The Complete Guide to Contract Drafting Technology
Edition 2024
Download as PDF file.
Successfully implementing a document automation tool will provide a significant boost to efficiency, quality, and consistency of the first drafts lawyers produce. Unfortunately, most document automation tools stop being useful after the first draft has been produced. Depending on the automation power of the tool being used, the type of document being drafted, the negotiation power between the parties, and potentially a few dozen other factors, document automation can create a first draft that is within 60-100% of the final version of the document you aim to produce. That potentially leaves a lot of work left to be done.
Nevertheless, this is a significant boost to the default of just drafting on the basis of the last precedent you or your colleagues remember using in a similar type of situation. In our experience, these precedents are first drafts that contain anywhere between 40-60% of the content retained in the final version.
Full disclosure: these are not scientifically researched numbers, in the sense that we do not have peer-reviewed research to point to. Consider these somewhat reliable estimates – they are the product of decades of experience using automation software as lawyers and selling it as legal tech vendors – but take them with a grain of salt.
Either way, you’ll find that some additional tweaking and tailoring may be necessary to go from first draft to final version. Lawyers will recognise this as the typical process of
Below, we look at the tools that assist in these stages of the drafting process.
Every lawyer knows the arts-and-crafts process of finding clauses in other precedents and copy/pasting them into a new document. It’s a terribly repetitive and time-consuming process that forces you to search through your emails, the organisation’s document management system, and the memories of colleagues to find useful material.
The search functionality in the document management systems, while a prime candidate for making this information easily accessible, usually leaves a lot to be desired because of inherent slowness (type in keyword, wait for the system to present a list of documents, double click on each document, scroll to a hopefully relevant paragraph, and so on). And since most organisations don’t invest in curating this information centrally, many lawyers resort to building their personal collection of preferred precedents. Some even build a structured collection of individual clauses.
Today, the legal tech market boasts a range of different tools that help lawyers find useful clauses from the collective drafting knowledge of their team. We see two different approaches these tools take to making that information available.
Much as the name suggests, clause libraries allow lawyers to make their preferred clauses centrally available in a structured format, which takes the form of a collection of folders and sub-folders that yields a comprehensive overview of available information.
Users can search for useful material in this library with the help of keyword searches, or by opening individual folders and browsing through the collection therein.
Clause libraries also allow users to augment their clauses with additional information like the indication of which party the clause favours, whether a clause should be used as a standard clause or a fallback clause, in which jurisdiction the clause should (not) be used, etc. – the type of information you wouldn’t immediately be able to glean from just reading the clause.
While a structured clause library can be a tremendous resource for lawyers everywhere to tap into the premier drafting knowledge of the organisation, such a library does require manual labour. Lawyers have to contribute knowledge when the library is still being built, and they have to periodically update clauses in light of regulatory and jurisprudential changes.
As a result, they are primarily interesting for legal teams that have knowledge management high on their agenda – e.g. law firms with a dedicated knowledge management function or in-house legal teams focused on bringing consistency and standardisation in their legal drafting as a risk management exercise.
Efficiency and Time Savings: Clause libraries significantly speed up the legal drafting process by providing extremely quick access to a wide range of vetted clauses, often within mere seconds.
Quality and Relevance Issues: The quality and relevance of clauses in a library can vary, and some clauses may become outdated if the library isn't regularly reviewed and updated. This can lead to the use of inappropriate or legally outdated language in contracts.
Consistency and Standardization: By using standardized clauses, lawyers ensure consistency across various documents and contracts.
This uniformity helps in maintaining a professional standard of drafting and ensures individual lawyers across a team are aligned on how to deal with different legal options or commercial negotiation positions.
Initial Setup: Establishing a clause library requires a significant initial investment of time and resources, which can be a burden for smaller law firms or legal teams that don’t already have a culture of knowledge sharing in place.
