps shape this obligation for the whole company.

Who is accountable for an AI-assisted result in the company?

Organisationally, accountability lies with the person who approves a result and with the head of legal or compliance who defines the workflow. The liability consequences in an individual case are a matter for legal counsel.

Let’s talk about your initiative.

In 30 minutes we clarify where your organization stands and which step will pay off most now.

Request a consultation

30 minutes · free · no obligation