Turning data overload into smart insights with AI

At Enhesa, ongoing improvements to how we screen and prioritize regulatory content have made the process faster and more focused. These enhancements directly support our customers by ensuring the right intelligence reaches them sooner and with greater precision. In this article, Enhesa data specialist Candy Quarry examines the business need for concise, up-to-date compliance requirement data — and how Enhesa has developed integrated AI solutions to overcome these challenges for our customers.

Sifting through the regulatory deluge

To stay compliant, businesses need timely, relevant regulatory insights they can trust without having to sift through overwhelming volumes of information. Yet every day, thousands of regulatory notices, publications, and announcements are released worldwide, making it increasingly difficult to identify what truly matters. For many teams, this means hours spent searching for a handful of critical updates and trying to identify meaningful regulatory signals within a rapidly expanding data set.

At Enhesa, we addressed this challenge by transforming data overload into smarter regulatory intelligence, combining deep expertise with AI-driven screening to deliver faster, more focused outcomes.

Our Data Analytics team plays a crucial role in supporting regulatory intelligence across the business. Each day, we run screening procedures that analyze thousands of rows of data to identify relevant regulatory updates and insights. These outputs underpin solutions such as our Focused Compliance modules, giving businesses practical tools to monitor, manage, and apply regulatory change with confidence. By connecting timely intelligence with expert interpretation and compliance solutions, we help customers stay ahead and strengthen their programmes.

These results feed into two key areas:

• Our Editorial team turn complex regulatory updates into clear, insightful articles that help customers’ understand what the changes mean for their business.
• Our Expert Services team provide actionable intelligence to support compliance within our customers’ organisations.

Together, this work forms part of Enhesa’s broader commitment to advancing compliance through technology, using data-driven processes and advanced tools to strengthen global sustainability solutions and power the next generation of regulatory discovery.

The challenge: Volume and relevance

With hundreds of new sources added each year, the challenge isn’t just volume; it’s relevance. Each day, the team sifts through thousands of pieces of regulatory source content, including:

• Government notices
• Draft laws
• Consultation pages
• Agency press releases
• Guidance PDFs
• News articles

Yet only a small fraction of them are relevant for our scope. Out of around 2,000 incoming updates, maybe only 75 are relevant. It can be like searching for a needle in a haystack.

Imagine you’re searching for the latest chemical legislation in Asia. Now picture having to scroll through thousands of global regulatory updates first, most of which have nothing to do with this region. That was our reality, and the question became clear:

How do we cut the time spent on this manual process without sacrificing accuracy as the dataset keeps growing?

The answer was to turn data overload into smart, automated data screening using AI.

Before automation: The manual method

Before outlining the screening process, it’s important to explain how the data is sourced.

From the very beginning, this workflow was built on innovation. Before AI entered the picture, one of the toughest challenges wasn’t analyzing regulatory updates; it was collecting them. To address this, we implemented a specialized web-scraping solution that captures information from hundreds of regulatory sources worldwide. While the software provides the framework, the real work lies in configuring, maintaining, and continuously enhancing it to ensure accuracy, coverage, and resilience as new sources and formats emerge. This ongoing effort by our Data Analytics team means we don’t miss critical updates and can adapt quickly as the regulatory landscape evolves.

Our earlier screening process relied on this incoming data but was entirely manual. Simple keyword matching helped narrow down potential updates; however, the team still needed to read all or parts of each document to determine whether it truly fell within scope.

While effective, this approach required extensive visual checks and typically took 4-5 hours each day, depending on the volume of incoming content.

The AI solution, with expert oversight

As the volume of regulatory source content continued to grow, it became clear that a fully manual process wasn’t sustainable. We needed a way to reduce the daily screening workload without losing the human judgement that keeps our analysis accurate. That’s where AI came in. Our goal wasn’t to replace expertise, but to build a tool that could support it by highlighting the updates most likely to matter while leaving the final decision to our specialists.

Harnessing predictive models to maximize impact

We built our own scoring tool, designed specifically for our workflow and scope criteria.

The tool now scores every incoming item, helping us assess scope much faster. As a result, daily screening time has dropped from 4-5 hours to 1.5-2 hours, saving roughly 670 hours each year — the equivalent of more than 80 working days.

This extra capacity allows our team to focus on other important projects and continuous improvement efforts, while also delivering insights to internal teams more quickly. This efficiency means insights reach our Editorial and Expert Services teams earlier, enabling us to publish updates faster and provide timely intelligence to customers.

Defining a learning AI model

In this case, when we talk about AI, we mean a type of machine learning model that learns patterns from text. Instead of being programmed with fixed rules, the model improves by analyzing examples and spotting the kinds of wording that usually signal relevance.

These labelled examples act as the model’s ‘teaching set’, showing it what we consider in scope and what we don’t.

The objective: Managing increase volume without sacrificing human expertise

The goal was to create a tool that could manage the growing volume while keeping expert oversight central to the process.

We wanted to design a tool that supports the team by assisting with the repetitive workload and highlighting the updates most likely to matter, but where the final decision would remain with our team of experts.

To tackle the growing workload, the AI and Machine Learning Team built an AI-powered scoring tool to streamline the process. Instead of manually assessing thousands of rows, the model evaluates each one and predicts whether it’s likely to be relevant to our scope. This narrows down the likelihood of something being in or out of scope and organizes the data we need to consider in a manageable way.

Specialized training for a specialized tool

At the heart of the tool is a scoring model, trained on examples of past news items that had previously been labelled as either “in scope” or “not in scope” by our regulatory expert journalists. All these labelled items formed the training data and showed the model what we considered to be valuable.

How it works

The model works by looking at the words available, either from inside the publication or based on the title and summary. Every word is given a weight. Some words push the score up because they’re usually linked to regulatory changes we care about. Others push it down because they tend to appear in unrelated content.

The resulting score works a bit like a probability: the higher the score, the more likely the document is something we need to report on.

During training, the model learns by trial and error. It looks at the words in each labelled text, checks whether its prediction matches the human-assigned label, adjusts its weights, and repeats. Over time, it finds a balance where it performs well on the labelled data and can handle new, unseen documents — without simply memorizing examples.

Testing the model

Once trained, we tested the model against real, human-marked data to assess its accuracy. It achieved approximately 99% accuracy and missed very few positive hits, meaning nearly all relevant updates were caught. Most of the remaining flagged items were false positives — a few extra items the team could quickly review, ensuring that no critical updates were overlooked.

To reinforce this, we also built a safety layer of extra human checks for high-priority sources, keeping the team vigilant and making sure important updates are never missed.

What's next?

This AI-powered screening tool is just one of many innovations we’ve developed at Enhesa. Looking ahead, we aim to scale this approach to additional processes and content areas, driving a smarter, more connected way of working across the business…

Greater efficiency

A key priority is bringing greater harmony and efficiency to screening workflows across our expert business units. By aligning processes and sharing insights seamlessly, we can maximize consistency and unlock even more value for our teams and customers.

Clearer, more tailored insights — faster

We’re also exploring next-generation search and discovery capabilities that go beyond traditional keyword lookups. Imagine surfacing the most relevant regulatory intelligence instantly, no matter how complex the query or how vast the dataset. Combined with AI-driven screening, this will transform how quickly we deliver insights internally — and ultimately to our customers.

A clear goal

Our focus is set firmly on driving transformation that saves time, improves accuracy, and keeps compliance intelligence ahead of the curve.

By continuing to refine workflows and integrate smarter solutions, we’re building a foundation for faster, more intuitive access to the information that matters most to our customers across myriad industries.