That said, this effort should not be overestimated. Clause libraries focus on quality, not quantity. The average clause library for a specialized department will usually not contain more than 500 clauses of that team’s best clauses and this collection will come about organically, with clauses being added on-the-fly as lawyers come across useful material. If a team of 10 lawyers each adds 5 clauses to the library per week, then a maximum of 10 weeks is needed for a “complete” clause library.
Up-to-Date Legal Language: Clause libraries can be regularly updated to reflect the latest legal standards, regulations, and best drafting practices. This ensures that all documents drafted using the library are compliant with current laws and that the entire team is easily brought up to speed on important changes.
Maintenance effort: Lawyers need to be able to trust that the content in the library is good, up to date, and from a reliable source. To continually earn that trust, clause libraries must be periodically maintained. This requires time, which many lawyers don’t have.
To get started with curated clause libraries, you don’t need to break the bank to buy a shiny tool. Many lawyers in your organisation are probably already building their own curated clause libraries with basic tools in your MS Office suite, like Word, Excel, and Sharepoint.
Obviously, these tools are not purpose-built for legal drafters so they will fall short on a number of fronts. Nevertheless, they may already provide an easy and cheap place to start. We actually wrote an entire guide on creating clause libraries, including those you can build with tools you already have lying around. Check it out.
Litera’s Content Companion (previously “Clause Companion”) allows you to create a standardised library of clauses, email snippets, paragraphs in legal opinions, etc. This library consists of a personal library, a firm library, and any practice areas.
Content Companion’s primary benefit comes from the fact that it lives inside lawyers’ trusted MS Word environment and from its inclusion in the Litera suite of legal tech products, which include proofreading, reviewing, and document assembly tools. Beyond that, there is little attractiveness to Content Companion. The lawyers we spoke to mentioned sky-high license pricing (over 600 USD per year) and limited innovation on this product. For example: it is currently impossible to create libraries in Content Companion with folder structures that are more than 3 levels deep.
Perhaps most importantly, however, is that Content Companion still relies on the old plugin architecture we discussed above. With limited support and expansion for this product, as well as an outdated technical setup, it is likely that Litera does not consider this product important enough to continue investing in.
Much like Content Companion, ClauseBuddy allows you to create libraries to store your standardized clauses, emails, paragraphs, etc.
ClauseBuddy goes beyond the default functionalities of Content Companion in a number of ways, however:some text
Many legal teams, law firms in particular, have incentives and cultures in place that prevent them from spending a lot of (unbillable) time in setting up clause libraries and creating templates. For those teams, dismantling and rebuilding the broken process of contract drafting may not be possible, but there are still a few tools that can assist in making it less painful.
Those tools, which we’ve called “clause extractors” crawl through large repositories of precedents and dissect them into individual, searchable clauses. These can be searched on using keywords and metadata attached to the original document that the clause was a part of, as well as metadata made available through the platform itself (e.g.: “how often is this clause used, how many more are there like it, etc.”).
The appeal is obvious – this technology meets lawyers on their home turf, aligning neatly with the way they already do legal drafting. Instead of manually juggling through dozens of templates to find useful clauses, clause extractors provide an easy interface to search through potentially decades of legal drafting knowledge.
Clause extractors are most beneficial in scenarios where legal professionals know exactly what they are searching for. If a lawyer remembers an exact sequence of keywords from a clause drafted years ago, then these tools can quickly locate the exact match or very similar clauses within the database. This also makes clause extractors less suited for exploratory searches where the user is "just browsing" or not entirely sure of which clause they need. In such cases, the sheer volume of clauses available just makes for a bigger haystack to find a needle in. The time saved by not curating a clause library is then just lost again diving through said haystack, and reading each of the clauses in the haystack.
Finally, it should also be mentioned that clause extractors are potentially risky from a compliance perspective. After all, they require the complete transfer of your chosen selection of precedent documents (or indeed, the entirety of your document management system) to a vendor’s server in order to be able to cut the documents into individual clauses. From a cybersecurity, confidentiality, and data protection standpoint, this presents a lot of potential issues.
