Getting an instant, accurate answer to questions is not only a relief for every customer; it’s also an investment in the future of your business, as HR service provider Fhris realized. Pipple translated their wishlist for a GenAI assistant (chatbot) into a secure tool that lightens the workload for both clients and employees.

Fhris is a service provider that automates HR processes and payroll in AFAS. Since their launch in 2019, their customer base has been growing exponentially each year. Soon, the number of daily users on their platform will reach 15,000 professionals — impressive figures that would make many startups envious. But founder and director Tim Claassen remains realistic: “Our profit margin depends on efficiency. So, an increase in users is only interesting if we can prevent our workload from rising proportionally.”

Huge Knowledge Base

The workload at Fhris is partly driven by questions users ask the Fhris back office. Currently, there are 3,000 tickets per month. Handling these tickets well requires the necessary skills from a back-office employee. The process starts with correctly interpreting the question. Then, the employee has to find the answer in an internal wiki full of documents, including previously given answers, entire collective labor agreements, company-specific regulations, and thick tax authority handbooks. The employee has to gauge where the answer is located and select relevant information from it, and then formulate it clearly to assist the questioner.

Automating with AI

A process prone to errors and time-consuming, especially with complex questions or when the employee is less experienced. That’s why Fhris decided to automate this process, Claassen explains: “Our ultimate goal is for users not to have to submit tickets at all. Through a search function on our platform, they should be able to get an accurate and clear answer immediately. And if the chatbot doesn’t know the answer? It should at least formulate the customer’s question fully so the back office can provide the right answer in one go.”

A Necessary Investment

Since Fhris didn’t have the in-house expertise to set up an AI-driven database, Claassen had to bring in external experts. A big investment for a startup, but according to Claassen, it was essential: “I am convinced this will improve our customer satisfaction and productivity. Moreover, it gives us a crucial advantage over competitors. Accountants generally prefer dealing with the past rather than the future, but no company can ignore the rapid rise of AI. Otherwise, you’ll hopelessly fall behind.”

Wishlist

Claassen approached three possible partners, one of which was Pipple. Each was given the same set of requirements:

• Control over the answers provided;
• The ability to easily manage source material without needing a programmer;
• An off-the-shelf application to continually leverage the latest developments from Microsoft and Open AI;
• Everything within their own Microsoft environment due to the privacy-sensitive data.

Clear Explanation

Pipple won the project because they were the only ones who could explain their solution in simple terms. Claassen: “If you’re making a big investment, you want to understand exactly what you’re paying for.”

Phased Rollout

Pipple helped map out a phased rollout that allowed Fhris to control their development speed and spread out investment costs. First, the back-office employees would work with the AI assistant, training the application as they went. Only when the assistant could answer almost every question perfectly would Fhris release it to end users.

The database was also filled step by step. The first chunk of information that Pipple fed into the new AI assistant was the extensive “Payroll Tax Handbook” from the Tax Authority, the cornerstone for payroll processors. Now, the collective labor agreements and company-specific regulations also hold no secrets for the AI assistant.

Beyond Expectations

Although Fhris has only been working with AI for a few months, Claassen already sees the success: “Without any training, employees can find the right information at the press of a button. Seniors familiar with the source documents check the assistant’s answers and correct them if necessary. This only needs to happen once; the assistant will provide the correct answer from then on. We are already on track to halve the number of tickets.”

“We’re much further along than I, and perhaps even Pipple, had expected. This is largely thanks to the enthusiasm with which Pipple approached our issue. They were, of course, already skilled in AI solutions, but not yet in our field. They really immersed themselves in our business. This understanding shows in the result: the solution they created for us is exactly what we needed.”