Case studies · The working machine
The AI marketing ops build.
Three case studies from the marketing machine I run as CMO of a multi-brand AI company — and a fourth, run on this site, because an audit I sell should be an audit I take.
01 · HubSpot lifecycle & scoring · 3 weeks · September 2026
Three lead scores into one.
The portal had three lead scores — one per market — each with its own workflows. Three scores meant three definitions of a good lead, and sales had no single number to act on. I folded them into one fit-plus-engagement model across about 1,370 contacts, two sales pipelines, and one handoff to sales.
What I built
- One universal score: fit becomes a letter (A, B, C), engagement a digit (1, 2, 3). An A1 is a great-fit buyer leaning in; a C1 is very engaged and probably not a buyer. A single number would have blurred the two.
- A workflow turns the grade into Hot, Warm, or Cold — a rep doesn’t need to know the scoring model to know who to call.
- Lead→MQL→SQL on score, with guardrails: an A1 can skip straight to sales, but only with the rep notified the moment it happens. Leads that aren’t ready recycle into nurture instead of disappearing.
- A closed-won deal opens a pilot onboarding ticket automatically, with a mid-pilot review scheduled 60 days after the pilot goes live.
Every contact now carries the same score, and the scoring logic is one page instead of three. It’s too early for conversion rates — and I won’t publish a number that a few leads could swing. Every workflow was built switched off and turned on with sales in the room.
02 · AI search visibility (AEO) · tracking daily since September 24, 2026
AEO measured against the buyer.
Buyers increasingly ask an AI assistant before they ask Google. The first setup measured the company, not the buyer — prompts for a product we’d stopped leading with, and nothing separating two brands that sell to completely different buyers. I rebuilt it from the buyer’s side: one profile per brand, 25 prompts each earning its slot — 21 unbranded questions a buyer asks before they know us, four branded health checks — mapped to the buying journey and wired into daily reporting.
The baseline, published honestly
100%
Branded visibility — ask about us by name and every assistant knows who we are
0%
Unbranded visibility — the category questions, and the number the work has to move
2.3%
Owned citations — 160 of 6,905 sources the assistants read are our pages
825
AI answers tracked in the first 30-day window, across two brands
30-day window ending October 4, 2026 · HubSpot AEO, tracking ChatGPT, Gemini, and Perplexity.
0% is a baseline, not a failure. Branded visibility at 100% feels good and means almost nothing about pipeline — the unbranded number is the one that pays. One finding no dashboard would have shown: an assistant kept confusing the brand with a look-alike domain, cited 24 times in the same window.
03 · AI-agent reporting · live since September 16, 2026
The daily marketing pulse.
Three brands shared one HubSpot portal, named however each person liked, so no report could say which campaign produced what. The naming convention came first — Year-Brand-Type-Name, doubling as the UTM tag. Then an AI agent that reads HubSpot every weekday afternoon and posts one Slack message ending with the one thing worth looking at today, plus a Monday narrative report with fixed health thresholds.
What it caught in its first twelve reports
- A “spike” of 27 new contacts that was an attribution gap, not demand — 26 had no campaign source.
- Nine MQLs arriving within the same ten seconds, flagged as a bulk change dressed up as pipeline.
- A newsletter that was read and never clicked — the two recorded clicks were bots. The fix was the call to action, not the list.
- An AI assistant describing a different company with a similar name — an identity problem in AI search no dashboard would have surfaced.
Most days the pulse says “quiet day.” That’s useful too — flat by default is different from flat because something broke. The agent reads, flags, and drafts. People decide.
04 · Self-audit · this site · October 2026
Practice what you sell: this site, audited.
Before taking the AI Visibility Audit to clients, I ran its first step on this site. Same method as the offer: start from the questions a buyer would ask, check what the assistants’ sources can actually find, and publish the baseline honestly — zeros included.
What the audit found
- Search my name plus “marketing” and the top results are an outdated author bio with my old title — and several other people who share the name. Nothing connected this site to my current CMO work.
- Search “Emily L. Snyder” plus “fractional CMO” and nothing about me comes back at all — the buyer’s query is an empty room.
- This site wasn’t yet indexed: a new build on a rented subdomain is invisible to the sources AI assistants cite. Owned citations: zero.
- An older, half-finished version of this site was still live on another subdomain — generic page title, old job title — competing with me for my own name.
The 90-day fix list, in progress
- One name everywhere — Emily L. Snyder — across this site, LinkedIn, and author bios, with the full-time CMO role stated publicly.
- A custom domain and branded email, so citations accrue to an address I own. Prompt tracking starts the day it switches — measuring a subdomain that’s about to change would be theater.
- Publish the machine: the three case studies above exist because assistants can only cite what’s on the record.
- Retire the old site and update the stale third-party bios.
The honest baseline for this site is mostly zeros — which is exactly where a first audit usually lands. This page is the “before.” The offer is getting a client to the “after” faster than they would alone.
