Request A Demo

Legal AI: What can it do and what can it not do?

The expectations of AI are enormous. How can users use AI successfully, what are its limits? Information provides clarity, which is why we have provided you with an overview of Legal AI’s competencies.

Legal Artificial Intelligence (AI) refers to the use of AI technology in the legal system. Its areas of application differ: Legal AI systems can be trained to make predictions about cases or to facilitate administrative processes. However, AI in the legal system can also be specialized in assisting with research or assisting with document review and analysis. Contract review is particularly well suited to the use of Legal AI, as it is here that frequently recurring patterns can be identified and repetitive work can be accelerated by the use of AI.

How does Legal AI work in contract review?

What can Legal AI do and what can it not do? AI applications work particularly well where patterns are repeated. That is why the use of AI is particularly useful in the legal sector: Contract review is particularly well suited to use AI in the review process. This is because specific information is selected here by an automated Contract Review from a large amount of data, compared with learned patterns and summarized for the user.

We have compiled concrete examples for you and used them to explain the functions and possibilities of Legal AI in an understandable way.

Legal AI in the training process

In order for Legal AI to be applied, recognize the right patterns and highlight deviations, the artificial intelligence is trained using machine learning. In the process, the algorithm is trained with hundreds of documents in different languages.

During the training, legally trained experts analyze the individual sentences of the documents and define relevant clauses. In this way, clauses such as applicable law, governing law, guarantee period and warranty period are stored and recognized on the basis of their similarity to trained patterns in contracts. Usually one sentence marks one clause, but it can be the case that two clauses are recognized in one sentence.

CTA Legal Tech Guide ENExample sentences in the training:


One clause per sentence:

1. Governing Law - Anwendbares Recht

EN: “The law governing this agreement is Swiss law.” 

This sentence is labeled as governing law (clause governing law).

DE: “Dieser Vertrag untersteht ausschliesslich schweizerischem Recht.” 

This sentence is labeled as governing law (Anwendbares Recht).

2. Warranty Period - Gewährleistungsfrist

EN: “This Warranty covers the defects resulting from defective parts, materials or manufacturing, if such defects are revealed during the period of 12 months since the date of purchase.” 

This sentence is labeled as warranty period.


DE: “Die Verjährung von Mängelansprüchen beginnt mit der Abnahme der Lieferung und endet nach 12 Monaten.” 

This sentence is designated as warranty period.

Zwei Klauseln pro Satz:

3. Liability Exclusion und Force Majeure - Haftungsausschluss and Höhere Gewalt

EN: “Neither Party shall be liable for any damage that is caused by an event unforeseen at the time of signing this Agreement and whose occurrence or consequences such Party can neither avoid nor overcome by reasonable means ("event of Force Majeure"), including, but not limited to hostilities, riot, explosion, fire and flooding.” 

This sentence will be labeled as the clauses liability exclusion and force majeure.

DE: “​​Keine Partei haftet für die Nichterfüllung oder die verzögerte Erfüllung ihrer Verpflichtungen, wenn diese auf höhere Gewalt zurückzuführen sind (dazu gehören unter anderem staatliche Maßnahmen, Streiks, Krieg, Überschwemmungen, Pandemien oder Embargos).” 

This sentence will be labelled as the clauses Haftungsausschluss and Höhere Gewalt.

Contract Review: How does Legal AI work?

During the contract review, the contract review AI is presented with data, which the artificial intelligence has not processed before. The algorithm analyzes every single unknown sentence from the documents presented and compares it to the known patterns. In this way, the algorithm searches for similarities between the unknown sentence and the trained structures. Any sentence which has similarities to the trained patterns is identified as a clause.

Example sentences: Training and application

1. Governing Law - Anwendbares Recht

EN: The law governing this agreement is Swiss law.” 

This sentence has similarities to other sentences that have been labeled as governing law. It is therefore identified as governing law.

DE: Dieser Vertrag untersteht schweizerischem Recht” 

This sentence has similarities to other sentences that have been labeled as governing law. It is therefore identified as a governing law clause.

In addition, company settings can be adjusted so that the contracts can be analyzed internally and it is possible to check whether clauses comply with the defined corporate policies. To do this, you define the relevant policies of your organization centrally in a Contract Playbook. Based on this, the contract review software can decide whether the clause is acceptable for your company or not.

In the next step, the algorithm can identify concrete values (such as locations, maturities or amounts) within a sentence - for example, the law of Switzerland in the above example. Based on the previously adjusted company settings, the contract review software knows whether the value found (e.g., location Switzerland) is acceptable for your company or not.

If inadmissible clauses are found, the software instructs the user to correct the text with the standard text of their company. If important values or clauses are missing, the software prompts the user to insert the missing text using the company’s standard text.

Competencies of AI: What can Legal AI do?

  1. Standardized contracts or concepts:

    The more similar the training sentence is to the sentence to be reviewed, the better the performance of an algorithm. That’s why the AI performs best on contracts such as NDAs, DPAs, GTCs and EULAs. Frequently recurring concepts are also well recognized in more complex contracts such as limitations of liability, warranty disclaimers or risk transfer clauses.

  2. Finding inconsistencies in long contracts:

    Especially with long contracts such as a 20-page DPA, it is more difficult to identify relevant contract language on different pages or in unexpected places, such as in the preamble. It also requires effort to ensure that no important clauses or values are missing from the document. For these tasks, an AI only needs a few seconds.

