Lindy Community Icon

Optimizing Automated Email Responses for Multiple Brands

·
·

Afternoon! I'm working with the out of the box automated email agent. our use case is somewhat unique in that we aggregate brand information from multiple brands, so for example, we have to keep warranty information in our knowledge base related to multiple brands. if an email is sent with a question about one brand, Lindy generally returns the accurate response and documents. however, if a question is asked like "hey what are the lead times for brand 1? and can you send me the warranty information for brand 2?" it often sends back the warranty information for the wrong brand. I'm currently only using one knowledge base. is there another way I should be doing this? tia!

  • Avatar of Alon J.
    Alon J.
    ·
    ·

    Hey Travis! How many brands do you aggregate for? If the knowledge base is getting confused, another option you could do is use a condition with 3 exits where:

    • If X is asking about Brand 1, go to knowledge base 1.

    • If X is asking about Brand 2, go to knowledge base 2

    • If X is asking about Brand 2, go to knowledge base 3

    Etc etc.... And each knowledge base will have its respective brand information. Let me know if that helps!

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    Thanks Alon J.! I tried testing that earlier with just two brands, however when I tested an incoming email asking for information from Brand 1 and Brand 2, it was able to return the information from Brand 1, but was unable to return the information for Brand 2. It's like the Lindy is grabbing the answer for Question 1 and continuing down the workflow without checking for answers in parallel Brand Knowledge bases from Brands 2, 3, 4, etc. (if that makes sense). I'm new to this, so it's completely possible that I'm just missing some logic that should be there! But here's a screenshot of my flow (as I mentioned, it's basically the out of the box workflow).

  • Avatar of Alon J.
    Alon J.
    ·
    ·

    Do you mind sending over the task link and I can take a better look? How to share your Lindy task URL: 1. Click your Lindy agent → click “tasks” 2. Find the task you need help with 3. Click the specific task (you should see the full execution) 4. Copy that page URL from your browser 5. Paste the URL in your support request

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    here you go, Alon J.!: https://chat.lindy.ai/travis-kerns-workspace/lindy/tk-test-email-responder-684c708e[…]stRunId=6850796ba6fd3744f766a9a9&task=685074c54d14d7d989359c85 "source" is one of our brands while "9to5" is another. you'll notice in the execution that Lindy does a great job returning the GSA (contract) details for Source in part 1, however, in part 2 it says it's returning the 9to5 warranty details, but it actually links to the Source warranty details.

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    thank you so much for your help and attention by the way. I really appreciate it!

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    Alon J. additionally, now when I'm testing, it's not even pulling back the info for the "second" question in an email. it's just saying "i'll get back to you shortly with the answer" or creating a draft with "insert information here" where it should return the info automatically. example here: https://chat.lindy.ai/home/?templateId=6851c35f4fb7b79aafda5b50

  • Avatar of Alon J.
    Alon J.
    ·
    ·

    Hey Travis! Apologies about the delay. For the last think, do you mind sending the Task URL (not the template URL). If you click on the task, you can just copy the URL for that specific task. The task URL for that example would be great so I can take a look here. Also, How big is your knowledge base and how is it structured? A couple things to do here: So this seems like the information is being answered incorrectly based off a couple things

    1. 1.

      The first thing you can try is upping your model. I see you're using GPT 4o-mini -> If you have a complex database with different brands and a lot of information within those brands, you can probably up the model here. Gemini 2.5 Pro could be a good option here. (it’s robust, cost-effective, and excels at nuanced searches across large context windows)

    2. 2.

      Another thing would be to update the prompts to make it suuuper clear as to what the information needs to be added.

    Pro tip: If you write a prompt manually, you can put it through Chat GPT and ask it to improve it, making it clearer and more concrete (this has saved me a few times) For the prompt, you can also advise it to answer each question and double check that all the answered were replied to and responded to with each brand respectively without confusing the brands. Let me know if that helps!

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    Thank you Alon J. this is super helpful! At which "node" do I update the prompt? There are quite a few opportunities to do so.

  • Avatar of Alon J.
    Alon J.
    ·
    ·

    No problem! In the knowledge Base -- That's where it pulls all the information. You'll want to make it as clear as possible there on how to formulate the response from the information in the Knowledge Base. In the last email 'Respond' action, you can use mostly what's already pre-written there. You may just want to customize a little bit 🙂

  • Avatar of Travis K.
    Travis K.
    ·
    ·

    thank you so much! I'll do some more testing with your tips.

  • Avatar of Alon J.
    Alon J.
    ·
    ·

    No problem - Let us know if you hit any roadblocks