Initiating Conversations (P2)
Your Guide to Asking the Right Questions in consumr.ai
Picking up from where we left off in the article, “Initiating Conversations (P1),” you are now aware of the directions you need to take to reach your destination in the map of consumr.ai and begin the conversation. But the last article was purely navigational. This article will be make for an interesting read because we will learn about the types of questions we need to ask, how, and when.
So, let’s dive right in. As described in the last article, asking the right question will set you in the right direction. If you ask a question that heads you down south, where you wanted to be headed north, then you are bound to lose time, money, and put in a lot more effort. Worse, give up. So, let’s learn how to ask the right questions.
The Scope
Before we start asking questions, let us first understand the scope of conversations in consumr.ai. “In consumr.ai” being the keyword. Conversations as a term seem very chatty and broad, and you do not want to be asking dating questions to a consumer intelligence platform. So let's define our boundaries. Conversations with consumr.ai ideally work on the grounds of knowledge that the platform accumulates about the brand in discussion. This knowledge is stored and used as per the portfolio we are working with. The knowledge bank comprises 3 types of intelligence:
Internal: Knowledge that the platform draws from a series of reports created using consumr.ai modules.
Organizational: Knowledge that the user uploads onto the platform as a file. This is intelligence that is known at an organization level and is internal or even confidential to the organization. Uploading such information through a file enables consumr.ai to learn from it and help you answer a question using either the file alone, or by merging the knowledge with Internal and/or general.
General: This type of knowledge is usually gathered by the system if you allow the system to take creative liberties, or if you ask a question that is too broad, and the system still tries to get you an answer using the knowledge available on the internet. This happens when consumr.ai initiates a search for and gathers the top results in search and finds out the knowledge from it.
Now, going back to my casual example of asking dating questions, if your brand in the portfolio is not ‘Tinder,’ then it is likely that consumr.ai will use the third type of knowledge to still curate an answer for you, but it will not be relevant to your brand.
Precision
Now that we have our constraints in place, let us discuss the ‘Precision’ that is connected to the scope. LLMs are interesting with the way they can put things into perspective so quickly, but it also has a creative side that makes them hallucinate a bit. This is why precision is a measure that can restrict them from going to their dreamland and pinch it every now and then to keep them on track. In consumr.ai, conversations have 3 settings of precision as shown below. The user will have to select anyone of them based on the expected responses. By default, it will always be tuned to ‘Analyze’.
NOTE: Precision modes are only applicable for co-pilot
1. Analyze: In this mode, the AI will answer questions related to the "assets in use" exclusively. This is the default mode when you choose assets from the history table and start a conversation. In this mode, if a question is asked that cannot be answered from the existing assets, the user will be suggested to switch to another mode. The advantage of this mode is that its super-fast and is 100% factual with very little chance for AI hallucinations.
2. Investigate: In this mode, the AI will first look up the "assets in use" if it cannot find an answer within it, it will do a best effort search across all reports in the last 30 days in the current portfolio. If it finds appropriate assets that can answer the question it will, else it will prompt the user to switch to Explore mode. In this mode, since there are additional searches, the response time will be slightly slower than in Analyst mode. Hallucinations are possible, since it might think a report is relevant, but it might turn out to be exactly what was implied by the user. This is the default mode when you start a question directly from omnibox.
3. Helpful: This is the mode powerful and intelligent mode in conversations. This mode combines the analyst and investigate capabilities, along with the ability to generate "custom reports". These are not your typical behavior, intent or mentions reports. Rather they are a combination of cards that are specifically created to answer the user's question. This is an agentic flow. Which means that the AI creates a plan, selects the exact tools that are required to answer the question and then orchestrates it all to produce a final answer. Its highly recommended to see the demo video to understand how this works better. Needless to say, this is the most expensive mode. In the sense, that its computationally heavy, commercially more expensive and time consuming.
This means that if you ask a question that it cannot answer in a particular mode, it will ask you to change the mode to get the answer.
