The AI vision behind Ondorse: Redefining KYB for the future

From hype to reality: Deploying AI for compliance at scale

LLMs hold vast power for compliance professionals - but without proper software sitting on top, they are practically useless. Here is what we have built and what we are currently building at Ondorse in order to harness the new power of AI.

Compliance departments across the globe - who are used to deal with tons of paperwork, boring repetitive tasks, vast amounts of badly structured data, internal and external audits, policies and regulations spread across 10 different jurisdictions along with endless reporting - are watching the current AI revolution and asking themselves:

How can I turn this to my advantage ?

By now, most of them have experienced the power of these tools by themselves. They sent a few cases to ChatGPT (the most adventurous have sent real client data) and they were amazed by the results.

So they have turned to their CTO (or the CDO or their strategic IT consultants of choice, Accenture we are looking at you here) and told them they wanted to deploy this at scale, with a model that can understand their business and functions, and which has access to their data.

To which the CTO probably replied with some concerns over:

• the security of such a project: access to proprietary data, user rights, detailed logs, troubleshooting issues and transparency are not what come to mind first when dealing with LLMs
• the ability for such a project to handle errors correctly
• the lack of exact results due to the non-deterministic nature of LLMs

At Ondorse, our goal is to build AI tools for compliance professionals that can securely be deployed at scale.

Here is how.

The pre-requisite

Throwing an LLM at a KYC onboarding flow is like using a flamethrower to light a candle. LLMs need an opinionated perspective on how to solve the problem.

Imagine you're in a new city, and you want to ask a local for directions. Before you ask, you need to understand a few things about the city: what the main landmarks are, how the streets are laid out, and maybe even some local customs. This understanding helps you ask the right questions and interpret the answers correctly.

Similarly, when an organization wants to use LLMs, they need to first establish a clear framework that explains how their business works, what their goals are, and how they think about their domain. This framework is like a map of the city that helps the AI understand the context in which it will operate.

By providing these facts as inputs, the LLM can explicitly understand the relationships within the domain rather than discerning them statistically.

At Ondorse, we have spent the last 3 years building this framework for compliance operations: we call it Ondorse Workflows.

Ondorse Workflows are structured processes and objects designed to automate and streamline compliance and verification tasks within an organization. These workflows are essential for ensuring that operators follow internal requirements efficiently and effectively.

Ondorse Workflows serve as the essential framework—or ontology—required for an LLM to effectively operate in compliance-related tasks.

Just as a tour guide helps you navigate a new city by providing context and local knowledge, Ondorse Workflows provide the LLM with the contextual understanding of how compliance processes operate within your organization. This includes knowing the specific tasks, their sequence, and the conditions under which they are activated.

The Holy Trinity

Now that we have this framework in place, we have identified a "holy trinity" of principles that will guide our development of AI features at Ondorse in the near future.

These principles ensure that AI is leveraged optimally, enhancing compliance processes without overwhelming users with unnecessary complexity.

Here is the holy trinity.

1. Nourish the beast just right
2. Advise broadly, act precisely
3. Customize and conquer

Nourish the beast just right

To function effectively, an LLM requires a clear and concise understanding of its operational context.

Ondorse ensures that the LLM receives just the right amount of information—not too little, and crucially, not too much.

This balance is essential for maintaining the LLM's focus and preventing it from being overwhelmed by irrelevant data.

For example, if you want an LLM to suggest discards on potential sanction list screening hits, it needs the context of the specific case you are working on—not the entire client base. This focus ensures that the LLM's recommendations are tailored and accurate, avoiding the risk of suggesting to discard a hit based on irrelevant or broader data.

Ondorse is also committed to maintaining the highest standards of data security and privacy. The LLM is never provided with sensitive information such as credentials, passwords, or any data that cannot be shared. We ensure that the AI operates within safe boundaries, respecting the confidentiality and security of all information.

With the right context, the LLM can offer suggestions and insights across an entire case, acting like a knowledgeable advisor that understands the broader landscape.

Advise broadly, act precisely

We use a nuanced approach to AI integration :

• We allow it to suggest improvements across the entire process
• But we allow it to act ⚠️ only ⚠️ at the task level.

This strategy ensures that AI performs specific, well-defined tasks—such as identity verification or fraud detection—where its precision and reliability are most valuable.

Meanwhile, the AI can act like a knowledgeable coworker, offering insights and recommendations on optimizing the broader compliance process.

This dual role allows the AI to suggest improvements, identify potential risks, and propose efficiencies without taking control of the entire workflow. A compliance workflow must be deterministic (ask the regulator about AI bias).

In compliance, human oversight and decision-making remain central. And, as illustrated below, accuracy is non negotiable.

It’s equally important to not get lost in the tech hype either: an AI copilot still requires, well… a pilot. It is at the service of – not a replacement for – these teams.

By limiting AI actions to delimited tasks built into a human-made workflow, Ondorse benefits from the AI's comprehensive understanding and analytical capabilities within a controlled framework.

Customize and conquer

We are enablers : we are making advanced AI capabilities accessible to the compliance officers so they can leverage them effectively in their day-to-day operations.

Our platform already allows you to experiment with various pre-integrated models. Therefore, we ensure sensitive information remains on-premises and compliant with regulatory standards.

This approach enhances data protection and builds trust in today's geopolitical climate.

The prompts are easily adjustable. This level of customization means that you can fine-tune the AI's behavior without needing extensive development resources.

Ondorse also guarantee reliability and support, ensuring that compliance processes run smoothly and efficiently. Our commitment to supporting our clients’ operations means that they can focus on what matters most—maintaining compliance and driving the business forward.

Like what we have done with Workflows, which are highly customizable, it is about giving you the right toolbox.

Where does it takes us

Ondorse is paving the way to a new world of compliance and user experience.

By integrating AI seamlessly into compliance workflows, we eliminate waiting times, enhancing the overall user experience.

Compliance officers are freed from mundane tasks, allowing them to focus on highly creative and strategic work. Our platform ensures that human operators maintain full visibility into the execution rules and history of automations.

With Ondorse, LLMs operate discreetly, hidden in plain sight, driving processes without overshadowing human oversight. We go beyond chat.

This balance allows teams to take a significant step toward the autonomous enterprise, where AI and agents optimize operations while keeping human fulfillment and security at the forefront.

We are building this.