You open your Google Ads account, click into keywords, and the one you care about says 2.37 conversions for the month.
Not two. Not three. Two point three seven. Nobody ever called the office and bought thirty-seven hundredths of a service, so the natural conclusion is that the tracking is broken.
It isn't. That decimal is Google telling you, quietly, that it took a single sale and divided it between several of your keywords. It decided which clicks deserved how much of the credit, and it handed out the slices.
That would be a harmless quirk if it stayed in the report. It doesn't. The same slices decide what your automated bidding pays for, and they are the numbers most owners use when they decide which keywords to pause. Read them wrong and you can switch off the keyword that found your best customer, because a different keyword got the bigger slice for closing the deal.
Nobody Bought 0.37 of a Water Heater. Google Ads Still Reports It.
Picture a homeowner whose water heater has started making a noise. One Monday night they search "water heater replacement cost," click your ad, read for a few minutes, and close the tab. Two days later they search "tankless water heater installer near me," see your ad again, and click. That weekend they type your company name into Google, click the ad at the top, and call you. You book the job.
One customer, one sale, three clicks on three different keywords.
The old way of counting, called last click, handed the whole sale to the last one: your company name. The first two keywords got nothing, even though one of them is how this person heard of you.
Data-driven attribution, which is now the default in Google Ads, splits it. Say it decides the first search was worth 0.37 of the sale, the second 0.18, and the name search 0.45. Those three numbers add up to exactly one, because there was exactly one sale. Each keyword's conversion column gets its slice. Add up enough slices across a month and you get numbers like 2.37.
(Those percentages are an example to show the mechanics. Google calculates real ones from your own account, and they change from one sale to the next.)
There is one more wrinkle that confuses people comparing reports. Google files a conversion under the date of the click, not the date of the sale. Its reasoning is sensible: your cost is recorded on the day of the click, so the result should sit next to it. But it means the slices of one sale can land in different weeks. The first click and the last one might sit in different reporting periods, and the week you ran the job might show fewer conversions than the week the customer started looking.
None of this is an error. It is the account telling you, more honestly than it used to, that most purchases worth advertising take more than one search.
Google Hands Out Credit by Guessing What Would Have Happened Without the Click
So how does Google decide that the first search was worth 0.37 and not 0.10?
It does not watch the customer and know. It estimates. Google's own description is that the model compares the paths of people who converted with the paths of people who didn't, and looks for the ad interactions that show up more often in the paths that ended in a sale. If people who clicked your "replacement cost" ad early on go on to buy noticeably more often than similar people who never saw it, that keyword earns real credit. If a click shows up just as often in paths that went nowhere, it earns little.
The useful way to think about it: for each click, the model is asking how much less likely the sale would have been without it. A click that moved the odds a lot gets a big slice. A click that was along for the ride gets a small one.
That makes it a far better guess than handing everything to the last click, and it is still a guess. Google sees its own ads. It does not see the neighbor who recommended you or the truck the customer drove past. It divides credit among the things it can see.
It is also no longer optional in any real sense. Google Ads used to offer a menu: first click, linear, time decay, position based. All four are gone. Any conversion action that used one was moved to data-driven automatically. What remains is data-driven, which is the default for most conversion actions, and last click. If you have never opened your attribution settings, you are almost certainly on data-driven already, and that is why the decimals appeared.
The model covers more than search. Google applies it across Search (including Shopping), YouTube, Display and Demand Gen. So a YouTube view or a Display click can take a slice of a sale that closed on a search ad, and the search ad's number shrinks to make room.
Last Click Gave Your Brand Name Credit for Customers It Never Found
Go back to the water heater customer. Under last click, the keyword that got the whole sale was your company name.
Over a year, that makes your name look like the best keyword you own: cheap clicks, lots of conversions, a cost per conversion nothing else can touch. The generic keywords that introduced you look expensive and weak, because the customers they found were credited to someone else when they came back.
