Audit Google Ads against your sales

The account in this guide: an Australian education provider I audit monthly, inside someone else's business. Leads come from a form or a phone call, and the sale is an enrolment weeks later. I can't name it, but the numbers are real.

01 Reconcile the numbers before you read them

Start by checking that the spend in your export matches the account total for the same dates. If it doesn't, stop. A missing campaign or a time zone mismatch will change every conclusion after it.

Use complete days only. Today's half-finished row makes yesterday look like a cliff.

Then check how fresh the sales data is. On this account the CRM export was 13 days behind the end of the audit window. Most recent campaigns showed no sales at all, because the sales hadn't been loaded yet. The audit went out marked provisional, with budget changes on hold until the export was refreshed. The rule I use now: if the sales data is more than a week behind, the audit can describe, but it can't decide.

02 Connect clicks to sales, and measure the gap

The join is the click ID. With auto-tagging on, Google adds one to every ad click (the GCLID). A hidden field on your form captures it, your CRM stores it against the lead, and when the lead becomes a sale you can trace the sale back to the campaign, keyword and search that started it.

Then measure coverage: what share of sales you can trace to a click. On this account it was about three in four. Most of the rest were phone enrolments and clicks too old to match. Write that number at the top of the audit. It tells you how far to trust everything below it, and which campaigns are under-credited. Campaigns that drive calls always look worse than they are.

If the CRM has no click ID at all, that's your first finding. Fix it before anything else, because nothing else in the audit can be checked against sales until you do.

03 Score every campaign on cost per sale

For each campaign: spend, sales, revenue, cost per sale and return on ad spend. Rank by spend, because that's where a wrong call costs most.

Use a mature window. Leads take weeks to become sales, so a recent window understates everything. I use six months for verdicts and the last month for direction. Then give each campaign one of three flags: below target, protect, or too new to judge. "Too new" is a verdict in its own right, and it means you look again next month.

On this account one course took two thirds of the spend. Its search campaigns returned about 3.3 times their cost. Its Performance Max campaign (Google's automated campaign type that runs across search, YouTube and display) returned nearly 10 times. Some of that was probably brand searches it would have won anyway, which is why the checklist asks for brand exclusions before you scale one. Even discounted, it was the best thing that course had, and the search campaigns were where the money leaked.

Two search campaigns with click-through rates of 11 to 15% sat near the bottom. People clicked the ads and then didn't buy, which points to the landing page or lead quality. More budget would only make that problem bigger.

04 Verify every cut against sales

Most audits skip this step.

Search terms containing "free" looked like tyre kickers. In fact they cost less per enquiry than the account average, and searches for free or government funded courses produced about 15% of the traceable enrolments in 90 days. A blanket negative keyword would have cut about one traceable sale in seven. I recommended routing those searches instead of excluding them, with their own ad group and a landing page that answers the funding question.

Disability searches were the other near miss. The first draft of the negative keyword list, which AI wrote, marked them as waste because the provider doesn't sell a disability course. They weren't waste. In Australia, disability support work sits under community services qualifications, and some of those searchers were enrolling in the neighbouring courses. I caught it because I know the sector, and the model doesn't.

The rule: never pause a keyword or add a negative without checking it against sales. A keyword with sales gets judged on its cost per sale.

05 Retarget bidding, then tune

A conversion action is anything you tell Google to count: a form, a call, a sale. The one set as primary is what Smart Bidding (Google's automated bidding) chases. If it's the form fill, the account is buying cheap enquiries. On this account about one enquiry in nine could be traced to a sale, so most of what bidding chased never became one.

Then look at conversion values. At the time of the audit, every conversion Google had ever recorded on the account was valued, in total, at less than 0.04% of the revenue from the previous 60 days alone, so Smart Bidding had almost no sale value to aim at.

The fix, in order:

1. Send each sale back to Google with its value, its click ID and the lead's email or phone, which Google hashes and matches. Google calls this enhanced conversions for leads, and it replaces the older import that used the click ID alone.
2. Make the sale the primary conversion once each campaign gets enough sales to learn from, roughly 15 or more a month. Until then, bid on the closest stage that has the volume, such as a qualified lead, and carry the sale value with it. Run a new conversion alongside the old one for at least four weeks before you switch.
3. Track phone calls as conversions, and send back which calls became sales, so call-driven campaigns stop looking worse than they are.
4. Move to value-based bidding once there's enough volume for it.

Only then tune structure, keywords and budget, with the smallest reversible change first. Move budget in steps of about 20%, so you can still tell what caused what.

Look past the account

The biggest leak on this account wasn't in Google Ads. About one lead in seven had sat uncontacted for more than two weeks, and "uncontactable" was the most common reason for a lost sale. An audit that stopped at the ad account would have recommended more spend into a funnel that wasn't calling people back. Ask for the lead report by status before you write a single recommendation.

Where AI fits

AI is good at the grunt work: grouping thousands of search terms into themes, rolling up repeated words across them, drafting the audit from the numbers, and spotting the campaign whose spend had doubled since the last review. It is bad at knowing that a disability search can become a community services enrolment.

On the monthly audit I run, a script pulls and reconciles the numbers, a model reads them and drafts, and I make every call that removes traffic. The disability near miss is why that last part isn't negotiable.