c***z
发帖数: 6348 | 1 This was a project that I attempted, but didn't finish because I changed
company. Still, it would be great if you can share your insights with us.
The company spend money on many channels of advertising, such as TV, radio,
newspaper, etc
We do keep track of the sources of each purchase and leads (possible
purchases). However we do believe that the interaction between the channels
is huge, as well as the temporal effects. For example, you might have heard
of us in the radio last week, and decided to purchase online today because
you saw our commercial on TV.
On the form of purchase, there will be an optional item called "source of
information", you may have filled in "TV" since you already forgot about the
radio.
Even worse, if you didn't fill in the item, since your purchase is online,
we will assume your source of information is "online".
We want to adjust our budget and invest more on promising channels. What
would you do?
Thanks! | r*******y
发帖数: 626 | 2 This is actually a big challenge in marketing: how to track users online and
offline behaviors.
Some thoughts:
1. Put different information through different media channels. For example,
use different URLs or different promotion codes in ads.
2. Run ads at different time and then observe online traffic. For example,
run TV ads in week 1, radio ads in week 3, no any ads in week 5 etc.
3. Study a small sample of representative users
,
channels
heard
【在 c***z 的大作中提到】
: This was a project that I attempted, but didn't finish because I changed
: company. Still, it would be great if you can share your insights with us.
: The company spend money on many channels of advertising, such as TV, radio,
: newspaper, etc
: We do keep track of the sources of each purchase and leads (possible
: purchases). However we do believe that the interaction between the channels
: is huge, as well as the temporal effects. For example, you might have heard
: of us in the radio last week, and decided to purchase online today because
: you saw our commercial on TV.
: On the form of purchase, there will be an optional item called "source of
| t*********l
发帖数: 52 | 3 我研究经济，如果我做这个项目，会加上一个步骤，文献综述，从而找到各种媒体之间
的关系，然后决定项目的方向。
,
channels
heard
【在 c***z 的大作中提到】
: This was a project that I attempted, but didn't finish because I changed
: company. Still, it would be great if you can share your insights with us.
: The company spend money on many channels of advertising, such as TV, radio,
: newspaper, etc
: We do keep track of the sources of each purchase and leads (possible
: purchases). However we do believe that the interaction between the channels
: is huge, as well as the temporal effects. For example, you might have heard
: of us in the radio last week, and decided to purchase online today because
: you saw our commercial on TV.
: On the form of purchase, there will be an optional item called "source of
| a******a
发帖数: 57 | 4 Some people use 1st click model, putting most weights on the 1st click. Some
people use last click model. Some use even attribution model.
,
channels
★ 发自iPhone App: ChineseWeb 7.8
【在 c***z 的大作中提到】
: This was a project that I attempted, but didn't finish because I changed
: company. Still, it would be great if you can share your insights with us.
: The company spend money on many channels of advertising, such as TV, radio,
: newspaper, etc
: We do keep track of the sources of each purchase and leads (possible
: purchases). However we do believe that the interaction between the channels
: is huge, as well as the temporal effects. For example, you might have heard
: of us in the radio last week, and decided to purchase online today because
: you saw our commercial on TV.
: On the form of purchase, there will be an optional item called "source of
| w*****a
发帖数: 218 | 5 本人几个月前作了一个同样的项目
觉得找到一个比较可靠的办法来解决这个问题
并写了一篇小文章
因为公司太大，Politics 的原因，
很难弄到真实数据 - 后来才知道，我要的数据虽然看起来简单，就是
（每天或每周的 公司总利润， 和各个 channel 的花费）
但这样的数据连公司 CMO 都弄不到
中间看过一家小公司的数据
感觉能提高他们原来的 ROI by 20%-30%
在下对这一问题非常有兴趣
如果可能的话，愿意 Part Time （测试结果出来前免费）参与各位的类似项目
更重要的是通过这样的机会
想在真实市场条件下检验一下本人的模型和方法
各位若有兴趣，欢迎站内信联系
,
channels
heard
【在 c***z 的大作中提到】
: This was a project that I attempted, but didn't finish because I changed
: company. Still, it would be great if you can share your insights with us.
