Prior to starting StatBid (my paid search agency) in 2015, I was VP of Digital Marketing for Leslie's Pool Supplies, a national chain with over 900 stores in 35 states. I've managed digital marketing and engineering teams through a variety of challenges over those years.
I recommend taking a look at my LinkedIn profile (and recommendations) for more context!
Let's take the example of a single keyword, which we aren't sure how it'll perform.
We'll need to make a few more assumptions to get started. Hopefully, you can produce these based on similar activity in the account, or on your site.
You'll need an Average Order Value assumption first. If you're targeting a specific product, that product's price is a reasonable place to start. We'll assume $150.
You'll also need a Conversion Rate. You can either use data from a similar campaign, or your site's average. Don't worry--it can be a ballpark, as this will correct course quickly enough. We'll assume 3%.
You'll need to know how much you're willing to pay for advertising, as a percent of revenue. Too high, and your margin erodes. Too low, and you give up market share. This figure is called your Cost Of Sale (COS), and is just the reciprocal of your ROI target. Let's assume you're looking to spend no more than 15% of revenue for these ads.
Now, with the COS and the AOV, you can arrive at a Cost Per Acquisition (CPA) target. This is how many dollars you can spend per conversion. In our case, it's $150 * 15% = $22.50. That's how much we can spend to generate the clicks it takes to produce a conversion.
If we have the Conversion Rate, then we know how many clicks (on average) that takes. At 3% conversion rate, we average an order for every ~33 clicks, so the $22.50 is all we can pay for them and still hit our goal. That's $0.68 per click, and that makes for a GREAT starting point.
But how do you know when to give up? You can use Wolfram|Alpha to find out how many clicks it would take to get to 95% probability of a conversion:
The .03 is our conversion rate, and the .95 is the certainty we'd like to hit. This solves to ~98 clicks, so we'll run our test to at least that many.
If you get a conversion sooner than that (with some sort of upper bound), you can bid based on the real conversion rate. For example, if you got a conversion after 25 clicks, you can bid using that 4% number. If that's actually too high, you'll have collected the additional clicks to know that fairly quickly, so it turns itself back down before it spends very much. If you aren't getting a conversion, you can also bid as if you had only one. If you are at 90 clicks, for example, and still haven't seen a conversion, you could bid as if you had one, bid using the 1.11% rate, and your bid will be cautious, skeptical even, and your spend will be controlled.
Now, this was all for a single biddable entity, perhaps a keyword. What I'd recommend is to keep your long-tail keywords separate from your higher-action keywords, in different AdGroups. As you sift out higher-performers, move them to the main group. Slower-action keywords can stay in the long-tail. You may not even need to bid at the keyword level, if you use this strategy, as you can apply exactly the same logic to the AdGroup level. If the entire Group is a new test, then this will accumulate clicks more rapidly, and reduce costs.
Now, how ever broad you go will set how much risk you're taking on. If you have one keyword, like this example, you're probably only going to spend about $50. If you do that thousands of times, then that number goes up with it.
If you'd like to talk through this, I'd be happy to do so. My VIP link is: https://clarity.fm/roysteves/statbid
I think you might be comparing apples to oranges. The cohort ROI report will be using individual conversions, whereas the LTV report will be using multiple conversions. If your revenue is highly driven by recurring revenue, then you could end up in the situation you're describing.
For example, if you sell an average of 5 $100 items to a given customer, your LTV might be $500. However, if you spend $150 to acquire that customer, then you'd see those reports produce a negative ROI, with a higher LTV.
I wouldn't start giving AdWords credit for non-tagged sales. Rather, determine what your target CPA is for a transaction, averaged across your transactions (regardless of how many past orders they have, unless you're prepared to do some overly complicated targeting). Then, you'll never run up against the challenge of "should I just stop", as you can always optimize your bids to target that specific target.
If, however, you can only compete in your niche if you consider the LTV, then you'll want to model your customers, and exclude the outliers on either end. Build a plausible "model customer", using some median values, and target their LTV like it was an AOV.
Once you have an AOV (literal or LTV-based), then you just need to know how much of that you want to spend. This percent is called the Cost of Sale (COS). Let's say your AOV is $300, and your COS target, based on your budget, is 30%. That means you can spend $90 per sale within that AdGroup, Product Group, or other biddable entity in your account.
With a $90 CPA, you can arrive at a reasonable bid by multiplying by your Conversion Rate. If your conversion rate is 5%, then you're saying that it takes about 20 clicks to get a sale (in aggregate), so you can only spend 1/20th of that $90 for each of those clicks. That's a CPC target of $4.50.
If your average CPCs are coming in lower than that, you're going to beat your ROI target, but you may be giving up market share. If your average is higher, then you might be capturing a lot of market, but you'll be short of your ROI target.
If you update those bids regularly, then they'll tend to normalize at producing what ever return you're aiming for, with the proportional share of market.
If you'd like to get a little more help with the specifics of your situation, feel free to reach out. Here's my VIP link, so it'd be on the house: https://clarity.fm/roysteves/statbid