RTB rips the black box part 3:beyond surplus

In this article, I will explain that if you want to maximize profits for the entire promotional program, deciding whether you should bid should be the expected return rate (ROI)rather than the expected profit, which is different from what we described earlier. At Datacratic, we've cut our ROI-based strategy at the end of the year .

Too Little of a good Thing

The full profit of the promotion plan can be expressed as:

By this formula, it seems that maximizing the desired profit for each auction is maximizing the overall promotion plan. But it is not, because it has a constraint: all of the consumption plus is equal to the budget. If you want to maximize the expected profit, you need a higher bid for each request, then you will have fewer bids, and the result is that you may have a lower overall profit. However, this does not mean that you should bid at a low expected profit, I mean you should weigh the profit and consumption.

This problem has been noted by economists long ago, and the measure we want to use is ROI, which is the calculation of ROI.

If you consider each request to be an investment, your feedback control system will tell you that you are now speeding or sub-consumption, and you can increase or decrease the minimum desired ROI threshold. You use this threshold to decide whether to bid, rather than the minimum expected profit threshold. In the financial sector, this threshold is called the lowest acceptable return ( minimum acceptable rate of return (MARR)) or threshold yield (hurdle).

A thought experiment

The following is a hypothetical experiment, which we use to illustrate that ROI -based strategies are better than profit-based.

Assume a push plan budget is $1000 , one-action charge < Span style= "FONT-FAMILY:CALIBRI;" > (CPA) is $1 , actions can be post-display events, such as clicking or converting or watching video. Suppose a and b has an unlimited number of identical requests, unlike their response rate for example, focus on click, Response rate is CTR, focus on conversion, response rate is conversion rate ) and price. We further assume that we have a very accurate response rate estimation and consumption prediction algorithm. We are located in a and b The exposure expectation value is higher than the market value, then the winning probability is 1 .

	Response Rate	Value	Cost	Surplus	Roi	Spend	Imps	Actions	CPA
	0.1%	u$1000	u$750	u$250	33%	$1000	1.333m	1333	$0.75
	0.05%	u$500	u$250	u$250	100%	$1000	4m	< P align= "left" >2000	$0.50

profits are the same in A and b , but in b There is a higher ROI, so it has a lower CPA.

Now suppose another experiment: if the requests in a and B are mixed, there is also a CTR particularly low-priced high X -Flow, What about our two strategies?

pacing Strategy	a Imps	b Imps	x imps	actions	CPA
surplus-based				1500	lang= "en-US" >$0.66
Roi-based	0	4M	0	2000	$0.50

Both strategies have been unsuccessful in buyingXFlow, because whether from profit orROIPoint of view, it is not optimal. However, profit-based strategies cannot distinguish between a and B traffic ( they have the same profit ), so separately purchased 1M of traffic from a and B. ROI -based strategies ignore traffic from a , just like ignoring X, and focus only on high roi Type B traffic.

Therefore, the ROI-based strategy will have a lower CPA than the profit-based strategy .

Conclusion

Advertisers, like fund managers, use ROI to determine what channels or strategies they use to invest. ROI -based strategy, your budget will be spent on the most effective exposure, that is, to minimize the CPA of the promotion plan , and maximize ROI and profit.

RTB rips the black box part 3:beyond surplus