Forecasting using Quicksight and Q

[Vim]

hello, i have got my Quicksight and Q up and running and able to generate various graphs and visuals using natural language query. my marketing users are now asking how they would use this tool for forecasting. using a dataset containing beauty sales info by country for the last five years, I asked “forecast sales of shampoo products over the next four quarters”

first it says " Your exact question cannot be answered".

then it give me this useless graph showing a range vary from tremendous growth to steep losses and notes this as the question it understood " Forecast of Total Revenue by Ship Date week" which is nonsense.

I found some references in some 2022 posts but wondering if there is a recommended way to do that now.

[duncan]

Hello @Vim !

How much review have you done on your Q topic after creating it i.e. have you created synonyms, preferred aggregations, etc… on the q topic level?

I haven’t tried forecasting yet but from my experience with Q topics so far it is best to do a thorough review prior to using the generative aspects of Q.

[Thomas]

Hi @Vim,
As your question seems to touch on multiple different challenges, let me try to answer the ones that I could spot and please let me know if I missed something. Also given that the structure and configuration of your dataset and Q topic is unknown, it is difficult to anticipate the expected behavior that you are looking for (feel free to provide more context if my answers are too generic for you).

Generally speaking, Q leverages Q topics that you can fine-tune and adjust to help Q to understand your business terminology and provide fine-tuned semantical meaning to your data. If you are completely new to this, I can recommend to have a look at the
Mastering Q Topic Creation for Amazon Q in QuickSight recording from our learning series and check out the documentation.

You can use Synonyms in the Data Fields section of the Data tab of your topic to define which measure reflects what you are referring to as “Sales” and you can also define the default date for each dataset in the Dataset section of the Data tab of your topic, as highlighted with the red rectangle on the following screenshot:

As forecasting is no exact science, QuickSight shows a lower and upper bound around the expected value. If you want to understand what algorithms we are using, you can find more details here. You can also configure forecast properties, such as forecast length, seasonality, and boundaries as described here. Looking at your specific question to Q, it’s important to understand that as of now, Q does not understand the instruction given regarding the forecasting properties. Until that changes, you can configure them easily via the properties configuration.

To summarize and provide a similar example: After configuring your Q topic to understand what “Sales” means to you and setting the correct default date, you should be able to ask a question like “forecast quarterly sales of energy products” and receive a response looking like:

After selecting “Add to Analysis”, you can select the light-bulb icon and Edit forecast…

… to adjust the Periods forward property of the forecast length and other properties you want to tweak.

As a result you should be seeing a visual showing quarterly sales numbers, filtered on the type of products you wanted it to be filtered by (in my example only energy products), and a forecast for the next 4 quarters.

Did this answer your question? If so, please help the community out by marking this answer as "Solution!

[Vim]

thank you for the detailed suggestions. will review and respond.