Case Notes from Sydney Mid-Market Pilots: 5 Experiments That Worked (and 3 That Didn’t)

These are anonymised Sydney mid-market case notes from 90-day AI marketing pilots.
Every experiment had a clean baseline, a holdout, and board-grade KPIs.
I’m sharing the hypothesis, what we changed, how we measured, and the result.
Use what worked.
Dodge what didn’t.

‍

What “good” looked like in every pilot

We locked a 28–56 day baseline.
We defined ROI = (Incremental Gross Profit – Total Pilot Cost) ÷ Total Pilot Cost.
We used a geo or audience holdout to prove incrementality.
We reported one board page: Revenue/Marketing $, CAC, Payback, Non-brand organic, SQLs, Unsub SLA.

1) Answer Packs for AEO + SEO (Worked)

Who. B2B SaaS, ~$40M revenue, North Sydney.
Hypothesis. Structured “Answer Packs” will win AI Overviews, snippets, and SQLs.
Change. 1 hub + 8 spokes with definition → steps → small table → FAQ → JSON-LD.
Design. Topic-level holdout; no paid support; weekly crawl budget fix.
Result. +38% non-brand organic, +24% SQLs, CAC ↓12%, Payback ↓1.2 months.
ROI. Incremental GP ≈ $176k on $40k pilot cost → 3.4×.
Why it worked. Machines could quote us.
Copy this. Ship compact answers with schema and the same facts across web + PDF.

2) Reactivation With Consent-Tight Journeys (Worked)

Who. Multi-brand retail, ~$70M, Inner West.
Hypothesis. First-party “reason to return” sequences will revive cold buyers without spam risk.
Change. Two-step email + SMS fallback, incentive ladder, product-agnostic tips, zero dark patterns.
Design. Audience holdout by last-purchase month; five-day unsubscribe SLA tracked.
Result. Revenue/Recipient ↑41%, complaint rate flat, deliverability ↑.
ROI. Incremental GP ≈ $75k on $25k cost → 2.0×.
Why it worked. Value first, frequency capped, consent clean.
Copy this. Treat unsub SLA as a product requirement, not a legal footnote.

3) Transparent Pricing Calculator + Zero-Party Fields (Worked)

Who. Services, ~$55M, Parramatta.
Hypothesis. A public calculator will lift qualified form fills and SQL win rate.
Change. Calculator with 5 inputs, explainer table, and email me-the-estimate option.
Design. Split-URL test; sales blinded to form source.
Result. MQL→SQL CVR ↑32%, AOV ↑6%, sales cycle ↓9 days.
ROI. Incremental GP ≈ $90k on $30k cost → 2.0×.
Why it worked. Expectations set before the call.
Copy this. Show the math and send the estimate PDF with the same numbers.

4) Field-to-Feed: Event Notes → 24-Hour Publish (Worked)

Who. Fintech, ~$30M, CBD.
Hypothesis. 24-hour event recaps in Answer-Pack format will boost branded search and assisted revenue.
Change. Pre-templated outlines, public quotes with citations, one small benchmark table.
Design. Suburb-level holdout; no ads; LinkedIn distribution only.
Result. Branded search ↑19%, assisted revenue +$46k in 6 weeks.
ROI. $46k GP on $18k cost → 1.56×.
Why it worked. Speed + entity consistency across web + PDF + author pages.
Copy this. Publish fast, refine at 72 hours, keep facts identical everywhere.

5) Speed-First CRO (Worked)

Who. D2C, ~$25M, Northern Beaches.
Hypothesis. Sub-2s LCP and friction fixes will lift checkout CVR more than new creative.
Change. Image policy, inline validation, Apple/Google Pay, “save cart” email.
Design. 50/50 split on top 5 templates; same promo calendar.
Result. Checkout CVR ↑14%, bounce ↓11%, Revenue/Session ↑9%.
ROI. Incremental GP ≈ $80k on $22k cost → 2.64×.
Why it worked. Latency kills intent.
Copy this. Fix speed before you brief a new headline.

6) Over-Personalised Banners From Third-Party Data (Didn’t)

Who. Marketplace, ~$60M, Alexandria.
Hypothesis. Deep persona banners will raise CVR on landing pages.
Change. Third-party segments, 12 creative variants, aggressive rotation.
Design. 50/50 split on top 10 landing pages.
Result. CVR flat, unsubscribes +28%, complaints +0.2pp.
ROI. Incremental GP ≈ $5k on $20k cost → -75%.
Why it failed. Creepy + mismatched intent + legal risk.
Do instead. Context and cohorts, not covert profiles.

7) “Let the Bot Rewrite Everything” (Didn’t)

Who. B2B services, ~$45M, Macquarie Park.
Hypothesis. Full AI rewrite will 10× content at same quality.
Change. Bot-only drafts with minimal human edit.
Design. 1:1 replacement across 30 URLs; same internal links.
Result. Time-on-page ↓23%, snippet losses, backlinks stalled.
ROI. GP ≈ $12k on $28k cost → -57%.
Why it failed. Voice drift, shallow claims, no unique proof.
Do instead. Human taste + AI throughput, with citations and method notes.

8) Social Volume Without Distribution Architecture (Didn’t)

Who. Hospitality, ~$20M, Inner City.
Hypothesis. 5× more posts = 5× more bookings.
Change. Daily posts, no cluster plan, no UTMs, no owned hub.
Design. City-pair holdout (Sydney vs Melbourne brand).
Result. Impressions ↑, bookings flat, team burnout, no attribution.
ROI. GP ≈ $10k on $15k cost → -33%.
Why it failed. No entity strategy, no answer assets, no analytics spine.
Do instead. Publish to a hub first, then syndicate with UTMs.

Patterns I now trust

Structured answers beat long prose.
Consent-tight lifecycle beats blast.
Calculators with transparent math change sales calls.
Speed beats clever on conversion.
Creepy personalisation backfires fast.

Your 30-60-90 if you want these wins

Days 1–30. Lock baseline and holdout.
Ship one Answer-Pack hub and one CRO fix.
Stand up consent ledger and unsub QA.
Days 31–60. Publish 3–5 spokes.
Launch the reactivation journey.
Add a pricing calculator.
Days 61–90. Run Field-to-Feed on two events.
Kill underperformers.
Present the ROI pack and scale only what pays.

What to ask your agency this week

Show me one Answer Pack you shipped with schema.
Export CAC, payback, SQLs, unsub logs now.
Propose a clean holdout.
Quote a fixed 90-day price.
Name the human owner for every agent.

Conclusion

AI marketing in Sydney compounds when you ship structured answers, respect consent, and measure like a CFO.
Copy the five winners.
Avoid the three traps.
Make your next 90 days about ROI, not hype—and let the board see it on one page.
Book a demo at https://hoook.io to see how our customers getting up to 100% traffic growth and up to 20% revenue increase.

‍