September 2024
Release Calendar & Notes for September 2024
Last updated
Release Calendar & Notes for September 2024
Last updated
Calendar
Live on App
16-Sep-2024
Live on App
16-Sep-2024
Test
23-Sep-2024
Live on App
16-Sep-2024
Test (Has failures)
23-Sep-2024
One the most crucial phase in any research process is the communication of findings. This is a phase if not handled properly leads to low adoption of the actionabilities from the research. When we speak to an AI twin, we are conducting research. Communicating these findings is hence important. However, while you can share a conversation, it almost feels like watching the entire replay of a match, rather than seeing the highlights. Shareable Cards allows the user to quickly/easily share their findings with others in their organization.
These are found in the contextual (three dot menu) of an answer.
One of the most often asked questions is how many twins do I need? Do you need 1 or 2 or 3? And why limit to only three? These are great questions, and we have some answers based on math.
When you go to create an AI twin now and select a behavior report, the platform will within a few seconds recommend how many twins you need to create. While its only a few seconds, there is a ton of math and AI behind it.
Let's dig into nitty gritties
The basic idea is that you want to ascertain "how much of the behavioral data can be explained" by a twin. The aim is that you want to be certain that a large majority of the cohort you have created can be explained by the AI Twins you are about to create.
This depends on how evenly distributed the data is. Which means if you have a 50/50 age gender split, you would almost always need 2 twins correct? Or if the age distribution is very evenly spread out. Conversely, if you have dominant buckets like 55+ female is hugely dominant, you might just need one twin. Think of the interests as well, if the interests are similar, like Apparel, Fashion, Beauty and dominant or are they very diverse. All of these factors will determine how many twins you need. There is a special mathematical concept called "entropy" that can be used to calculate this diversity. We use AI to of course cluster aspects and then apply this entropy concept with various weights and thresholds that determine how many AI twins are required.
How do I use it
Nothing needs to be done from the user's side. The moment a behavior report is selected, the analysis is initiated and completed in 2.1 seconds. And the number of AI Twin slider is automatically placed where its recommended. This can be overruled by the user and a different number is selectable.
At consumr.ai we are all about real time intelligence. All our API calls are live and we always pull the freshest data available. So, how can the AI twin be any different? The ability for an AI twin to stay "relevant" is key for it be trustable by business. Now when you create a new AI twin, you will be given the option to make it auto-updateable.
What does this do?
If you select this option, then periodically (the checks to see if we need to update are done once a month), the twin's persona & memory will be updated and a new version will be made available. The updation process takes into several things into consideration:
For Behavior: If the check proves that there is a significant shift in the demographics or a spike in certain interests/themes, it will create a new persona, else it will retain the same persona and wait for more significant changes before updating.
For Intent: It will see if new trending keywords are appearing or if the questions/doubts/comparisons have changed and accordingly update its recent memory.
For Mentions: It understands what changes have happen in perceptions across various categories and platforms and either re-enforces an earlier memory or creates new recent memories.
Versioning
Each update creates a new version of the twin. You can go back to previous versions of a twin and also speak to it. There is also a compare mode, where you can chat side by side with different twin versions. The Shareable Card is available in this mode as well to quickly see what is different about the two responses and share that as a learning.
Why we built this?
From the time we showed AI Twin to clients and also played with ourselves, we have realized that the AI twin would be great for providing feedback on communication. The good folks at BBQ guys have been using the AI twin for this as well. Now what if we could take that further and give the AI twin a pair of eyes to "see" visuals. So if you ever wanted the AI twin to react to an image (say a catalogue cover or a print ad), you can attach it and ask the twin what it thinks of it :)
How to use it?
Click on the attachment icon, a normal file window will open, select a png or a jpg image file. Once you select it, ask your question and you should get a response like this:
Why we built this?
Demand Gen is an upcoming product of Google and our clients like Hartford are using it. Like Performance Max, Demand Gen is a complementary but similar product by Google that enables advertisers to focus efforts on brand new prospects (more mid/upper funnel) then Performance Max campaigns. As the name suggests, they lean towards increasing the upper funnel and hence have more focus on engagement ads and channels. Demand Gen campaigns are structured quite similar to Performance Max Campaigns, however they differ in one way. Demand Gen campaigns have "ad groups" instead of asset groups like in P. Max.
What did we build?
The following abilities:
Listing of Demand Gen campaigns and ad groups in Improve
Ability to see optimization opportunities in search themes
Ability to see/recommend optimization opportunities in URL ad groups (which are created from consumr.ai)
Ability to push segments into Demand Gen campaigns in "Draft Campaigns"