The above question was asked in "Analyze" mode. Since adding new assets intelligently is not a function of Analyze it asked the user to change the mode or add new assets. You can click on the other modes and hit "Proceed".
In "Investigate mode" it explores the whole list of assets with in the organization outside of what is added in the conversation. The above message is what you will get if there are not enough assets in the organization. You can tweak the question or switch to explore.
Switching to explore more will activate the true nature of co-pilot, where you can either let the AI understand the situation and create required assets to get an answer to the question, or you can take control and manually create reports. Lets ask AI to handle it.
Asking AI to handle it will open a right frame that will show you the complex processing it does to get you the answer. As you can see the right frame has a toggle. If you select Analysis, it will show you the whole complex code that it is writing to get you the answer. Selecting preview will give you graphical insights of the data, as shown below.
Once the answer is ready, it will give you the answer inside the conversation
Use the text box below to ask more questions and take the conversation from there.
Asking Questions
Finally, let's jump into asking questions now that you are already aware of the pre-sets and constraints. There are two ways of asking questions:
Selecting one of the recommended questions.
Asking a custom question.
Let’s cover both in depth.
Recommended Questions:
Situation: You are on the conversations pop-up and start a new conversation. Strangely, you find the chat box to be empty and no assets are loaded. Well, it is how it is designed to be. The engine needs to know your intent and cannot guess it. So, when you initiate a new conversation, you either must ask a question and let it find the right assets to add, or you add the assets and let it recommend questions for you. In this section, let’s go with recommendations.
To get recommendation of questions to ask, click on the Add Asset button on the top of the left panel of the box.
A 2nd left panel will slide in giving you the option to add 3 types of assets:
Add URL as an asset: Just put in a URL you want to use the information from and add it as an asset. We recommend putting in a specific page. Giving the homepage URL for a large brand for example, will give you vague information with in the asset.
Browse an Asset: This will allow you to search for a pdf or pptx file on your machine or cloud.
Add from asset library: This will give you the option to select any asset that you may need to make this conversation more intelligent. This library will mostly include the insight reports created by consumr.ai
Select the option you like and "add" the assets to the conversation. Once you do, watch consumr.ai study the assets and in seconds curate 4 recommendations of questions you can start with.
Of course, you don’t have to stick with them, but if you do, then click on any one of them and get an answer to it in seconds. Let’s select the first one.
You can see that the conversation is auto named and the engine has given you a textual reply that is simplified. However, if you want view the answer in detail, click on the 3 vertical dots next to each report and click on " Open Asset in a new Tab".
You will get an option to open this insight in a new tab. Click it and a new tab will open with the report, just as it read in the menu.
You click on summary and the chat will summarize the report for you. Let’s click it.
In addition, you want to action an answer or add it to the bank of Research intelligence, hover on the answer to see 3 horizontal dots appear in the row of the answer. Click on the first option and the answer will be save to the knowledge bank of research, to be pulled up at another date in a conversation or other flows that permits research-based knowledge to be inherited.
What questions should I ask?
Asking questions is not the difficult task. Asking the right one is. However, since this is about your business, you are the best person to do so. I have put down a list of Do’s and Don’ts for you to follow, and I'm sure it will all work out just fine. You may even find a new buddy in the form of consumr.ai.
Dos’
Be Precise: Broader questions will get you very broad answers, and that may not help you.
Start your questions with what.
Keep your questions related to the assets added.
Before you ask questions, read about the modules of consumr.ai: This will help you with the scope of your curiosity.
Stay in the marketing arena: consumr.ai is a consumer intelligence platform, and it may not give you the right advice on other matters.
Ask a series of questions that require shorter and specific answers. Then put them all together.
Don’ts
Avoid starting with why: Why will bring in speculation, and the model will not take you down the right path.
Don’t ask for opinions on strategy: There is a reason why humans are still managing companies and not AI.
Don’t add just one asset or too many assets: Too many assets may include too much information that can confuse the model, due to contradictory information. For example, adding 20 behavior reports created with the same input, over a period of time, and then asking it to give one answer at a point in time and not asking about trends.
Last updated