One of Spain’s largest real estate portals has two help chatbots. One serves anyone who visits the portal, mostly private individuals listing their home. The other serves real estate agencies and professionals.
In early 2023, the portal’s team asked us two questions. Should the two chatbots have the same personality? And what should change to improve the experience?
The challenge
A help chatbot usually starts as a tree of questions and answers, and grows by adding content. Over time, reasonable sentences it doesn’t understand start to appear, along with answers that are too long and shifts in tone from one answer to the next.
It had to be measured methodically, and the team needed clear criteria to improve the answers while respecting the style guide they already had.
What we did
We evaluated both chatbots with established conversational quality and usability methods, and documented each problem with real examples.
In a workshop with the portal’s team, we defined each chatbot’s personality and tone. They share a common base but differ in warmth: the one talking to professionals who come back often can be more familiar than the one handling one-off questions.
We wrapped up with a recommendations report, a guide to how each chatbot should behave in the most common situations, and indicators to measure their quality.
Results
The portal’s team received a diagnosis of each chatbot, a defined personality for both, and a guide to improve their answers. We also recommended repeating the evaluation with real users.