We aim to be fully transparent about this so for our own clause extractor tool, we’ve written a complete breakdown on potential compliance issues here.
Improving a broken process – Clause extractors significantly reduce the time and effort required to search for specific clauses within a large volume of legal documents. Since clause extractors approximate how lawyers already do their legal drafting, it will also feel like a natural fit.
Compliance – Despite their advantages, clause extractors pose significant challenges in terms of compliance with security, confidentiality, and data protection laws. Especially with tools that link to your document management system, you are essentially making available every document your organisation has ever produced to a third party. Ensuring that these tools comply with relevant data protection regulations and safeguard client confidentiality is paramount. This requires rigorous security measures and often a detailed assessment of the legal technology provider's practices and policies.
No upfront effort for lawyers– Especially for those clause extractors that connect to your document management system, only the IT team has to spend time getting the tool up and running, but this pain will not be felt by the average lawyer in the team.
Limited use case – Clause extractors are primarily useful when you are “sniping” for clauses that you know you have drafted in the past, because you can then use an exact sequence of keywords that you remember from the clause (or the name of the client) to instantly surface what you need.
If, on the other hand, you are merely trying to find some drafting inspiration, then the time-savings will be limited, as you still need to browse through potentially hundreds or even thousands of clauses. Well-organised quality-based libraries will then be much, much faster: think seconds (quality library) instead of a minute (clause extractors) instead of several minutes (no dedicated tool used).
Incentive alignment – While more applicable to attorneys than in-house lawyers, it’s important to note that clause extractors align much better with the incentives placed on lawyers in the average legal team. Their main incentive, after all, is reaching their billable target. Compared to the time spent building clause libraries, the time spent searching for that needle in a haystack is time that can be billed directly to a client.
No institutional knowledge gained – Since clause extractors are primarily useful for retrieving material from a lawyer’s own drafting history, an opportunity is missed to tap the collective drafting knowledge of the team. Sure, the clauses made available by the extractor will be drawn from many lawyers and many documents, but the background knowledge behind why a clause came to be written the way it was, is typically lost to everyone but the people involved in the drafting. That knowledge vanishes if those people leave the organisation.
Also, do not underestimate how daunting the hundreds of search results of clause extractors can be for inexperienced lawyers, and how easily they will choose the wrong clause, because those extractors do not offer guidance.
Henchman has made headlines in recent years after raising a total of 10 million EUR in funding for its clause extractor tool of the same name. While not the first clause extractor to hit the market, they have in many ways paved the way for this type of technology.
Like ClauseBuddy, Henchman also focuses on extracting clauses from a large collection of precedents. The main difference is that it adopts a “data maximisation” approach. Henchman integrates directly with virtually every purpose-built legal document management system (e.g.: iManage, NetDocuments,…) and several non-purpose built document management systems (e.g.: Dropbox, Sharepoint,…) to ensure that the entirety of a team’s precedent contracts are covered in the extraction. It then transfers all these files to its own servers, where it extracts the clauses from each file that looks like a contract, and stores them in a database optimised for text searching. It will then periodically synchronize your document management solution with its own clause database, so that new files (Word-documents, email attachments, etc.) that get added to your document management solution will automatically become available in Henchman’s clause database.
The beauty of Henchman’s approach is that all of this happens automatically, without any involvement of lawyers. Even though the IT-department will have quite some work to do to establish integration towards Henchman's servers, and the security department will probably also need to be involved (e.g., to drill a hole in the organisation's firewall to allow for Henchman's document extraction, or to exclude some folders that should not be in scope of Henchman's contract crawling), little billable time is spent here.
