Agentic Web Scraping

Efficiently scraping information from the web using a multi-step Agentic workflow.

The problem & situation

Our client is a knowledge center on sustainability for the Dutch market. They offer a practical guide on various things that you can do today, to make a positive impact on the environment. Many of these initiatives are locally subsidized by municipalities. Not many people know this and it can be hard to find the right information on government websites. Therefore, our client offers an easy-to-use tool to the public where anyone can find if they could be sponsored by their municipality for a sustainable project. Staying on top of all sustainability related subsidies for every municipality in The Netherlands is a substantial effort and costly exercise. To make sure our client offers up-to-date information to their users, all known subsidies are regularly checked by hand.

The solution

Leveraging the latest tools in AI, we developed a multi-step Agentic workflow, to greatly reduce time and money spent on this issue. Our first Agent crawls the web and finds the location of the information we need, without human intervention. Saving us a lot of time as opposed to conventional scraping methods where we would've had to build the required logic for every website individually. Our second agent is optimized to gather the information and place it into a desired format. The last Agent in this process analyses the content we gathered and flags changes that should get human attention. There's many small steps in between to make this work, but the process is essentially simple which makes it efficient and reliable.

The outcome

Using this new way of working, this entire process can be managed by just one person in a matter of minutes. Greatly reducing the cost for this particular tool and enabling our client to focus their time on more rewarding parts of their job. Additionally, accuracy and reach were both improved.

Ready to take off?

Schedule a free data scan or contact us for your next challenge!