Bot is clueless and fantasizing a lot

My bot gives junk answers and can’t find a simple reference (“CRP”) which is clearly explained in the knowledgebase. It also doesn’t remember its own company name (“Leanmap”) and is mixing the company name with her own name (“Lea”). This is very disappointing, essentially make the bot useless. Can this be fixed?

Hi Jmuenzing,

This is happening because the quality control was triggered. The agent thought this was an irrelevant question. You should create a summary explaining exactly what CRP is without using the extended name because the agent is not good at understanding acronyms. You can also use the Q&A dataset to fine-tune questions like “What’s CRP?”

i would suggest also to include a better description of the business inside the quality control menu option, prevent unrelated conversation.

Does’t work. I entered terms in QA-database, but the bot is too stupid to find them - it’s annoying! In addition, the bot thinks the agent’s name is the company’s name - how is this possible? It’s so clearly defined but even the most basic information is messed up!

…here another example: when asking "What’s ALSS? The bot is clueless, despite the term was defined in QA-database, precisely as asked. How the heck is this possible?
Screenshot 2024-03-02 at 17.45.37

The bot is appreciating my curiosity, but doesn’t know the answer (example: private browser session) - same result, issue is persistent
Screenshot 2024-03-02 at 17.51.51

Hi jmuenzing

I just checked your dashboard and some Q&A works and some others don’t i think there must be some issue when saving the Q&A with your account i have just opened a ticket with the developer to have your account checked.

Thanks Simone for the follow-up. Answerly is well designed but the functionality bugs are really hurting. Hope this can be solved. J

Hi Jmuenzing,

We have just released a quick fix. Could you please check if the Q&A are now being triggered correctly?

Partially solved, 50/50 ok/not.
Screenshot 2024-03-04 at 17.26.50

I believe in this case, it’s working as expected because you entered the keyword “explain crc,” and the Q&A is used only to match certain keywords. If you want to explain what CRC is, you need to create a normal dataset, summarized like this:

Question: What does CRC stand for?
Answer: CRC means Cost Reduction Course (CRC).

The Q&A dataset is only used to target specific keywords or phrases. You will have to create a dataset with Q&A and a normal summary dataset in order to trigger a response with the keyword: “explain crc.”

Also, please keep in mind that if you start duplicating the term “crc” across your datasets, this will confuse the agent regarding the correct match and response. So, I would recommend keeping this in only one set of the dataset. In fact, I believe you don’t really need to use Q&A.

Bot is clueless, not giving the right answer, even when asking the most basic/pure question.

Hi Simone, thanks much for the detailed testing and feedback.

Interestingly, now it seems to work (did only two quick checks).

The duplicated content was auto-generated by Answerly (scraping website), carry over from previous to newer version.

I added new information (links) where missing, but did not manually go through legacy knowledgebase to look for duplicated content.

I’d add this issue to developer’s list to prevent false responses from duplicated content through integrated Poka Yoke (error proofing).

Thanks/best, Joerg

1 Like

Hi, I just checked your dashboard again and tested it. The agent was performing just fine when I asked those questions. I think the problem, in your case, is the way you have built your knowledge hub, there is too much duplicated content. The datasets should be constructed around what real user questions might be. Here is a video I just made for you.