Aside from the inherent security, data protection and confidentiality issues of a data maximisation approach, the main downside of Henchman is that it does not really venture beyond mere clause extraction and does not contribute towards more robust knowledge management, e.g. to increase quality and offer better guidance to junior lawyers. After all, the product focuses on maintaining existing contract drafting processes, optimizing them without addressing the underlying chaos that may exist. Several legal teams therefore expressed their reluctance to give junior lawyers access to Henchman, fearing that the wrong clauses will get selected, due to the tool's abundant search results, over-reliance on a clause's popularity, and lack of context/guidance for individual clauses.
Tip: see also our in-depth review of Henchman.
ClauseBuddy is a MS Word plugin that is part of the ClauseBase suite of legal drafting tools. Its clause extractor module gives users access to millions of sample clauses from the SEC’s EDGAR database and also allows users to upload their own precedents to extract clauses from.
Users can search for clauses using keywords and switch between different browsing windows – one where clauses are presented individually and one where the clauses are shown in their original context. Entering a clause into a document automatically adjusts its style, numbering, paragraph spacing, indentation, etc., like a chameleon adjusting itself to its environment.
One of the biggest differences in design choices for ClauseBuddy compared to other clause extractors is that ClauseBuddy deliberately does not transfer all available material in your document management system for use in your clause library. Instead, ClauseBuddy will require users to make a selection of documents to upload.
This can be considered a benefit or a drawback. On the one hand, it forces users to already engage in some knowledge curation which will significantly improve the quality of the search results, as users will not be confronted with thousands of clauses from inferior sources (e.g., counterparty or outdated documents). On the other hand, this takes up preparation and maintenance time when many feel that the entire purpose of a clause extractor is to not have to waste any non-billable time.
The design choice fits within the overall strategy of ClauseBase as a company, though, which is generally geared towards legal teams who want to distinguish themselves through thorough knowledge management. ClauseBase explicitly pitches the clause extractor within ClauseBuddy as a first step towards this goal, with legal teams starting from the chaos of a pile of precedents and gradually and organically moving towards an ordered clause library.
Ever since the launch of GPT-3 in November of 2022, the legal sector has been consumed with Generative AI. Conferences, webinars, blog posts – you can’t swing a dead cat without hitting Generative AI.
With research like that of Goldman Sachs claiming that AI could automate 44% of legal tasks and recent research from Onit showing that Large Language Models (LLMs) outperform senior lawyers in speed and even quality of review, it makes sense that lawyers everywhere are keeping a close eye on how this technology develops.
In the wake of this surge of interest, hundreds of new AI solutions have hit the market – many of them expansions on existing products, but most of them new solutions that have an LLM integration as their key focus.
The Generative AI hype may have started cooling down in early 2024, but we consider this is a good thing. Lawyers are now finding the real solutions to actual problems and LLMs are beginning to provide tangible value to legal service providers and their clients, free from inflated expectations and marketing hype.
In the spirit of its mission to create a legal drafting toolbox for lawyers, ClauseBase has launched tons of experiments with implementing Generative AI into its offering. Here are the use cases which we have found provide real value to the contract drafting process and a few that are mostly just hype.
LLMs are very good at creating summaries. In fact, they are so much better than humans, that humans actually prefer summaries created by LLMs over summaries created by fellow human beings.
Except for a few very specialised legal domains, LLMs should therefore become the go-to approach for (first drafts of) summaries, similar to how translation engines have become a household staple within law firms.
LLMs are very useful for receiving drafting inspiration, especially when useful precedent clauses are hard to find or a useful precedent clause needs to be tweaked a bit more.
While they accelerate the process, human review remains essential. In fact, this technology should only be used for drafting in legal domains that you are an expert in, so you can accurately verify its output.
Additionally, they are surprisingly useful in negotiations. Explaining to the LLM what compromise the parties have agreed to in an oral negotiation is usually enough to receive a redrafted clause.
This one hurts because we have to be honest – we developed our own tool to leverage GPT4 in drafting contracts from scratch and the results are pretty underwhelming.