: The company spend money on many channels of advertising, such as TV, radio,
: newspaper, etc
: We do keep track of the sources of each purchase and leads (possible
: purchases). However we do believe that the interaction between the channels
: is huge, as well as the temporal effects. For example, you might have heard
: of us in the radio last week, and decided to purchase online today because
: you saw our commercial on TV.
: On the form of purchase, there will be an optional item called "source of
| y******u
发帖数: 804 | 6 有点像HMM，但又不是很典型，不知道能不能捣鼓出来 | c***z
发帖数: 6348 | 7 各位的发言都是高质量的啊
我当时的研究还在初级阶段，主要在考虑configuration theory和fussy stat， 后来
没有继续了 | w***o
发帖数: 110 | 8 http://www.turn.com/downloads/white_papers/TURN_Tech_WP_Data-dr
最近的一个project与这个相关。
,
channels
heard
【在 c***z 的大作中提到】
: This was a project that I attempted, but didn't finish because I changed
: company. Still, it would be great if you can share your insights with us.
: The company spend money on many channels of advertising, such as TV, radio,
: newspaper, etc
: We do keep track of the sources of each purchase and leads (possible
: purchases). However we do believe that the interaction between the channels
: is huge, as well as the temporal effects. For example, you might have heard
: of us in the radio last week, and decided to purchase online today because
: you saw our commercial on TV.
: On the form of purchase, there will be an optional item called "source of
| w*****a
发帖数: 218 | 9 这篇文章之所以影响大
是因为它差不多是这个领域的开山之作
但凭心而论很初级
也可能是大数据原因，只能是这个玩法
另外，这篇文章是解决这个问题的一个办法 - 如果你有user impression / click /
conversion 等等所有的数据
但不是唯一的办法
【在 w***o 的大作中提到】
: http://www.turn.com/downloads/white_papers/TURN_Tech_WP_Data-dr
: 最近的一个project与这个相关。
:
: ,
: channels
: heard
| M***e
发帖数: 531 | 10 这个基本无解吧。。相当于数据未必真实，那也没有地方去找groundtruth..
有什么数据就用什么数据，然后加一句：based on our current data.. | | | p****o
发帖数: 1340 | 11 读了一下。这篇文章更多是估计每个channel的attribution。
但这个问题本身是整体资源优化的，简单attribution并不对。
【在 w*****a 的大作中提到】
: 这篇文章之所以影响大
: 是因为它差不多是这个领域的开山之作
: 但凭心而论很初级
: 也可能是大数据原因，只能是这个玩法
: 另外，这篇文章是解决这个问题的一个办法 - 如果你有user impression / click /
: conversion 等等所有的数据
: 但不是唯一的办法
| m******e
发帖数: 21 | 12 hand 同做这方面的工作
radio
because
of
【在 w***o 的大作中提到】
: http://www.turn.com/downloads/white_papers/TURN_Tech_WP_Data-dr
: 最近的一个project与这个相关。
:
: ,
: channels
: heard
| h***t
发帖数: 2540 | 13 Indeed this paper is not as influential as you said, cited only 10 times so
far, probably on KDD industry track, which is 3 times acceptance rate of KDD
research track.
【在 w*****a 的大作中提到】
: 这篇文章之所以影响大
: 是因为它差不多是这个领域的开山之作
: 但凭心而论很初级
: 也可能是大数据原因，只能是这个玩法
: 另外，这篇文章是解决这个问题的一个办法 - 如果你有user impression / click /
: conversion 等等所有的数据
: 但不是唯一的办法
| r***e
发帖数: 1840 | 14 coupon code:
During period 1: distribute the coupon code on radio for example.
During period 2: distribute a different code on TV.
Of course the discount rate should be the same. And we should adjust the
sale number by the burnout rate as well. | R*******y
发帖数: 86 | 15 marketing工作重点应该是减少每个user在channel之间的交替，最好一次命中。
我们公司前两年有这个挑战，最近来了个牛人，硬是把乱生生的channel理顺了。
话说回来，我是做数据分析的，
但我选择不作为（暂时接受last paid solution)。
不是所有数据优化了就好，有的时候big picture更重要。 |
|
