RAG over voice of customer
The Listening Post
Ask questions of scattered customer feedback in plain language and get an answer grounded in the actual verbatims, cited back to source.
Open the labAI systems for customer intelligence & operations
AI systems that help your customers, and turn every customer conversation into insights you can actually use.
Try the Listening Post now
A working system, not a mock-up. It searches 513 pieces of customer feedback from a fictional B2B SaaS company and cites every claim back to the verbatim it came from.
01
Labs
CX Matters Labs
Working demos showcasing how AI can turn customer-experience data into decisions.
Each one is a real, functioning demo, trained on a synthetic dataset.
MCP server for voice of customer
Signal Line
Puts this dataset's tools and data behind the Model Context Protocol, so an AI assistant can query it and act on it directly, inside the tools people already use.
Open the labAutonomous agent for C-suite briefing
The Monday Brief
A scheduled agent investigates at-risk accounts on its own initiative and surfaces what nobody thought to ask about, before anyone has to ask.
Open the labVISUALLY EXPLORE CLUSTERED FEEDBACK
The Observatory
Every piece of customer feedback grouped into clusters and summarised, sized by volume and positioned by meaning, coloured by sentiment, theme or segment, so patterns become visible at a glance.
Open the labAgent evals for safe deployment
The Proving Ground
Simulated customers with secret briefs put a support agent through the same six hard conversations, one prompt change at a time — catching the silent failures that never reach a support queue.
Open the lab
02
The moment of truth
Marketing builds the reputation. Every conversation is where it's earned or lost.
Marketing makes a promise. Product sets an expectation. Every customer conversation — a sales question, a stalled onboarding, a support ticket at 2am, is where a real person, with a real problem, finds out if what they were promised is true.
What's changing now
Customer conversations used to be split across departments (sales, support, success), each team with its own tools and its own tickets. AI is pulling that together into one ongoing conversation across the whole customer relationship. A single agent remembers the customer and moves between departments as needed.
That's an enormous opportunity and a quiet risk. Done well, it absorbs the repetitive volume so your people are free for the moments that actually move growth and retention. Done badly, it just turns customer conversations into AI-powered deflection — cheaper and worse — and you lose the richest signal you own.
The companies that win the next few years won't be the ones who deployed AI fastest. They'll be the ones who used it to understand their customers better: to build better products, deliver more value, and prove it in the numbers.
03
One customer comment. Three things at once.
Humans. Data. Revenue.
Every customer conversation is three things at once, and most companies capture none of them.
"I've been trying to export my report for an hour and nothing works. About to give up and find another tool."
A human moment
Someone stuck and about to walk, forming the memory they will repeat to other people.
A data point
The export flow is failing. How many others hit this exact step and said nothing?
A revenue event
"Find another tool" is churn forming in real time, and it is still reversible.
Humans.
It's someone at a real moment — frustrated, doubtful, or just deciding if they're making the right call — forming the memory they'll repeat to others. That memory becomes word of mouth — and word of mouth is the cheapest marketing you'll ever get, or the most expensive you'll never see.
Data.
Your teams sit on the richest unstructured dataset in the company and let it evaporate one closed conversation at a time. Mined properly, it tells you exactly what to fix, what to change, and what each one is costing you.
Revenue.
It's the hinge between the revenue you keep and the revenue you win — the deal that stalls, the renewal decided in week one, the account that churns or expands. Customer experience isn't a cost centre. It's the cheapest growth you have.
04
What I do
Diagnose. Build. Lead.
The same work, at three depths, without the full-time hire. Every engagement is build-led: I go into the problem, prove the value with something working, and leave the function sharper than I found it.
01 / Diagnose
Voice-of-Customer Diagnostic
I take your real conversations (tickets, calls, churn notes, sales objections) and turn them into a prioritised read: what is breaking, what it is costing, and what to fix first. A decision document your product and exec teams can act on, not a dashboard nobody opens.
02 / Build
AI & Conversation Enablement
I build the AI layer your customer operations run on, starting from the problem, not the tool. Sometimes that means configuring a platform you already run, like Intercom Fin, so it resolves instead of deflects; sometimes it means building custom on top, like the retrieval systems, MCP servers and agents. Most deployments need both, and knowing which the problem calls for is the job. I build to prove the value, then partner with your engineers to harden it for production: the point of a forward-deployed engagement.
03 / Lead
Fractional Customer Intelligence Lead
The senior seat, without the full-time seat. I shape strategy with your leadership, stand up the measurement, sharpen the AI, and run the loop that turns daily conversations into decisions. Part operator, part coach. The function holds its edge after I have gone.
05
Proof
Examples of my work
Atlassian
$61M
Rovo, Atlassian's flagship AI product, was launching to ~200 enterprise accounts, representing $61M in annual spend, with no fast way to know what they actually thought. I built a custom SQL and R reporting pipeline, with an AI agent synthesising raw feedback into structured, bi-weekly reports for product and go-to-market. Signal reached the teams who needed it, shaping real product decisions.
Stack
Custom VoC pipeline
AI Agent
Internal AI Gateway
Databricks
R
SQL
Heidi Health
75→87%
Heidi's AI front line sits in a regulated clinical environment, where a wrong answer isn't an annoyance. It's a safety issue. I designed the Conversation AI flows and agentic escalation paths running across it, on a Databricks VoC pipeline with Intercom's Fin as the resolution engine. CSAT rose 75%→87% while support volume grew 70% — accuracy holding as scale increased.
Stack
Databricks VoC pipeline
Intercom Fin
Conversational AI
Claude API
Python
SQL
Immutable
0→90%
Immutable needed a 24/7 global support function in a nascent, fast-moving web3 space — built from nothing, without a large team to run it. I built Intercom from the ground up, with chatbot-first conversational design engineered to resolve rather than deflect. Self-serve resolution went from 0%→90% at scale — proving the model: volume to the machine, moments to people.
Stack
Intercom
Chatbot-first architecture
Conversational Design
06
About
Most people who can lead the function cannot build it.
Fifteen years ago I started on a support desk at Apple. Since then: a developer, Intercom's first Sydney hire, CX leadership at Immutable, Shippit and Heidi, and Voice of Customer programs inside Atlassian. I hold a postgraduate qualification in data science, and SQL, Python, R & Ruby are how I let the data set the priorities. The idea that has run through all of it is simple: customer conversations, not tickets. The best products are built by teams who genuinely listen.
Used to deflect rather than resolve, AI walls customers off from the help they need and throws away the signal in the same move. Used right, it is what makes staying human affordable at scale. The AI era needs a hybrid skillset, the player-coach: someone who can read what an agent is getting wrong, redesign the handoff to fix it, coach the team through the change, then sit with the C-suite and explain what it means for the business.
I am forward-deployed by instinct. I build working systems to prove value, then partner with engineers to harden them. I have done both, leading the function and building what it runs on, with numbers behind it.
07
Contact