AI systems and infrastructure

Building AI into the work,
not around it.

Every system here started with a specific GTM problem. None of them required an engineering team. All of them are running in production. This is what it looks like to operationalise AI inside a revenue marketing function, not as a side project, but as the way the work gets done.

Most conversations about AI in marketing focus on individual productivity: faster drafts, quicker research, less time on email. That's useful. The bigger opportunity is different, and it's the one I'm focused on.

When AI is designed well and connected to the right data, it doesn't just help one person move faster. It gives an entire sales team a clearer picture of where to focus, makes personalised outreach possible at a scale that previously required a much bigger team, and converts hours of manual analysis into a 15-minute workflow anyone can run. That's the shift from tool to infrastructure. That's what I've been building.

01

APAC insights hub

The problem

17+ account executives across five APAC markets, each managing large account books, with no consistent weekly signal on what to prioritise. Intelligence was manually assembled, inconsistent across reps, and depended entirely on one person pulling it together. When that person was offline, nothing went out.

What I built

An automated account intelligence system that pulls intent signals from Demandbase and Salesforce, scores accounts by likelihood and timing, and sends each AE a personalised brief every Monday. MQAs ranked by pipeline score, closed-lost reactivations, new contacts showing buying intent, and accounts with recent engagement. Regional leads get roll-up views. The system runs without me in the loop.

What changed

Rolled out across ANZ, then expanded to Japan, GCR, India, and SEA. The APAC SDR team ranked it their top tool for weekly account prioritisation. EMEA started building the same model independently, without being asked.

"Extremely useful" was the team's word for the refresh. When sales tells you they'd notice if it stopped arriving, the system has earned its place.
CursorDemandbaseSalesforceAccount intelligenceSales enablement
02

AI readiness audit workflow

The problem

Enterprise outreach for a major Shopify event needed to be genuinely personalised to each merchant's situation. A template wouldn't land. Doing it properly required reading each merchant's audit results, understanding their specific gaps, and writing copy that connected to their commercial context. At scale, that was weeks of work.

What I built

A workflow using Cursor and connected data sources that generates 18 personalised merchant-facing assets from a single audit dataset. Each flyer opens on a different narrative, mapped to that merchant's score, pain points, and stage. LinkedIn DM copy and seller call scripts included, each specific enough that the merchant knows you read the brief. What would have taken two to three weeks took two days.

What changed

Personalised outreach that scaled across the full account list ahead of the event, and drove some of the highest-quality conversations on the day.

The test for this kind of work is whether a merchant reads it and thinks it was written for them. The meeting bookings suggest they did.
CursorStandard LLMsPersonalised outreachEnterprise GTM
How I think about it

Six things I've learned about building AI that actually gets used.

The gap between AI that's impressive in a demo and AI that a sales team trusts and uses every week is wider than most people expect. Here's what I've found actually matters.

01

Start with a real problem

Not "how do I use AI?" but "what recurring workflow is slow, manual, or inconsistent and costs the team something meaningful?" The tool comes second. The problem comes first.

02

Connect it to data the team already trusts

Generic AI output doesn't earn seller trust. Every system I've built pulls from Salesforce, Demandbase, or internal data sources, things the team already believes. The AI is the engine. The data is what makes it credible.

03

Structure the workflow before you automate it

If I can't describe the inputs, outputs, owners, and review points on paper first, I'm not ready to build it. Automation makes a bad process faster. It doesn't make it good.

04

Keep humans in the loop where it matters

AI accelerates synthesis and speeds up research. Final judgment on customer sentiment, account strategy, and commercial decisions stays with a person. That's not a limitation. That's the design.

05

Make the output seller-ready

Success isn't "AI produced an answer." It's "a seller read it, trusted it, and acted on it." If the output requires interpretation or cleaning before it's useful, the workflow isn't finished.

06

Package it for reuse

If I've done something more than twice, it becomes a workflow, a template, or a dashboard. The goal is infrastructure that works whether I'm there or not. If it only runs because I'm maintaining it, it's not done yet.

By the numbers

Hundreds
Accounts covered by the APAC Insights Hub
18
Personalised merchant assets generated for one campaign workflow
5
APAC markets covered by the weekly signal system
17+
Account executives receiving personalised weekly briefs
0
Engineering dependencies across all systems built

AI usage across Cursor, Gumloop, and standard LLMs sits materially above median across the marketing org. Exact figures pending verification from internal dashboards.

Want to talk about any of this?

Happy to walk through how the systems work or what I'd build next.