Is your spending bring you revenue?

Follow along and do this analysis for free with your data @ www.humanlytics.co.

The relationship between spend and revenue seems simple. Yet, the more you peel behind the veil and think about the detailed mechanics of their relationship, the more complicated it becomes.

On the one hand, spend does not directly result in revenue for your company — the only thing spend on advertising, organic, or other resources do is bringing more eyeballs to your offering and business. It is up to your offer and website to convert those eyeballs into cash for your business.

On another side of the story, revenue does not need spend to be generated. There are plenty of cases where companies achieve great success without almost no spend on marketing (Sriracha comes to mind as a great example).

Once upon a time, I had a client that had cash flow issues and had to pause all of their advertising for a couple of months. To their great surprise, they still generated a significant amount of revenue with close to zero marketing spend — the company produced more profit without spend than with spend.

This is not to say that you should stop all of your marketing spend immediately. However, it does come into question whether your marketing spend is driving positive revenue for your business. That is, is it bringing more revenue/profit than the money you spend on it?

This article will show you two analyses that can serve as a great starting point for you to figure this out for your own business. While I am not claiming those analyses to be exhaustive or deep, many of you will find surprising results when you look through your data (which you can do for free at www.humanlytics.co, no strings attached).

Setting up the analysis

The first analysis we are going to do is a simple time series analysis that compares your trend of revenue and spend over time.

You will need to produce a “double y-axis time series chart” with the axis on the left representing revenue and the axis on the right representing spend. Being a time-series chart, the x-axis of the graph will be the date of this analysis.

We use two y-axis instead of one because the differential between your spend and revenue is vast. If we plot both on one y-axis, your spend will look flatter than it is — obscuring the analysis.

For this analysis, I recommend using a full year of data so you can view the complete trend of those two variables, though any more or less will produce slightly different insights and are time periods all valid.

For the second analysis, we will need to create a scatter plot of spend and revenue, with spend on the x-axis (a common place for variables that we can vary, or independent variables) and revenue on the y-axis (a common place for variables that we do not have direct control over, or dependent variables).

Each dot on the plot represents a specific day of the analysis period, and I recommend doing this analysis at least over your data of 90 days in length.

After the scatter plot is made, we want to draw a line (mathematically) through the dots and create a simple linear regression line — this will show us the movement of revenue as we observe different levels of spend for the business.

Finally, there are also two important variables that we want to compute here, the R² and the Correlation Coefficient.

The R², ranging from 0–1, represents the strength of the relationship between two variables — that is, how confident we are that they are related. When you are at 0 R², the two variables are completely uncorrelated. If you are at 1 R², looking at a perfect correlation, we usually expect an R² of over .2 in the real world to be a relatively strong relationship.

On the other hand, the correlation coefficient represents the “power” of the relationship between two variables, that is, how much does your change in spend result change in revenue (this is an approximation of your ROI).

For both analyses, recommend using the 7-day running average of your data instead of in its raw form — you will find a much smoother curve and a stronger relationship since this move removes the impact of weekly seasonalities while not harming the implications of the analyses significantly.

I have intentionally left out a lot of details on how to create the linear regression line and how to compute the two variables — since we have a free tool that can do it for you, and explaining it will take quite some time — but use this article as a guide, or simply email me for more questions at bill@humanlytics.co.

https://www.ablebits.com/office-addins-blog/linear-regression-analysis-excel/

The Ideal Outcome

We want a strong relationship between your spend and revenue for both analyses.

For our time-series analysis, this comes in the form of revenue and spend moving with each other, as the graph above has indicated.

The strength of this relationship may depend on how much of your revenue is dependent on spend (if you primarily market through organic means, this relationship will be much weaker than a DTC brand that only spends on paid marketing). However, you should nevertheless see a lift in revenue when you lift your spend.

Now, it is entirely normal to see a slight lag effect of spend on revenue. That is, your revenue lifts a day or two after an increase in spend.

However, you do not want to see your spend increase after 1–2 days of rise in revenue — this is usually caused by your marketing team raising spend in response to an increase in revenue — unless you have an excellent reason to do so, it is going to harm your profitability more than helps.

For our linear regression analysis, we want to see the regression line we produced trending upward with a strong R² (over .2) and correlation coefficient (the higher, the better, over 1 should be a baseline). This indicates that a rise in spend is correlated with a rise in revenue in your business.

Suboptimal results

Unfortunately, most of your data will not come out as ideal as the previous section has shown — so let’s talk about some suboptimal outcomes identified by the analysis and the steps we can take to fix them.

The first common outcome we see is “the passive” outcome — that is, spend remains flat throughout the year while revenue is varying significantly. There is almost no relationship between spend and revenue.