  3. Extending competencies to less experienced reviewers or new employees:

    Complicated contract language and different company policies for different negotiation scenarios - this can be overwhelming for less experienced reviewers. With the intelligent assistance of an AI-powered tool, the task of contract review becomes much more manageable and does not require an experienced attorney every time.

Legal AI and humans: What can Legal AI not yet do?

  1. Identify unusual or company-specific clauses:

    An algorithm needs a large number of training examples to provide reliable results. If it encounters unusual language that is very different from anything it has seen before, the AI results may not always be correct - this is true, for example, of poorly translated clauses or a clause, which uses many internal company abbreviations.

    It is not possible to predict when and where such unusual texts will appear. Therefore, human beings should always double-check the contract language that could not be detected by the algorithm.

  2. Make automatic corrections:

    Generating grammatically appropriate correction texts when an error has been found in a document belonging to a third party is still very advanced for any AI solution today.

    Corrections require that the algorithm recognize the correct names of the parties (e.g., recipient), correctly use singular and plural when using verbs and pronouns, recognize the correct gender and the correct case of words in languages such as German. The correction of texts is still best done today by human beings.

  3. Understand references to documents or legal regulations:

    Contract clauses often refer to other clauses within the same contract (e.g. notwithstanding the terms described in clause 2.1), to related contracts (e.g. according to the liability section in the framework contract) or to legal regulations (e.g. according to Section 970 General Civil Code (ABGB)).

    As a human reviewer, you will be able to identify the relevant clauses and check whether there are contradictions between texts. This task is still very advanced for any machine learning solution since it requires not only very thorough text recognition but also a basic understanding of the law.

How does Legartis work?

Step 1

The users set up a Contract Playbook in the Legartis app for each type of contract. They go through a list of pre-written clauses in English and German to define what is required and what is acceptable for their organization.

Step 2

The users go to Microsoft Word, open the contract to be reviewed (also PDFs after conversion) and click on the Legartis logo.

Step 3

AI checks the contract within seconds. The users then go through their contract using an intuitive to-do list, which shows them which parts they need to review or correct. After completing all tasks, the users can be sure that all company policies have been reviewed and no important clauses are missing.

CTA Legal Tech Map EN


FAQs about Legal AI

  1. Do I still need to read the contract if I use Legal AI?

    Yes - Check your documents. We strongly recommend that you continue to read your contracts when you use Legal AI. AI is getting better and better over time and covering more and more legal concepts. But there are still many ways to write contracts. This can confuse an algorithm. Just as with a spell checker, you should make the final decision and reject any suggests with which you disagree.

  2. How can I save time then?

    With the intelligent support of contract review software, you can save time in three ways.

    An algorithm can find inadmissible or missing terms in seconds. This not only saves you time because you no longer have to perform this final review yourself. It also ensures that nothing is overlooked in the process.

    When analyzing a third party’s contract, reviewers often compare the third-party text with their own templates. Switching between documents and searching for matching paragraphs can take a lot of time. The AI-powered review tool makes all required information available as needed.

    Highly standardized contracts such as NDAs or DPAs do not always need to be reviewed by an experienced lawyer. By providing step-by-step support, Legal AI also enables less experienced reviewers and professional staff to independently review contracts. The legal department only needs to be contacted in borderline cases.

  3. What if the AI makes an error?

    The AI may misclassify or “overlook” a text segment every now and then - if such a case occurs, this can be reported directly to Legartis in the software. Legartis evaluates these errors and uses them to further improve the AI. In this way, we make sure our AI learns from its errors and constantly improves. Normally, after only one more training session at least 90% of the errors are solved and do not occur again.

  4. Can Legal AI also help me if I need to check topics or clauses, which the AI has not yet learned?

    Yes! Legartis offers the possibility to also store checkpoints and standard texts for user-defined clauses in the Contract Playbook, on which our AI has not yet been trained. Even if the presence and content of such clauses in the contract must be checked manually, this function provides guidance on which parameters should be checked and ensures that nothing is forgotten. In addition, we are constantly working on the expansion of our offer and are constantly training new clauses.

  5. Can Legal AI also help me if the contracts I want to review are special and not highly standardized.

    Yes. Even special, complex contracts contain, at least in part, recurring legal topics such as liability provisions or regulations on jurisdiction and termination rights that can be identified by AI. “More exotic” topics can be covered by means of user-defined clauses, which also lead to time savings and increased security during the review thanks to their checklist function. If you are looking for solutions tailored to specific topics, industries, or types of contracts, our experts will be happy to discuss possible solutions with you as part of a development partnership or a special design!


Key Takeaways

Legal AI can accelerate processes in the legal department and especially during the contract review and relieve your legal team. Based on extensive training, the algorithm is able to recognize and identify known structures and similar patterns in new documents. In this way, AI can reliably check contracts in particular that are designed according to standardized patterns and summarize the results for you.

But AI can’t do everything yet: Despite the algorithm’s performance, you should always read your contracts as well and leave the final decision on the review results to humans. At the same time, you can expand the contract review competencies, delegate them to professional staff and review them in a structured way using to-do lists. When AI and humans work together, you achieve the best results. 

Would you like to know the best way to implement Legal AI in your company? Then read our Legal Tech Guide now and find out the 7 most important steps in introducing Legal Tech

Recommended Articles

Subscribe to the Legartis Newsletter

Sign up for our newsletter and benefit from the latest insights from the world of LegalTech, artificial intelligence and contract review automation.

Subscribe Here!