Legal AI product: Key success factors5 min read

Guest post by Sergii Shcherbak, CTO Maigon.io

 

Are you passionate about LegalTech and Legal AI? Do you have a product in mind that you would like to create? There are few things you need to know that will help you in your journey.

Starting from the development and launch of Maigon PPC, which analyses privacy policies for compliance with the GDPR and has been used in more than 70 countries, up until the recent release of Maigon DPA, a corporate solution which provides a comprehensive compliance review of data processing agreements, we at Maigon.io, a Legal AI company, have accumulated a lot of practical experience.

The below checklist distills the key TODOs to make sure your Legal AI product succeeds. And by “success” I mean, largely, a growing client base and strong customer retention.

Verified product-market potential. Discuss with existing or potential clients, preferably legal departments, what solution they need. What software tool would considerably simplify their daily legal work? If you have an NLP (natural language processing)-focused product in mind, ask what legal documents they spend a lot of time on, for example. Then outline main features that a potential product could offer, such as key information extraction, clause extraction overview, deviations analysis, etc. Discuss with the client which of the proposed functionality would be the most useful. Always be open to suggestions. The solution does not have to be overly complicated or comprehensive to deliver the value your clients are looking for.

Data. For any high-quality Legal AI product, a lot of carefully labeled legal data is necessary. And a considerable amount of legal resources is required to make sure that the data meets these criteria. Lawyers’ work is far from cheap, so allocate enough resources in advance to build datasets of high quality. For instance, a generic NLP-focused product which reviews contracts of a certain type would require at least hundreds of such contracts collected. If it is a basic clause extraction tool, each of these collected contracts would need to be split into clauses and the clauses put into separate “data boxes”, all done manually.

Compute. If it is indeed a Legal AI product, training neural networks of a modern architecture like BERT requires a powerful GPU-powered machine. You could use a cloud platform for this: many cloud infrastructure solutions provide on-demand GPU instances where you can train your deep learning models. In practice, however, it is more cost-efficient to have at least one dedicated powerful PC in-house. Again, this all depends on the machine learning model or neural network architecture you choose for the product: for some solutions, a simple scikit-learn classifier could be enough, and you do not need a GPU instance for that at all. In the end, it is all about the value and quality of the product, not the fancy architecture it is using.

Best talent. Make sure the engineers building your product have a good track record. If you outsource the development, ask the candidates to share their ML development portfolio. Some people claim to be “AI experts” after having read a couple of blogposts about neural networks. Practical experience is paramount: this will make sure that your resources are spent adequately and the set milestones are met, while best development practices are observed and the resulting solution is maintainable. At the same time, be realistic about your budget and do not hire more people unless absolutely necessary. “The more developers, the better the quality” principle does not always work — as Fred Brooks once put, “What one programmer can do in one month, two programmers can do in two months.”

Client participation in development. A few potential clients should be part of the product development from day 1. This client involvement makes sure that your time and resources are spent on something that really matters to the client, who ultimately will be the ones purchasing the product. Organize a recurring online meeting, once a week or bi-weekly, which will help the clients see the incremental progress you are making, appreciate the value build-up, and give them part of creative control over the end-product. At the same time, be open and realistic in discussing client expectations: nobody wants to anticipate an A and receive a Z.

Reliable infrastructure. A secure and robust infrastructure for your product will make sure the client data is safe. And to make sure the product always matches the client’s workload, the infrastructure must be easily scalable. For example, Amazon Web Services can scale the amount of instances (servers) hosting your product depending on the amount of incoming traffic. Make sure you have reserved an adequate budget for the infrastructure, it does not come very cheap.

Continuous improvement. Having built a product and signed up your first customer, be ready to actively work on the product improvement. The launch is just one of the first steps on the journey! Over time, there will be lots of client suggestions and requests that you will need to address in your product. Also, the legal AI world is fast-evolving, therefore constantly monitor for regularly appearing new technologies which would considerably improve the quality of your product and elevate it to the competitive standard. For example, as regards an NLP-focused solution, recurrent and convolutional neural networks (RNN and CNN) which were considered state-of-the-art just one year ago have now been replaced with an attention-based Transformers architecture such as BERT which demonstrates much more accurate results for most NLP tasks.

Noticeable. Now you have a great product, what’s next? For continuous growth, it is important to maintain and increase online visibility of the product. SEO (search engine optimization) is just one of the elements. Publish success stories. Reach out to your clients and ask them to share what real difference your solution makes in their work. An impressive success story is the best marketing.

On a final note, be passionate about what you are building. It usually results in high quality. And clients always appreciate that.

 

Sergii Shcherbak

CTO @ Maigon.io