I see this outcome more frequently as companies adopt a more “passive” style of advertising management and rely primarily on “marketing AIs” such as performance max and advantage plus to run most of their marketing.

Assuming that the conversion rate on your website does not change, and there are no impacts from weekly/monthly seasonalities of your business, this style of marketing is pretty optimal — as it provides a stable way of incrementally testing your spend level — however, we all know that the assumption of no seasonality and no change in conversion rate is unrealistic.

What I typically find is that AI algorithms do not significantly vary your spend on a day-to-day basis, nor do they operate in a way that maximizes your revenue and profit (more on that in the following article), resulting in loss in potential profit during a sales event, and loss due to overspending in the low season.

Furthermore, my biggest gripe with this management style as a data scientist is that you are unable to collect data to help you search for your most optimal spend level. If you keep your spend constant across all days, there is no way to know if increasing or decreasing your spend can result in better or worse results for your business — which spells complacency.

My recommendations to you, if you are seeing this “passive” spending pattern in your business, is to vary your spend more significantly and be more brave in testing the impact of different spend levels on your business — this could mean actions such as

• Raise your spend when you expect a higher conversion rate from your website, such as sales events.
• Decrease your spend when you expect a lower conversion rate from your website, such as right after Black Friday sales or during the low season of your business.
• Double or half your spend from a specific channel if you are wondering if it provides you enough value — you will find out quickly if that channel brings you revenue.

Now let’s talk about the second pattern, which I termed the “random.”

Random, like the name it is described as, sees almost no predictable relationship between the increase in spend and revenue — though unlike “passive” spend, it still changes throughout the year.

In some cases, we will see spend goes up for a company up to 10–20 days before the revenue goes up — resulting in a massive loss in profit due to this prolonged period of elevated spend.

Typically, this type of outcome results from the misconception that spend directly causes revenue increase.

Relating to the discussion we had at the beginning of the conversation — if you just increase your spend without changing the offers on your website, you are only sending more traffic.

If you are already converting your customers well from those traffic sources you are spending money on, more traffic may result in more revenue and profit for your business. Still, as each unit of traffic gets more expensive with increasing in spend, there will be a breakpoint in which you are losing money by spending more.

But in most cases, we see advertisers increase spend when the conversion rate from those traffic sources on the website cannot support such an increase in spend — resulting in a massive lift in spend with minimal lift in revenue.

When this happens, you may see your revenue being negatively correlated with your spend — this does not mean that your spend is driving negative revenue — it means that during a period of higher revenue, you are spending less than in a period of lower revenue.

Another cause of this “random” effect may be a practice of “priming” the algorithm ahead of the sale. While it is entirely understandable to start your ads 1–2 days ahead of the sale to build hype and traction, many companies begin way too early and lose too much money from overspending, which results in a decline in the profit of your overall sales.

If you are experiencing this “random” effect in your analysis, consider the following steps:

• Whenever you try to increase spend, ask yourself: do I have a high enough conversion rate on the website to justify an increase in such spend.
• Establish a shut-off point for your spend increases that, if you are not seeing a certain revenue level within a specific period, revert the change immediately to cut further losses.
• Closely monitor your spend and conversion rate daily.
• Prime your spend only 1–2 days before your significant sales event (just to put it here, I don’t think priming works as well as you think).

Caveats and Final Words

Let’s finish this article with some caveats of the analysis and final words.

The two analyses shown in this article are an excellent introduction to the practice of “Media Mix Modeling,” which, unlike attribution analysis, focuses on understanding the relationship between your marketing metrics and KPIs holistically.

One key caveat in the analyses shown here is that you cannot infer causality. No matter how strong the relationship may be, you should not conclude from those two analyses that your spend is causing an increase in your revenue — they are simply too general, and not based on repeated experiments in which those causational effects can be drawn.

This applies to the other directly as well — if you are seeing a lack of relationship between your spend and revenue, it may not mean that your spend is not causing your revenue to go up — there may be other factors, such as your organic presence, seasonality, that is overtaking the effect spend may have on your revenue.

However, with the disclaimers out of the way, there is a higher likelihood of a causational relationship if there is a relationship between your spend and revenue through the analyses above. The lack of a relationship between the two also shows a higher likelihood that the two may be unrelated.

Finally, you can do both analyses presented in this article for free at www.humanlytics.co by plugging in your own data, and you will be able to do this with all of our analyses presented in this series.

Next time, we will show you a couple of analyses you can use to examine the effectiveness of your AI algorithm (hint, they are not very effective).

In the meantime, follow us on Medium, Youtube (for a video discussion of this analysis), and TikTok (@humanlytics) for more marketing analytics tips.