This doc should hopefully provide you (all you legendary support reps) with a thorough overview of how the assistant conversation tool works so that you can provide clients with a good understanding of what it does. This will allow them to make informed decisions about if the tool is a good choice for their business. There are 2 main parts. The client walkthough and the RG troubleshooting. The first part is a high level overview and the second breaks down each process to assist you with troubleshooting.
Basic Client Walkthrough:
This section serves as a quick intro for any clients who are wondering what the assistant does and how it works. At a very high level, this tool serves to automate initial conversation with a new lead before passing it off to the assigned agent. That's it, that's the doc, let's go home folks. :)
Setup/ Config / Use:
It can be enabled for any company in the admin panel of their LM.
When enabled the assistant will start a conversation with any lead who receives a texting auto-responder AND replies to it. Messages sent by the assistant will have `(Assistant)` underneath, and while active the assistant will have a bar that notified the user that the `Assistant in Progress`.
Conversation "Goals":
It might help to think about this tool as a "flowchart". It sends replies based on the previous replies and responses received. These potential replies are "scored" based on the lead info it is being sent and it will then send the response the has the highest score. This is a way to describe what is happening, but we do not have the specific internals from structurely.
The end "goal" for the assistant is to schedule an appointment. (The name of one of my test agents is "person 1".)
If necessary it can also "bail" if the lead on the other end is proving too difficult to talk with: (ie if you try to break it)
It's relatively easy to break intentionally, but for a real lead, it shouldn't have many problems.
Conversations can be stopped by sending a manual text, drip text or hitting the "Stop Assistant" button.
Required criteria for conversation start:
- Tool must be enabled
- Lead needs to be assigned to an agent
- Agent must have auto-responders turned on for lead source
- Lead needs to reply to auto-responder
- Lead must NOT receive any manual text or drip text communication. (This will disable the assistant) This means turning off drips that have texts
- The lead's phone number has to not have ANY existing conversations in the company lead manager. (more on this in troubleshooting)
Caveats:
- There is no way to enable it for only 1 agent or specific sources.
- Real Geeks does not have any control over the responses being sent or content.
- We can't add or change messages or content.
- We don't have the scripts or scoring criteria for messages
_________________________________________________________________________________________________________
Real Geeks Troubleshooting:
This section of the doc is going to break down the tool action by action. It might be helpful to view the handy visual representation (thank you Aaron) while looking at this.
**Troubleshooting note: Before getting into the details it's worth noting that one of the caveats to start a conversation is that your lead's phone number does not have a text conversation associated with it. This means that you will only be able to test the Assistant Conversation 1 time per CRM. This is likely to be a common problem if we have a site owner or user who tries to test multiple times. You should also keep this in mind when trying to troubleshoot it yourself.
Overview:
At a high level, everything that happens with the assistant conversation uses leadrouter. Structurely connects to lead-router via Fargo. Fargo is our tool that allows external sources to send texts through lead-router. It does this by listening to activities and sending outbound texts to rookery, which then sends them to the lead via Twilio. Because of this, we can check the lead-router log groups to see activities during troubleshooting. (Find example activities and include locations here)
Activity/Log Walkthrough:
Conversation Start:
New Lead signs up and receives an auto-responder. Once the lead replies to that auto-responder it prompts LM to send an activity to Structurely.
Once Structurely has received tha activity above, they confirm receipt and start the conversation with the activity below:
For example: A lead (that meets all req criteria) fails to have their assistant conversation started. If you are not able to find the first activity, then the bug is on RG's end. (We failed to tell Structurely to start) If however, the first activity is there but we have no assistant_conversation_started activity, then the bug is on Structurely's end.
Conversation Stop:
There are two ways that a conversation can be stopped. Either RG stops the conversation or Structurely stops the conversation.
This activity is what RG will send to stop the conversation. An example of this would be if the agent manually texts a lead before the conversation has ended.
This is the activity that Stucturely sends to us when the bot has finished it's conversation.
Reason Codes:
^^ This need to be checked to confirm we still use them.
At the end of a successful conversation, Structurely will also send the following activity:
During Conversation:
Each text sent will have 2 activities. The first is Structurely telling us to send the text. The second is either confirmation that we were able to send the text or an activity letting us know the text failed. Remember, Fargo listens for the activity from Structurely and sends those activities to Rookery -> Twilio to be sent. You can see this process play out with these activities.
Structurely telling us to send:
Our Confirmation the message was sent. The messaging_text_id will match on activities for the same message assistant_conversation_id will be consistent for the whole lead convo.
Our confirmation that the message failed to deliver:
Finally, here is the activity that we send for lead responses that are sent to the assistant integration:
Conclusion:
Hopefully this serves to answer any questions you could have about the Assistant Conversation integration. It is still in beta, and is very likely to have bugs. Clients who are willing to try this out should be encouraged to provide us with Feedback. That feedback should be sent to Jared in product.