That is how an owner ends up cutting the wrong half of the account. The budget moves to the brand campaign, and for a few weeks nothing changes, because people who found you earlier are still in the pipeline. Then the brand searches thin out, because nobody new is learning your name.
Data-driven attribution moves the credit back up the path. When an account switches, generic keywords suddenly look better, because they finally get a slice of the sales they started. Brand keywords look worse, because they now have to share.
Neither campaign changed. The only thing that moved was the credit.
That creates the opposite mistake. When the brand campaign's numbers drop after a switch, some owners cut its bids, reasoning that it was never as good as it looked. Be careful. Someone searching your exact business name is about as close to buying as a person gets, and if a competitor bids on your name, the ad you skip is a customer you already paid to bring back walking into someone else's office. A smaller slice of credit does not make that click less important.
Smart Bidding Spends Real Dollars on Fractions of a Sale
If the decimals only changed the report, you could ignore them. They don't stop at the report.
If you use target CPA, target ROAS, or Maximize conversions, the bidding reads the same conversion column you do. Under data-driven attribution, a keyword that tends to start sales carries real conversion credit, so the bidding treats it as valuable and is willing to pay to show up for it. Under last click, the same keyword carries close to zero, so the bidding pulls back from it and pushes money toward whatever tends to close.
So the attribution model is not a reporting preference. It is an instruction about which kinds of searches are worth money. We made the same argument about targets in Your Google Ads Bid Target Isn't a Wish. It's an Instruction., and the model sits underneath the target. The target says how much a conversion is worth. The model decides which clicks get counted as part of one.
So changing the model changes what you buy. Switch from data-driven to last click and the bidding gradually shifts budget toward closing searches and away from the searches where people first find you. That might be what you want. It should never happen by accident, or because a report looked nicer one way. Before changing anything, add the "current model" conversion columns, which restate your past numbers under the model you are on, and compare them with the regular columns.
And the fractions are only as good as what they are fractions of. If your conversion action counts newsletter signups, accidental phone taps and duplicate form fills, data-driven attribution will carefully divide credit for things that were never sales. Fix what you count first. We walked through that cleanup in Your Conversion Number Is Going Up. Your Sales Aren't.
Google's Attribution Model Asks for 200 Conversions a Month. Most Small Accounts Have Far Fewer.
Here is the part most explanations of data-driven attribution skip, and it matters most to the businesses reading this.
Google's help page recommends at least 200 conversions and 2,000 ad interactions within 30 days for the model to work well. A local contractor, a dental office, or a law firm running a few thousand dollars a month might see 20 or 40 conversions in that time. The same page says the model will still function with less data, and it does. What it means in practice is that the model has much less of your own history to learn from, so its splits lean on patterns Google has seen elsewhere rather than on what your customers actually did.
Some practitioners take this as a reason for small accounts to switch back to last click. We don't think that is automatic, and the reason is in your own data.
Open the Path metrics report. It shows how many ad interactions people typically have before they convert and how many days it takes. In a lot of small local accounts, most conversions come after a single click: someone searches, clicks, calls. When there is only one click on the path, there is nothing to split. Data-driven and last click give the same answer, and the decimals barely appear.
Then open the Model comparison report and compare the two models for your campaigns and keywords. One of two things will be true.
The difference is small. Your customers mostly buy on the first click, the two models agree, and the choice barely matters. Leave the default alone and spend your attention elsewhere.
The difference is large. Your customers take several searches to decide, and the two models disagree about which keywords earn their keep. That disagreement is the most useful thing in your account. It tells you that some of your keywords are doing work that last click never showed you. Switching to last click to make the numbers simpler would mean choosing not to see it.
Research Keywords Look Worst on Cost per Conversion Because They Start the Sale
This is where the decimals cost real money.
Every few months, most accounts get the same cleanup: sort the keywords by cost per conversion and pause the worst. For most keywords that is right. A keyword that has spent real money for months and produced nothing deserves to go.
