RallySeller makes a great component to your marketing campaign. In a recent CuriousityStream affiliate campaign there was a 200% increase in CTR for Send to Messenger ads compared to the original flow. This means, where people were clicking the ad and subscribing to the bot (at some cost to you), then ignoring the first message and losing interest, RallySeller was able to entice 200% more activity from people because of the reinforcement learning strategy. See another recent case study here.
Each RallySeller instance is designed to handle one product or service at a time, each offer would contain 16 selling points. It would be possible to use RallySeller in a more generalized way, for example to answer general questions about a collection of products or offers, however it would work best to focus on just one at a time. One RallySeller instance can be used for some period to promote an offer and at any time this can be changed to another. If you'd like to focus on 2 offers at the same time, you can use the generalized approach (handling 16 selling points about both offers at the same time), or use 2 instances of RallySeller to handle each one separately although, this would require 2 monthly plans.
Not at all. This has nothing to do with the machine learning Google's DialogFlow uses to process user input. RallySeller does not respond to customer inquiries, instead controlling the path of the conversation with A.I. selected prompts -- this creates the sense of a conversation between subscriber and the bot.
Small edits to your ad copy such as a price change can be made without affecting the learning process. However, if making a major change to the topic then this requires resetting the chatbot and all previous learned behavior would be lost. Either way, you can upload a new Google Sheet with selling points at any time.
No, RallySeller can be connected to a chat widget, opt-in message or additionally placed in any other conversation or flow, for example you may have a flow with calls-to-action, or other messaging, and can just include another button “More Info” which would then go to the RallySeller sequence.
Template installation can generally be done in a few minutes. It involves replacing placeholder images, modifying any text to your preferences, and turning on the “rules” – this video walks through the process in just 2 minutes.
Yes it will, when you first install RallySeller and get everything setup no doubt some tests will be done to go through the sequence and see that it works – all of your interactions will then become encoded in the sales agent’s memory so that it will think these are good replies – in order to get around this you’ll have to first do the “testing phase” over some period of time, and then contact support to have your agent reset (generally takes less than 1 day), after the restart you can then release RallySeller into the wild. At this point, any interactions the bot has (whether for testing or real leads) will become part of the learning phase.
Given the amount of possible dialog trees it can easily take between 100 to 2000 interactions before the bot is trained. The default setting will be to go through extensive training for the first 200 leads (this means heavy experimentation and exploration with messaging with a mix of re-using past successes). After that, the balance will have decreased to almost zero (where 99% of the time RallySeller will just repeat what has worked before). You can request this setting to be adjusted to preference. For example, you may never want RallySeller to stop experimenting with possible combinations, in which case the agent can be locked at a 50/50 split at some point. Also, you can have the default setting of 200 dialogs decreased or increased. In previous instances, the chatbot was able to outperform other chatbots in just the first 15 interactions because it will at least not just keep repeating noneffective messaging. It is recommended that you give RallySeller the most possible live data to train on – hundreds or more conversations. Send an email to email@example.com to discuss your strategy and determine which learning rate setting is right for your case.
Besides measuring results on taps and message views within ManyChat you will have access to a performance chart to see cumulative scoring. Each interaction has a different score, for example more points are given for call-to-action taps, than for anything else. There are also “penalties” for not getting a response. Ideally, these scores will reach a plateau on the higher end of the y-axis which demonstrates the chatbots effectiveness. At this point, it would be possible to simply duplicate the best working sequences and not use the RallySeller API at all, having manually replicated what works best. This would then allow for the restarting of another RallySeller instance with all new messaging, or for a different offer.
A maximum of 5 messages are sent, with the 5th containing just a call-to-action with no further prompts. It would also be possible to make the last step a re-entry point to start another conversation. These limits are in place because if increased, the amount of possible dialog trees would continue to grow exponentially to a point where it would be unrealistic to train the bot in a reasonable amount of time.
Generally, yes in most Python based experiments tens of thousands of instances of a game are run before conversion or optimal behavior occurs. However, the "game" in this case being a dialog of 4-5 messages in length is much smaller and while it would be ideal to run the sales agent through training for thousands of instances before finding the best possible results RallySeller has proven to be effective in that, unlike most other flows (even using extensive AB testing manually) the agent won't repeatedly use non-working messages so good results can be seen in as little as 20-30 interactions.
Choosing a selection results in a full page refresh.
Press the space key then arrow keys to make a selection.