To start, LLMs are still incapable of ensuring compliance of their output with different laws in different jurisdictions. Not an insurmountable problem – all LLM output should be reviewed by an experienced lawyer. But it does mean that you have to do a review completely from scratch for every individual document generated and that there are guaranteed to be problems each time.
After all, inconsistency in generated output is part and parcel of the Gen AI experience. Much like with clause extractors, you save a lot of time up front in the first step of the drafting process, but a lot of that time is then lost in the next steps.
Funnily enough, this drawback to LLM-generated contracts has caused a surge in renewed interest for traditional document automation technology.
As set out above, LLMs are still incapable of reviewing a document against a specific legislative framework. Asking “review this contract under German law” will get you a response, but it won’t be one that is guaranteed to have actually checked the contract against all possible issues under German law. Some tools have begun to create databases filled with statutes and case law to try and feed the LLM with, but the results have been lacklustre at best.
On the other hand, recent advances in LLMs have meant that their commercial reviewing capabilities have taken a big leap. We actually launched a tool for reviewing and markup in early 2023 to test this use case. At the time, its performance was pretty abysmal. But then, with the upgrade to GPT-4 Turbo and a renewed approach where the focus was on explaining to the LLM what issues it needed to check for, the results were much, much better.
This is the holy grail of Generative AI. The idea is that you would be able to chat with the entirety of your firm’s document history and ask it to provide an answer to a specific legal question based on past work, generate a contract based on a precedent, etc.
We are not ruling out that this may one day be reality, but in May of 2024, we’re not there yet.
The main impediment to achieving this is the limit of context windows of Large Language Models. In short, an LLM can focus only on a limited amount of text at one time. That limit is constantly being pushed, but it’s not quite at the level yet where it can read through thousands of pages of text spread out across a database. New LLMs such as Google’s Gemini have a context window of about 700,000 words, but every answer in the chat conversation takes minutes to calculate and easily costs several dollars.
Techniques like Retrieval Augmented Generation have helped to circumvent this limitation, but they also serve to diminish the quality of the output. Until the context window is pushed to the thousands of pages, and the speed and cost issues are resolved, the dream of chatting with your DMS will not live up to the reality.
While LLMs may not yet have conquered the database frontier, they most certainly have conquered the document frontier. Recent advances in the context window limit have meant that most GPT-grade LLMs can now reliably read through an entire legal document.
This allows you to surface relevant information from said document, request explanations or summaries, redraft terminology across an entire document, and more.
Aat the top of most lawyers’ Christmas wish list in 2023 was Microsoft Copilot, the GPT-4 integration into MS Word. They actually got their wish quite a bit sooner, since MS Copilot launched to a general audience on November 1st, 2023. Up until then, Copilot had only been available for large enterprises.
While not a purpose-built tool for contract drafting, Copilot has some very nice entry-level features for that purpose. Front and centre is its ability to draft clauses based on a short prompt – much like you can already do with a tool like ChatGPT. It also allows you to engage with your document by asking questions about it (e.g.: “what are the obligations of the parties in case of termination?”) or asking for summaries. In our own tests, all of these features yielded positive results.
That said, lawyers should take care what version of Copilot they are using. Copilot is currently quite lacking on a few key features for contract drafters and only those with beta access to its more advanced versions have more capabilities.
Most notably, Copilot is incapable of redrafting a selected clause, unless that clause is shown in plain text (i.e.: not in a numbered paragraph). This of course makes it virtually unusable for contract drafting purposes, since most clauses in a contract will always be included in heading or numbered paragraph.
This omission is quite curious. Microsoft’s own tool is incapable of engaging with anything other than plain text in Word while many purpose-built legal drafting plugins for Word are capable of that. Furthermore, even if you change the numbered paragraph to plain text to allow Copilot to redraft the clause, you are not given an option to provide instructions for that redraft. Copilot just starts rewriting your selected paragraph. This of course misses the entire point of redrafting a clause from a legal perspective. You’re not trying to replace the selected words with synonyms, you’re trying to find different legal nuance or a different commercial position, which should be explained to the AI before it does its thing.
