I was wondering if this competition is really a prediction. \nWe are training our model using 2016-08-01 to 2017-08-01 data to predict the total revenue in the same year. Then we apply this model to the test set (2017-08-02 to 2018-04-30) in order to predict the total revenue per customer in this period: the company knows how much customers spent in this period. Do we need a predictive model to predict how much customers spent during last year? Is there any other application of such a model?
", "messages": [ { "id": 402786, "postDate": "2018-10-12T10:09:43.703Z", "content": "I was wondering if this competition is really a prediction. \nWe are training our model using 2016-08-01 to 2017-08-01 data to predict the total revenue in the same year. Then we apply this model to the test set (2017-08-02 to 2018-04-30) in order to predict the total revenue per customer in this period: the company knows how much customers spent in this period. Do we need a predictive model to predict how much customers spent during last year? Is there any other application of such a model?
", "rawMarkdown": "I was wondering if this competition is really a prediction. \nWe are training our model using 2016-08-01 to 2017-08-01 data to predict the total revenue in the same year. Then we apply this model to the test set (2017-08-02 to 2018-04-30) in order to predict the total revenue per customer in this period: the company knows how much customers spent in this period. Do we need a predictive model to predict how much customers spent during last year? Is there any other application of such a model?", "votes": 5 }, { "id": 403272, "postDate": "2018-10-13T08:09:05.167Z", "content": "Very good question. My take on it is that a model out of that data could be applied as input into something like 'real-time-offer' management system. Such a system would use the prediction whether and how much a customer is most likely to spend at the start when visiting a website again to offer a coupon/swag/free shipping/etc. to help nudge a customer to hit the buy button. This would require to omit totals.hits and totals.pageviews obviously but would allow to use demographics, previous/historic sessions statistics and word vectors of referralpath/searchwords/etc.
", "rawMarkdown": "Very good question. My take on it is that a model out of that data could be applied as input into something like 'real-time-offer' management system. Such a system would use the prediction whether and how much a customer is most likely to spend at the start when visiting a website again to offer a coupon/swag/free shipping/etc. to help nudge a customer to hit the buy button. This would require to omit totals.hits and totals.pageviews obviously but would allow to use demographics, previous/historic sessions statistics and word vectors of referralpath/searchwords/etc. ", "votes": 1 }, { "id": 402834, "postDate": "2018-10-12T12:32:11.770Z", "content": "Yes it is (skipping the leak issue for the moment :-) Think of it like this: the company has historical data and wants to predict the spending for existing customers over the entire next year going forward.
", "rawMarkdown": "Yes it is (skipping the leak issue for the moment :-) Think of it like this: the company has historical data and wants to predict the spending for existing customers over the entire next year going forward.", "votes": 1, "replies": [ { "id": 402839, "postDate": "2018-10-12T12:39:05.740Z", "content": "If you train a model to predict actual spending, using actual information like page views, clicks etc. How can you predict future behaviour. If you want to predict future behaviour, in my opinion, you should train model using actual behaviour and next year spending. Once you got the model then you can predict future behaviour.
", "rawMarkdown": "If you train a model to predict actual spending, using actual information like page views, clicks etc. How can you predict future behaviour. If you want to predict future behaviour, in my opinion, you should train model using actual behaviour and next year spending. Once you got the model then you can predict future behaviour." }, { "id": 402844, "postDate": "2018-10-12T12:46:30.123Z", "content": "Agreed - in order to be realistic, info like \"pageviews\" and \"bounces\" should be removed for the future observations.
", "rawMarkdown": "Agreed - in order to be realistic, info like \"pageviews\" and \"bounces\" should be removed for the future observations.", "votes": 1 }, { "id": 402849, "postDate": "2018-10-12T12:57:39.017Z", "content": "Yes. But if you remove customers behaviour from the model, do you really need a ML algorithm? Or is it enough an analysis?
", "rawMarkdown": "Yes. But if you remove customers behaviour from the model, do you really need a ML algorithm? Or is it enough an analysis?" } ] }, { "id": 402831, "postDate": "2018-10-12T12:24:23.883Z", "content": "Such a model is not predicting future customers spending.. it is predicting actual customers spending.
", "rawMarkdown": "Such a model is not predicting future customers spending.. it is predicting actual customers spending. ", "votes": 1 }, { "id": 402822, "postDate": "2018-10-12T11:57:21.130Z", "content": "@Armando Puglisi
\n\nYo got it : it is called a prediction, even if it has been occuring in the past...
\n\nScusami, ma l'opportunità è stata così bella ;-)
", "rawMarkdown": "@Armando Puglisi\n\nYo got it : it is called a prediction, even if it has been occuring in the past...\n\nScusami, ma l'opportunità è stata così bella ;-)", "votes": 2, "replies": [ { "id": 402833, "postDate": "2018-10-12T12:31:10.193Z", "content": "We are training model to predict actual spending. What is the value of such a model if in real life we know actual value?
