You must have done a similar project and be able to demonstrate this.
We have sales data for thousands of customers.
Some customers spend every day, some once per year.
We want to forecast weekly sales.
Some customers will have 0 sales in weeks. this is valid. this does not mean we do not have data, it means the customer does not buy every week.
We wish to re-forecast every customer on a schedule (e.g. weekly) and on demand.
The only data we have available is sales transactions, for example:
Date | Customer | Product | Amount | Quantity
We can feature engineer this data to show sales in any period and by features such as:
a) spent in last 7 days
b) spent in last 8-14 days
c) spent in last 15-21 days etc.
d) total unique spending days
e) days since last spend
f) days between spend (average, std dev, etc)
g) total unique products purchased
h) time 'chunks' of 7 days, 30 days, 365 days, calendar weeks, months, whatever, etc
g) overall propensity to spend (e.g. 18% of customers spend 3-5 times and this customer has spent x times)
You can feature engineer anything you like, or if you tell us what you want, we can do it to prepare the data.
What we cannot do is provide more data on the customer to assist the forecast. We only know the customer number, days on which they spent, amounts they spent, and quantities purchased.
The challenge we see is there is lots of sparsity in the data.
For every customer, the questions we wish to answer are:
1) will this customer spend again in the next 365 days (yes or no)
2) what is the probability score of this customer spending again in the next 365 days
3) what is the amount this customer will spend in the next 365 days
4) what is the probability score of this customer spending the 365-day forecasted amount in the next 365 days
5) will this customer spend again in the next 7 days (yes or no)
6) what is the probability score of this customer spending again in the next 7 days
7) what is the amount this customer will spend in the next 7 days
8) what is the probability score of this customer spending the 7-day forecasted amount in the next 7 days
If you can assist in answering all the questions with the available data let's talk.
Budget. Talk sensibly about the project and we can establish a budget acceptable to both of us. But we need to see experience with the problem and early prototype results to work together.
23 pekerja bebas membida secara purata $60/jam untuk pekerjaan ini
I have done a project on time series. So I feel i can do this and provide you the solution for your requirement. Looking forward to have a conversation with you.