Ironclad Journal icon Ironclad BLOG

Ironclad Layers on New No-code Features to Simplify AI Customization

May 9, 2024 4 min read
AI generated stylized image of a document with a pin in it

Despite how rapidly AI technology has proliferated in the past year, it has not yet joined the ranks of indispensable B2B software features like fast sales automation, robust customer relationship management, or digitized payroll. And while that status might still be a distant horizon, there’s one thing that could pull it closer: customization. Because people are much more likely to trust a tool–and use it–when it can configure itself specifically to their unique problems.

Ironclad believes that your AI should work for your business; that you should be in the driver’s seat when it comes to AI and maintain complete control. We also understand that your business needs are unique, so shouldn’t your AI accommodate that? (The answer is yes.)

What is Custom AI?

Custom AI is a set of features that allows users to improve the quality of their AI results and expand the model’s ability to absorb unique data, domain knowledge, and context, creating a cycle of continuous improvement that increases efficiency across the user base over time. The set was built around the customization concept of fine-tuning. Fine-tuning lets users tailor Ironclad AI models to their unique use cases–think custom clauses, industry specific data requirements, etc.– by feeding the models contract data that was not observed during the original training process.

Introducing Impact Scoring and Similarity Threshold Controls

That’s why we’re excited to announce the latest set of AI powered features for Custom AI Clauses to make training AI simple and fast – no technical experience required. Last November, we announced Custom AI Clauses, which gave users unprecedented control over their AI. And now, we’re enhancing this powerful functionality to give Ironclad users even more control over improving AI results across the platform.

Today, Ironclad is releasing two new no-code features improving global impact to Smart Import, Ironclad Playbooks, and AI AssistTM – which uses generative AI to identify problematic language in contracts and automatically suggest company approved clauses – to give legal teams the control and guardrails they need to unlock Ironclad AI’s full potential.

1. See how impactful your clause training is with Impact Scoring. When training custom clauses or one of Ironclad’s 175 default AI clauses, admins will now see ratings on the potential impact of their training samples. By adding industry specific clause examples to Ironclad, Admins will improve the  accuracy of Ironclad’s clause detection system in AI Playbooks and Smart Import that only gets more relevant and reliable over time.

2. Set the right level of training precision with Similarity threshold controls. Admins can now manage how narrowly or broadly Ironclad AI classifies clauses per custom definition. The “Exact Match” setting means AI will only identify exact matches to specific training samples provided. With “Somewhat Similar” selected, Ironclad AI looks for clauses with similarity to the samples. This similarity threshold control allows more precision when building platform-wide playbooks.

When customizing AI clauses with these new features, admins can significantly improve how contract data is captured in Smart Import and how their negotiation playbooks function.

How does Custom AI work?

Currently, Ironclad’s AI suiteIronclad Smart Import, AI Playbooks, and AI Assist™–can detect 175 clauses and 19 fields, collectively called “AI contract data points” out of the box. With the availability of Custom AI, users can train Ironclad AI on clauses that sit both within and outside of that box and are unique to their use cases and business requirements. And the more training examples they add to both, the more accurate and relevant results Custom AI will offer up. Its specific capabilities include:

  • Custom clauses. Users can train Custom AI with unique clauses. They simply create a name, description, and input examples of clause text to start training the model. They can start with three examples and add or delete additional ones whenever they’d like.
  • Consolidated data management. When Custom AI gets trained on new contract language, that training will be reflected across the entire Ironclad AI suite, reducing the time and effort to update each individual feature.
  • Out-of-the-box clause training. If a clause included in the out-of-the-box list doesn’t automatically get detected and labeled due to minor language variations, users can manually highlight and label the selected text as an example of the desired clause, and the updated language will get detected moving forward.
  • Restricted user permissions. Admins can limit who is allowed to make updates and train the models, giving them control of their accuracy and quality.
  • Controlled access to AI contract data points. Though there are 194 contract data points–plus countless custom clauses–to detect, not all of them are relevant to every contract, business, or industry. With Custom AI, users can hide or limit which data points are available depending on their relevance and the priorities of the business.

It’s important to note that we do not share new clause data with any third party AI companies, so Custom AI users can be confident that their data is protected and not being used to train external large language models (LLMs).

A first step

While Custom AI already has robust customization capabilities, it’s just the first set of many. As AI technology develops, as with most digital tech we’ve encountered in the last decade, it’s only going to get more specific, more tailored, more personalized. The added bonus that’s unique to AI training, though, is that the growing specificity makes the results more accurate and reliable.  And we’re excited to continue building on that journey.

Want more content like this? Sign up for our monthly newsletter.

Book your live demo