### Using AdWords Bid Simulator to Identify the Point of Diminishing Returns

Well, you can't talk about the AdWords Bid Simulator, cost of sale, diminishing returns without a lot using a lot of raw data. Don't worry, there's a puppy at the end.

When you increase your marketing budget, or your bids on a given AdGroup, the obvious intent is to capture more business at a slightly higher expense. This can backfire, however, and it's important to understand a subtle way it can do so. Everyone knows to watch their overall expense, but there's an undercurrent here worth exposing.

Let us consider this AdWords Bid Simulator screenshot:

This is Google's guess as to what the impact of a change in bid position might be, and while it's nowhere near perfect, they have quite a lot of data, and quite a lot of clever data scientists, so it'll do for our demonstration. The first thing we're going to do is export the data from the AdWords bid simulator into a spreadsheet, so we can play with it. Initially, that'll look like this, after I've expanded the columns to include derivatives of CPC (Cost per Click), CR (Conversion Rate), CPA (Cost per Acquisition), AOV (which I know for the period in question, and we'll use to estimate total revenue), Rev (Revenue, as determined by Conversions * AOV), and COS (Cost of Sale, percent of revenue spent on ads):

First, take a look at the CPC ranges. The AdWords bid simulator indicates a **300% increase in your Max CPC bid only results in an 89.8% increase in the actual CPC**. This is because you're starting to dominate the search auctions at that point, so there are fewer and fewer competitors willing to chase a bid that high, and the Actual CPC starts to lag behind your bid. Going the other direction, you can see that the competition actually gets stiffer, and the change in Actual CPC practically mirrors the bid change.

For the purposes of this exercise, let's say that the goal is to hit a COS of 35% or better--we want to spend no more than 35% of revenue on advertising. Based on this, it becomes clear that we don't want to increase the bids too much, as we won't be able to make up the expense in volume. If we presume that the break-even point is 55% (that is, a Gross Margin of 55%), then the **first three rows all represent shifts that push the Campaign entirely out of profitability**. That's a big scary thing to happen for simply increasing the Actual CPC by $0.53. As it turns out, those fifty-three cents matter quite a bit.

Here's how the COS plays out over the range of values:

Once again, you can see the auction saturation as the difference between +200% and +300% starts to level off. This looks like a pretty typical curve, as you'd expect from diminishing returns. Looks can be deceiving though, as these aren't linear jumps from each proposed bid change to the next. However, we can interpolate what those would be, to see if there's a more recognizable pattern.

This view certainly clarifies that Google is showing you a *lot* more information about bid increases, and a lot less about decreases. You could see that by looking at the data, but when you expand the graph to have a linear x-axis, that fact is even more obvious. This doesn't change very much, though--it's akin to switching between a linear and a log-scale. You still see the diminishing returns kick up rapidly, and then auction saturation kicking in around the midpoint of the graph, around +50% Bid Change.

Yeah, Roy, as I increase bids, I spend more. No kidding. So what?

Well, let's assume you *already *paid for the first 47 conversions (the +0% number), and you paid $3,069.62 for them. If you moved up to +10%, then you'd pay $3,977.07, and pick up 54 conversions. But that's a difference of $907.45 in additional costs, and $1,470.35 in additional revenue. **This means that you have an Incremental COS of 61.72%! **That means that you just added orders to your balance sheets that

*lost you money,*despite the fact that your overall COS is almost exactly on-target at 35.06%.

So, how can it be that the Campaign overall is still on-target and making money, but the orders I just added to the pile were harmful? Let's graph the Incremental COS, along with the overall COS, to see:

There's a gap on the right side of the Incremental COS line, where +0% sits, as it's not meaningful to compare something to itself, and the rest of the points represent the COS of the difference between cost and revenue between that point and the +0% home-point. As you'd expect, the values are lower to the left of home, and higher to the right.

### But wouldn't this logic tell me to turn all my bids almost all the way down, then?

That's the obvious counter-position, and is a point worth considering. Rather than looking at the difference between a proposed bid position and the +0% home-point, let's compare each iteration to its neighbors. This will tell you exactly how much more (or less) you're paying for *each step*. We'll call this the** Iterative COS**, to differentiate it from Incremental COS and Overall COS.

The Iterative COS line is stepped because it's like a derivative of the Incremental COS line, so when the slope of the red line changes, it sets the height of the yellow line. It also clearly demonstrates the weakness of the linear interpolation we used to fill in the other points between the ones that Google gave us--realistically only those points are meaningful. For example, if we look at the steps just below +0% home-point, it looks like there are plenty of values (down to -25%) where the Incremental COS is all 55%, which is of course not the case. Still, a bit hard to interpret...

While it's the statistical equivalent of smearing fingerpaint around, I'll do one more interpolation to smooth out the Iterative COS line, so we have a little easier time reading it.

With this, we can see that the Iterative COS at the home-point is 56%--a shred above our break-even point. So, we *may* want to slightly decrease our bids to shed those unprofitable clicks--but only slightly (and even then, the conversion counts probably put that difference well below a statistically significant level).

A perfectly tuned Campaign's *last* conversion makes a very small, but positive amount. That's how you can tell if you're going to "make it up in volume" or not--if your Incremental COS for the next conversion loses money, well, as they say...

### You can't make up a loss in volume.

**In the end, the whole idea is to take the revenue you'd earn at your current spend, and evaluate any additional revenue separately, to determine if it's revenue you can actually make money with.**

Oh, and I suppose I owe you one other thing:

*Have questions about using the AdWords Bid Simulator? Need help figuring out if your AdWords dollars are being spent correctly? Email me at roy@statbid.com, and I'd be happy to take a look for you, and give you recommendations on how to catch a few more wins along the way!*