Initial situation

Supercharging compliance workflows with generative AI

The various industry regulations, data protection requirements and other regulations create a complex web of requirements that companies must meet. And this applies not only to the companies themselves, but also to related companies of which they are suppliers or from which they in turn purchase goods or raw materials.

Our client offers a SaaS-based software solution that helps companies in regulated industries manage their compliance with these requirements. This involves storing both internal documents and compliance checklists in the software and linking the two. This process is usually done manually by employees and validated by external auditors.

Objective

A large part of these compliance processes relies on repetitive actions that have to be carried out by valuable experts:

Compliance questionnaires are answered over and over again in ever new variations, documents are manually sifted through for supporting evidence, and reports are assembled.

Advances in the field of Large Language Models offered the opportunity to achieve a high degree of automation without taking the experts out of the loop entirely.

Our customer wanted to integrate AI components into their existing workflows in such a way that a seamless, human-controlled, but extensively technology-supported process emerged.

Capture and vectorization of all relevant documents

The prerequisite for automated compliance workflows is the collection and preparation of the relevant compliance documents. These documents, such as codes of conduct, annual reports or certificates, are imported into a data room in our client's software and pre-processed to make them available for further processing by generative AI.

An essential part of this step is the conversion of PDF documents into a form that can be processed by the AI, which often involves customized OCR (optical character recognition) to extract the text from the documents and put it into a structured form.

NEOMO developed a component that breaks down the captured text content into meaningful facts or passages and vectorizes them on several levels using semantic embeddings.

The resulting vector database forms the basis for all further processing steps.

I-based answering of compliance questions

NEOMO implemented an Azure-based service landscape to automatically answer compliance questions. It was important that subject matter experts could intervene in the process at any time to control and correct it.

The close integration of our services and our client's user interface resulted in a workflow in which

the software automatically extracted and displayed relevant passages from the existing documents,
selected or all of these passages to a Large Language Model to generate an answer
or to automatically apply existing answers from other questionnaires to the new question.

At each step of the process, it is possible to check and adjust the results before further processing or to send them unchanged.

Fact catalog, document interviewing and validation

After the basic processes had been successfully run and tested with our customer's users, we had the opportunity to simplify or even enable other unpleasantly complex processes based on NEOMO technology.

For this, only minor extensions of the existing services were necessary. Among other things, the following were developed:

a semantically searchable catalog of facts
a component for “questioning” one or more documents, but free of hallucinations and with jumpable references to the source document
a component for qualitative evaluation of key documents.

The end of this development is not in sight yet: in every conversation with our client, we find new use cases and new possibilities to make the daily work of their users easier and more effective.

Result

Through intensive collaboration with our client, i.e. by combining deep domain expertise and technological know-how, a product was created that puts the AI completely at the service and under the control of the user. Without suffering from the “side effects” of LLMs, our client's users thus achieve:

Time savings and efficiency

The AI solution made it possible to answer compliance questionnaires in minutes, which previously took hours or even days. Employees who were previously occupied with the tedious task of answering questionnaires can now focus on more value-added activities.

Consistency and accuracy

By using a centralized database and semantic embeddings, answers were generated consistently and correctly. The risk of errors and inconsistencies caused by manual processing was drastically reduced.

Human-in-the-Loop

Despite the automation, a human remained involved in the control process, which ensured the quality and accountability of the responses. This allowed for continuous improvement of the system through human feedback.

Scalability

The solution is scalable and can easily be applied to other areas such as tender management or answering RFIs (Request for Information) and RFQs (Request for Quotation).

Minimized risk

By strictly adhering to the rule that AI is only based on human-generated information and does not generate facts independently, the risk of misinterpretations and hallucinations was minimized. This ensures that the AI solution remains reliable and trustworthy.

Core component

What started out as an add-on or an idea has become the core component of our customer's offering: AI-based treatment of compliance issues.

Jørgen Kadal, CPO & Founder, ComplAi

"As a compliance-focused company we are obviously under intense scrutiny in regards to security, data protection and reliability. So we could not just put arbitrary AI components into our product which might expose customer data to third parties or start hallucinating at any time. NEOMO helped us navigate the somewhat treacherous territory of current AI solutions with our specific use case and our specific parameters in mind. Their deep knowledge about the properties and implications of all types of large language models - both technological and legal - allowed us to build a solution that brings AI to our customer at maximum efficiency and reliability."

Start a conversation with us.

We are always interested in a good conversation, whether with a specific reason or simply to exchange ideas.

You can reach us via the contact form or you can book a meeting directly.