We expect that some of these weaknesses will be resolved in time, but at a price point of 30 USD/user/month, lawyers should ask themselves whether the tool can really provide enough value to make the investment worthwhile.
As stated above, the ClauseBase team has spared neither effort nor expense since the launch of GPT-3 to set up experiments around different use cases for Generative AI and to test if they make a valuable addition to ClauseBuddy’s legal drafting toolbox.
We’ll be the first to admit that some of these experiments completely flopped with our beta testers and never saw the light of day again after an initial round of testing. Others continue to be a part of ClauseBuddy to this day.
Perhaps the most popular AI drafting functionalities of ClauseBuddy are drafting and redrafting. While ClauseBuddy users can already tap their precedent bank for legal drafting inspiration and rely on a curated clause library for standardised and vetted clauses, even these resources may sometimes fall short. In those situations, being able to explain to the AI in a few words what you’re looking for can be invaluable.
Instead of drafting a clause out of thin air, ClauseBuddy also allows you to tweak existing clauses with a short prompt. This “redraft” functionality can be used on your own clauses as found in the clause library or the precedent bank, or they can be used on clauses found in a Word document. ClauseBuddy can even analyse your clauses and provide hints for potentially interesting legal variations or additions to that clause, which it will then also draft for you upon selection.
On the summarisation side, ClauseBuddy significantly improves upon Copilot, by allowing lawyers to create custom summary structures (e.g., focusing only on conditions precedent in a corporate transaction) and specify the layout of the summary.
On the document reviewing side, ClauseBuddy offers the ability to define rules for how a document should be reviewed and then let the AI cross-reference a document against that ruleset. ClauseBuddy then also allows you to define what actions should be undertaken by the AI if it finds infringements of your own ruleset in what is essentially playbook automation in the truest sense of the word.
Aside from these functionalities, ClauseBuddy also allows you to perform more basic AI-powered drafting requests, like asking for an explanation. We’re happy to call them basic because they we will be the first to admit that they don’t offer anything that Copilot doesn’t do as well, but still provide a nice way to help draft explanatory emails to clients or counterparties.
Spellbook recently made headlines after having raised 20 million USD to expand its international reach and is a leading company in the field of Legal AI Assistants.
Spellbook, like most other Legal AI Assistants, is in essence a GPT-4 integration built inside MS Word with an interface that is optimized for legal drafting. It offers a wealth of different ways to leverage GPT-4 for typical text generation and text reviewing tasks that contract drafters frequently find themselves confronted with.
All the basic, to-be-expected, functionalities are there and in a very crisp and visually appealing interface: drafting a clause with a short prompt, redrafting a selected paragraph, asking for an explanation of selected text, etc.
Spellbook offers a few functionalities that go beyond what most AI Legal Assistants offer in that respect. For example: the interface will provide a few premade options to engage with selected text – instead of selecting text and writing out something like “make this more concise”, Spellbook offers you to do just that with a click of a button.
Spellbook is also strong on the reviewing front. Like Copilot and ClauseBuddy, it offers lawyers the option to query their document. Like ClauseBuddy, it also offers the ability to define rules for how a document should be reviewed and then let the AI cross-reference a document against that ruleset. Spellbook also intends to offer that functionality in the future on a whole data room, making the tool a potentially powerful addition to the due diligence process. It also allows lawyers to request a “points of negotiation” review, which means it will search a document for imbalances and provide a list of suggestions for how the clauses could be tweaked for a more balanced end result.
Spellbook gives no concrete information on pricing on their website but indicates in their FAQ section that they are exploring different pricing models. We talked to a few user of Spellbook and found that they were indeed paying different license fees. One solo lawyer we talked to paid 100 USD per month. An alternative legal service provider we talked to said Spellbook charged 200 USD per user per month, which is definitely on the very expensive side.
