One platform for 50 years of educational expertise: How KNOLL is relieving its service team with an AI agent and hybrid search

Key facts at a glance

For over 50 years, KNOLL has been preparing aspiring tax experts for the tax consultant exam and further qualifications. In June 2025, the Munich-based continuing education institute went live with a newly built website and branchly: an AI agent intercepts routine questions, hybrid search delivers direct answers instead of pure link lists, and voice telephony also feeds into the same platform. Individual career advice deliberately remains with the service team. Eva Wegele, project manager in KNOLL's digital team, accompanied the rollout; her quotes are taken from the Branch Out podcast. The results after twelve months: 13,169 AI sessions processed, 85.9% of questions fully answered, and a knowledge gap of only 1.8%.

Why conventional answers were no longer enough at KNOLL

KNOLL supports participants through a demanding educational program for up to two years, from tax clerk to tax specialist to the ultimate discipline, the tax consultant exam. With exam preparation courses, in-person classes, comprehensive study booklets, and its own learning platform, one thing is clear: the product is knowledge, and that knowledge is highly complex.

This very abundance became an obstacle on the website. Different programs with sometimes very similar course names, numerous guide articles, constantly updated dates, and complex funding rules made it hard to find one's way. Prospects with specific questions first had to navigate the website on their own. At the same time, the service team was answering the same standard questions on the phone or by email day after day, leaving little time for personal career advice.

For KNOLL, therefore, two requirements were clear: personal advice remains the indispensable core of the institute. But recurring routine questions must be intercepted beforehand, right where they arise: on the website. The 2025 website relaunch offered the ideal opportunity: what was needed was not just a more modern web presence, but one that was noticeably more intelligent.

From idea to partnership: Munich, HelpSpace, and a personal kick-off

Eva Wegele is a project manager in KNOLL's digital team and is responsible for the institute's digital processes at the interface between technology and specialist departments. She describes the decision to use branchly with three core arguments: the existing knowledge database from HelpSpace could be integrated seamlessly, both teams are based in Munich, and the collaboration started with a personal meet-and-greet instead of anonymous support.

"There are now providers everywhere. But it was simply important to us not to buy just any solution from overseas, but to have a local contact person."

Data protection, EU hosting, and accessibility were firm requirements, but turned out to be uncomplicated to implement. The real challenge lay in cultural change: at KNOLL, several generations work closely together, and a digital assistant communicates differently than colleagues who, for many years, have been advising prospects with expertise and a feel for the subject. The project was therefore designed as a participatory change from the outset, not as a purely technical IT project.

Iterative rather than perfectly planned: implementation in just a few weeks

The technical rollout ran smoothly: around four weeks passed from setup to go-live on June 30, 2025, just in time for the launch of the new website. What mattered even more was the internal process: all specialist departments were involved in test phases at an early stage, could ask critical questions, and feed their feedback directly into the fine-tuning.

"We tested internally very early on and brought all colleagues on board. That was how we tried to take everyone along on the journey. The AI agent is now really highly valued and accepted by everyone."

How effective this dialogue is became apparent shortly after launch. Because many offerings have similar names, the AI agent occasionally confused course and program at first. KNOLL specifically trained the system to ask precise follow-up questions when something was unclear: Do you mean the program or the course? Today, the agent acts reliably, catches imprecise wording from searchers, and guides them purposefully to the right educational offering, without any blind guessing.

In addition to the AI agent, hybrid search went live. It combines semantic results with classic keyword search and delivers concrete answers instead of mere result lists. With dynamically displayed entry questions on every subpage, it functions more like an intelligent navigation aid. The knowledge base, which is maintained weekly, automatically incorporates curriculum changes, updates such as the reform of the tax consultant exam, and seasonal demand peaks before exam dates, without any double maintenance effort for the team.

What the numbers show after twelve months

branchly Analytics data from the past twelve months confirms the strong response: overall, the platform processed 13,169 AI sessions, 11,116 of them in chat and 3,376 via hybrid search. In addition, there were 317 sessions from voice telephony. By topic, questions about learning materials dominated (1,082 inquiries), followed by admission requirements (422) and details on exam simulation (355). Funding options (213), discounts (190), and examination regulations (152) were also frequently addressed. The distribution also shows a clear user preference: the dialogue-based chat is used significantly more often than the pure search field.

The resolution rate of 85.9% becomes even more meaningful in combination with the other figures: in only 1.8% of inquiries was the stored knowledge insufficient, while the agent asked targeted follow-up questions in 11.7% of cases to rule out misunderstandings. Calls via the integrated voice solution also converge on the same data foundation, allowing data to be evaluated across channels without KNOLL having to maintain parallel systems.

This puts the defined division of roles into effect: the system automatically handles routine inquiries about processes or payment terms, while individual educational decisions are deliberately handed off to the service team.

"When I look into the dashboard and see the questions, I think: good to know. If an email or a call had come in every time, then the agent has already provided relief."

The transparency gained provides additional value: the marketing team uses the analyses to refine website content in a targeted way and align information events with actual information needs.

Lessons: What other education providers can take away from this

The practical experience at KNOLL underscores that using generative AI is not a rigid one-time rollout but a continuous learning process. Three insights stand out:

First: a quick real-world test beats perfectionism. A focused start aimed at relieving routine inquiries provides more valuable insights than months of theoretical planning. Second: internal acceptance is decisive. Actively involving employees with years of consulting experience at an early stage reduces reservations and noticeably improves answer quality. Third: define clear interfaces. A successful AI deployment clearly separates automated information from personal expert advice and reliably passes inquirers on to a human at the decisive moment.

Building on this, KNOLL is already exploring next steps, including an AI-powered learning companion for participants. The ongoing dialogue with branchly shows how concrete user questions continuously lead to better offerings and deeper insights.