The trouble is the keywords at the start of a purchase. Searches like "how much does a new roof cost" are made by people who are researching, not ready to call today. Their own cost per conversion is usually ugly, because the person rarely converts on that click. They come back later through a more specific search or your name, and that later keyword gets most of the credit.
Pause the early keyword and its row looks like a saving. A few weeks later the specific keywords and your brand searches start to sag, and nothing in their rows explains why. The people who used to find you early are now finding someone else early.
Before you pause anything on its cost per conversion, run it through a short check:
- Look at its assists, not just its conversions. The assisted conversions report shows how often a keyword appeared earlier in the path of a sale that finished somewhere else. A keyword with a poor cost per conversion and a lot of assists is a door, not a leak.
- Check the top paths. If the keyword keeps showing up as the first step in paths that end in your name, it is how people are finding you.
- Read it over 30, 60 and 90 days. A small account's monthly numbers swing hard on one or two sales. A keyword that looks dead this month can look fine over a quarter.
- Fix before you cut. Check its search terms for junk you can add as negatives, and check whether the page it sends people to answers the question they searched. A research search sent to a page that only says "call now" will always look bad.
- If you do pause it, pause one keyword at a time and watch the campaign's total. Don't judge the decision by the paused keyword's row, which will obviously show zero. Judge it by whether total conversions hold up over the following month. If they drop, turn it back on. Pausing is reversible, which is exactly why it is worth doing as a test instead of a purge.
The Assisted Conversions Report Shows What Your Keyword Column Hides
Everything above lives in one place, and almost nobody opens it. In Google Ads, go to Goals, then Attribution. Twenty minutes with the four reports there will tell you more about how customers find you than a year of staring at the keywords tab.
Conversion paths lists the most common sequences of ad interactions before a conversion. If a generic search followed by your name is one of the top paths, you are looking at the brand-credit problem in your own data.
Path metrics shows how many interactions and how many days a conversion usually takes, which tells you how much the model choice matters for your business at all.
The assisted conversions report shows how often each campaign, ad group or keyword appeared earlier in a path that converted somewhere else. Sort by assists and hold it up against your pause list. It covers YouTube and Display as well as Search, so it will also show whether those are helping sales that close on a search ad.
Model comparison puts data-driven and last click side by side. Switch it to keyword level and look for the big gaps. The keywords that gain the most under data-driven are your starters. The ones that lose the most are your closers.
One more thing if you ever compare Google Ads with Google Analytics or your own sales records. The standard conversion columns use the click date, but Google also offers columns reported by the date the conversion happened, which line up far better with a sales log. If your numbers never match across tools, that is often part of the reason, along with the counting differences in GA4 Says 40 Conversions. Google Ads Says 61. Both Are Right.
Switching to Last Click to Make the Report Look Better Changes Which Keywords You Fund
The temptation with attribution is to pick whichever model makes the account look the way you want. Last click makes the brand campaign look brilliant. Data-driven makes the generic campaigns look like they earn their keep. Both are tempting for different reasons, and choosing either one for the report is choosing it for the wrong reason, because the model doesn't just describe the account. It steers the bidding.
Pick the model for the decision you are trying to make. For most small accounts that means leaving data-driven in place, checking Path metrics and Model comparison once a quarter, and never pausing a keyword without first looking at what it assisted.
And keep the whole thing in proportion. No attribution model knows for certain which ad made the sale. Google is dividing credit among the clicks it can see, inside its own platform, using its own estimates. We wrote about the limits of every platform's view of the customer in You Will Never Know Exactly Which Ad Made the Sale. The decimals are a better guess than last click ever was. They are still a guess, and the bank account is still the final answer.
If you would rather have someone read these reports for you every month and make the pause-or-keep calls, that is what we do for small businesses on Google Ads. And if you would rather learn to read them yourself, the four reports above are the place to start. Either way, the next time you see 2.37, you will know it is three keywords sharing one sale, and you will know to ask which of them found the customer.
