Rule or script
The weekly performance report across Google Ads, Meta and the online store
Clean platform data and known formulas: a connector and a dashboard build it every Monday. No model needed.
The EDY method
How we evaluate every AI use case.
Four steps, every engagement: the same method whether we are auditing a whole marketing budget or building one automation. The “AI or not?” scorecard sits at the centre.
-
01
Measure
List every AI and marketing-tech cost (seats, add-ons, usage bills, pilots, agencies) next to the metric it should move.
Output: the spend map
-
02
Decide
Run every line and every new idea through the scorecard. Each gets one verdict: cut, keep, fix, or build another way.
Output: a verdict for every line
-
03
Build
Build the lean version (rule, script, workflow or AI), with its monthly run cost stated before work starts.
Output: the lean version and its run cost
-
04
Govern
Review spend, renewals and model prices every quarter, so the savings hold.
Output: a quarterly review
The scorecard
Six questions decide whether AI earns its cost.
Each task gets 1 to 5 on six questions. Two set the horizontal position, two the vertical, and two can overrule the result.
| Question | Points toward AI when… | Points away from AI when… |
|---|---|---|
| How often does it run? | Thousands of times a month | A few times a month |
| What is each one worth? | A qualified lead, a sale, a budget decision | Minutes of someone's time |
| What does the input look like? | Free text, images, video, voice | Platform data, CRM fields, product feeds |
| Does it need judgment? | Brand, audience and context change the answer | The rule is known |
| What does a mistake cost? | Little, and it's easy to catch | A lot, or it goes out in public |
| What does the simple version cost? | Nothing simple works | A platform rule, template or script already does it |
Four verdicts
AI runs on every case only where judgment meets volume.
How often a task runs and what each case is worth set the horizontal position. Messy input and judgment set the vertical. If a simple version already works, start there. If mistakes are expensive or go out in public, AI assists a person instead of deciding.
Rule or scriptExisting toolCustom AIDon't do it
Examples
Four marketing AI use cases, scored.
Illustrations of the method, not client cases.
Existing tool
Writing about 2,000 ad copy variations a month for Search and Meta
Volume and creative judgment point to AI, but a wrong price or claim goes out in public: use the generator already inside the ad platforms, with a person approving.
Custom AI
Sorting about 10,000 WhatsApp and chat inquiries a month by intent
Volume, free text and leads worth real money: AI earns its cost here, priced per conversation before you switch it on.
Don't do it
An AI chatbot for the 15 questions a month the website gets
Fifteen questions with known answers: a FAQ page and a person handle them. The license costs more than the problem.
Before you sign your 2027 AI renewals, check what's paying off.
Thirty minutes, no slides: what you pay for AI today, what it returns and what's up for renewal.