", "rawMarkdown": "We are training model to predict actual spending. What is the value of such a model if in real life we know actual value?" } ] }, { "id": 402876, "postDate": "2018-10-12T13:47:59.820Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 402891, "postDate": "2018-10-12T14:22:12.440Z", "content": "So the competition is not Predicting Customer Revenue but: known the current customers behaviour, predict how much they would spend (using the model) in the same year and compare this value with the real actual value. In this way they could segment customers according to \"prediction\" and real behaviour.
", "rawMarkdown": "So the competition is not Predicting Customer Revenue but: known the current customers behaviour, predict how much they would spend (using the model) in the same year and compare this value with the real actual value. In this way they could segment customers according to \"prediction\" and real behaviour." } ] } ], "comments": [ { "id": 403272, "author_name": "glennneuber", "author_url": "", "post_date": "2018-10-13T08:09:05.167000", "content": "Very good question. My take on it is that a model out of that data could be applied as input into something like 'real-time-offer' management system. Such a system would use the prediction whether and how much a customer is most likely to spend at the start when visiting a website again to offer a coupon/swag/free shipping/etc. to help nudge a customer to hit the buy button. This would require to omit totals.hits and totals.pageviews obviously but would allow to use demographics, previous/historic sessions statistics and word vectors of referralpath/searchwords/etc.
", "votes": 1, "replies": [] }, { "id": 402834, "author_name": "Konrad Banachewicz", "author_url": "", "post_date": "2018-10-12T12:32:11.770000", "content": "Yes it is (skipping the leak issue for the moment :-) Think of it like this: the company has historical data and wants to predict the spending for existing customers over the entire next year going forward.
", "votes": 1, "replies": [ { "id": 402839, "author_name": "Armando Puglisi", "author_url": "", "post_date": "2018-10-12T12:39:05.740000", "content": "If you train a model to predict actual spending, using actual information like page views, clicks etc. How can you predict future behaviour. If you want to predict future behaviour, in my opinion, you should train model using actual behaviour and next year spending. Once you got the model then you can predict future behaviour.
", "votes": 0, "replies": [] }, { "id": 402844, "author_name": "Konrad Banachewicz", "author_url": "", "post_date": "2018-10-12T12:46:30.123000", "content": "Agreed - in order to be realistic, info like \"pageviews\" and \"bounces\" should be removed for the future observations.
", "votes": 1, "replies": [] }, { "id": 402849, "author_name": "Armando Puglisi", "author_url": "", "post_date": "2018-10-12T12:57:39.017000", "content": "Yes. But if you remove customers behaviour from the model, do you really need a ML algorithm? Or is it enough an analysis?
", "votes": 0, "replies": [] } ] }, { "id": 402831, "author_name": "Armando Puglisi", "author_url": "", "post_date": "2018-10-12T12:24:23.883000", "content": "Such a model is not predicting future customers spending.. it is predicting actual customers spending.
", "votes": 1, "replies": [] }, { "id": 402822, "author_name": "mezoganet", "author_url": "", "post_date": "2018-10-12T11:57:21.130000", "content": "@Armando Puglisi
\n\nYo got it : it is called a prediction, even if it has been occuring in the past...
\n\nScusami, ma l'opportunità è stata così bella ;-)
", "votes": 2, "replies": [ { "id": 402833, "author_name": "Armando Puglisi", "author_url": "", "post_date": "2018-10-12T12:31:10.193000", "content": "We are training model to predict actual spending. What is the value of such a model if in real life we know actual value?
", "votes": 0, "replies": [] } ] }, { "id": 402876, "author_name": "", "author_url": "", "post_date": "2018-10-12T13:47:59.820000", "content": "", "votes": 2, "replies": [ { "id": 402891, "author_name": "Armando Puglisi", "author_url": "", "post_date": "2018-10-12T14:22:12.440000", "content": "So the competition is not Predicting Customer Revenue but: known the current customers behaviour, predict how much they would spend (using the model) in the same year and compare this value with the real actual value. In this way they could segment customers according to \"prediction\" and real behaviour.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "402786": "I was wondering if this competition is really a prediction. \nWe are training our model using 2016-08-01 to 2017-08-01 data to predict the total revenue in the same year. Then we apply this model to the test set (2017-08-02 to 2018-04-30) in order to predict the total revenue per customer in this period: the company knows how much customers spent in this period. Do we need a predictive model to predict how much customers spent during last year? Is there any other application of such a model?", "403272": "Very good question. My take on it is that a model out of that data could be applied as input into something like 'real-time-offer' management system. Such a system would use the prediction whether and how much a customer is most likely to spend at the start when visiting a website again to offer a coupon/swag/free shipping/etc. to help nudge a customer to hit the buy button. This would require to omit totals.hits and totals.pageviews obviously but would allow to use demographics, previous/historic sessions statistics and word vectors of referralpath/searchwords/etc. ", "402834": "Yes it is (skipping the leak issue for the moment :-) Think of it like this: the company has historical data and wants to predict the spending for existing customers over the entire next year going forward.", "402831": "Such a model is not predicting future customers spending.. it is predicting actual customers spending. ", "402822": "@Armando Puglisi\n\nYo got it : it is called a prediction, even if it has been occuring in the past...\n\nScusami, ma l'opportunità è stata così bella ;-)", "402876": "" } }