01 the business
Online homeware retailer
25 staff, $4M yearly sales
The retailer advertises on Google, Meta and LinkedIn. Each platform reports its own conversions and they never agree with the shop's orders. Someone spends Monday copying numbers into a spreadsheet, and budget decisions for the week are made before it is finished.
02 before
The problem
Each ad platform reports its own version of the truth, and they rarely agree. Someone logs into four dashboards, copies figures into a spreadsheet and tries to work out which spend brought in real customers. By the time the report is ready, the budget for the next week has already been set.
- Four dashboards and a spreadsheet, rebuilt every week.
- Platforms claim more conversions combined than the shop actually received.
- Weak ad sets keep running for weeks because nobody spots them in time.
03 the build
What the AI does
- Step 1: Collects spend, clicks and conversions from Google Ads, Meta Ads and LinkedIn Ads on a schedule.
- Step 2: Matches those figures to actual orders in Shopify or new leads and deals in HubSpot, using your tracking tags.
- Step 3: Builds one consistent table by campaign and channel and refreshes a Looker Studio report.
- Step 4: Writes a short commentary on what went up, what went down and the likely reasons, citing the numbers it used.
- Step 5: Posts the summary to Slack with suggested changes, such as pausing a weak ad set or moving budget to a better one.
paused: waiting for a person to approve
How it was put together
- Pulled spend, clicks and conversions from each ad platform on a schedule.
- Matched them to real orders using the shop's tracking tags and UTM parameters.
- Built one consistent table by campaign and channel and refreshed a Looker Studio report.
- Wrote a short commentary on what moved and why, citing the figures it used.
04 human in the loop
Where a person stays in charge
The report only reads data and never changes campaigns. The marketing lead decides which suggestions to act on and makes any budget changes in the ad platforms themselves.
- The system only reads data and never changes campaigns or budgets.
- Suggestions are labelled as suggestions, with the numbers behind them.
- The marketing lead decides and makes any changes in the ad platforms.
05 rollout
From kickoff to live in 3 weeks
- Week 1Audit tracking tags and agree one definition of a conversion.
- Week 2Build the joined table and check it against last month's orders.
- Week 3Start the Monday summary and retire the manual spreadsheet.
06 after
What changes
Everyone sees the same numbers every week without a manual export. Budget decisions are based on customers and revenue, not just clicks, and the team acts on results while a campaign is still running.
Assume 5 hours a week on manual reporting and a $20,000 monthly ad budget, with 10% of it in ad sets that bring no orders.
5 hours × 52 weeks = 260 hours a year of reporting. $20,000 × 10% = $2,000 a month at risk; catching it two weeks sooner saves about $1,000 a month.
What to measure
- Hours spent on reporting each week.
- Cost per real order by channel.
- Days between a campaign going weak and it being changed.
- Gap between platform-reported and shop-recorded conversions.
07 evidence
What the research says
About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.
Gartner research put the average cost of poor data quality at $12.9 million a year per organisation.
Gartner, Data quality: why it matters and how to achieve it (opens in a new tab) · 2020
08 lessons
Mistakes to avoid
- Trusting platform conversions. Anchor everything to orders or deals in your own system.
- Changing the definition of a conversion mid-quarter.
- Reporting without a decision. Each summary should end with what to do next.
09 questions
Common questions
- Does it work for lead generation instead of ecommerce?
- Yes. It matches spend to leads and closed deals in your CRM instead of shop orders.
- What if our tracking is broken?
- The first week is spent fixing it. The report is only as honest as the tags behind it.