00 · diagnostic
Score your revenue engine in 2 minutes.
Six questions, one per layer of the engine plus two on who owns it. Answer as things are, not as the roadmap says. You get a score out of 100, the layer that is holding the others back, and a link to the fix.
01 · the reframe
Rented vs Owned
When a CRO asks me to look at a pipeline problem, I find the same shape every time. The demand gen motion is running. Campaigns fire and expire. Headcount turns over and takes the playbooks with it. The team is busy and the pipeline is thin, because the revenue motion is rented rather than owned. The audience belongs to LinkedIn and Google. The intent data belongs to a vendor who sells the same rows to your competitors. The throughput belongs to SDR headcount and ad budget, and it flatlines within one sales cycle of either being paused.
Revenue Engineering treats go-to-market as infrastructure you design, version and own. Matt McDonagh, writing under the term since 2024, draws the line against RevOps with a pit-crew metaphor: RevOps checks tyre pressure and runs clean stops; Revenue Engineering is the design team back at headquarters reading telemetry and shipping upgrades. RevOps runs the engine. Revenue Engineering designs it.
His version of the metaphor has a gap: nobody drives the car. Data integrity, process efficiency, forecasting, all engine work, none of it the race. A revenue engine nobody drives is a very clean CRM. The marketing-led school of Revenue Engineering, which is the one this guide describes, puts both halves in one seat. You build the enrichment pipelines, the scoring and the signal routing, and you run the ABM and demand programmes on top of them, and you own the pipeline number they produce.
Neither column of numbers happens without the other. The infrastructure sets the ceiling on how good the programmes can be. The programmes are the only evidence the infrastructure was worth building. Most organisations split this into two hires who never sit in the same meeting: an engineer with no demand number and a marketer who cannot see under the bonnet of their own campaigns. The handoff between them is where pipeline goes to die.
The core reframe: A campaign spends its value at the moment of delivery. A system shipped this quarter produces pipeline in eighteen months. You are building assets that appreciate, and your scoreboard moves from MQLs and cost per lead to sourced pipeline, compounding CAC and the share of the motion that runs without a human pressing a button.
| Dimension | Campaign model | Revenue Engineering |
|---|---|---|
| Primary metric | MQLs, cost per lead, CTR | Sourced pipeline, compounding CAC, citation rate |
| Time horizon | Quarterly cycles | Multi-year asset accumulation |
| Core asset | Rented audiences, purchased lists | Owned data, owned logic, owned content |
| Logic lives in | A marketer's head and a SaaS UI | A versioned repository |
| When spend stops | Pipeline flatlines within a sales cycle | Assets keep producing for years |
| Scaling mechanism | Add budget, add headcount | Add code, add agents |
| Failure mode | Volume without revenue; MQL-to-SQL blame loop | Slow start; needs engineering discipline up front |
🤖 Claude Co-Pilot Strategy
Use Claude to audit how much of your current motion is rented.
Act as a Revenue Engineering lead. Here is our GTM stack, channel mix and last four quarters of pipeline by source [paste]. Classify every input as rented (audience, intent, throughput we do not control) or owned (data, logic, content that keeps producing after spend stops). Estimate what percentage of this quarter's pipeline would survive a 90-day spend freeze. Name the three rented inputs that should be converted to owned assets first and what each conversion would take.Then: write the two-paragraph version for a CFO who thinks marketing is a cost line.
02 · the system
The Three-Layer Architecture
Every revenue engine I have built or rebuilt runs on three layers, and each one runs on the layer below it. Foundation is data you can trust. Modelling is the intelligence that turns that data into decisions. Activation is the execution that puts those decisions in front of buyers without anyone pressing a button. Skip a layer and the one above it fails in a way that looks like a tooling problem.
| Layer | Question it answers | What it contains | Tools I reach for |
|---|---|---|---|
| Foundation | Can we trust the record? | Dedupe, enrichment waterfall, field standards, ownership rules, CRM to warehouse sync | HubSpot or Salesforce, Clay, n8n |
| Modelling | Who should we talk to, about what, and when? | ICP scoring trained on your own deals, intent triangulation, propensity, the customer-truth file | Clay, Claude, the CRM's own closed-won history |
| Activation | Did the right action reach the buyer without a human in the loop? | Signal-triggered sequences, routing, ABM plays, answer-engine content, agents with job specs | n8n, Clay, HubSpot, Influ2, LinkedIn, Madison Logic |
The three layers are sequential to build and parallel to run. Enrichment jobs run on schedule. The ICP model gets retrained when the definition of a good customer changes. Activation loops get tuned as the signal-to-pipeline data comes back. The question to ask before any build is which layer is the constraint right now, because the answer changes what you ship first. I wrote up the three rungs and the two ways teams fall off them: activating on a foundation they never built, or polishing the foundation so long that nothing ever activates.
🤖 Claude Co-Pilot Strategy
Use Claude to find the constraining layer before anyone opens Clay.
Act as a Revenue Engineering lead running a diagnostic. Here is a description of our GTM stack, a sample of 50 CRM account records with field completeness, our current scoring rules, and how outbound is triggered today [paste]. For each layer (Foundation, Modelling, Activation) score readiness out of 10 with evidence from the sample, name the single constraint, and tell me which layer to fix first and why the others will not hold until it is fixed.Then: turn the answer into a one-page build order with a baseline metric for each item.
03 · layer one
Foundation: data you can trust
Most GTM failures I diagnose start here and get blamed somewhere else. The signal-based outreach fires on incomplete data. The ICP model runs on a CRM where 40% of records carry the wrong company size. The churn alert triggers on accounts that renewed last month. The team runs everything by hand because nobody built the layer that automation needs to stand on.
Foundation has four jobs, in this order.
- Dedupe before you enrich. Enriching a duplicate pays a vendor twice to make a routing error more confident. Deduplication comes first, with a merge policy the whole team has agreed.
- Run an enrichment waterfall, not a single vendor. Provider order matters more than provider choice. Cheapest reliable source first, escalate on miss, log which provider filled which field so you can renegotiate on evidence. The provider order I use and why.
- Standardise the fields the models depend on. Employee band, industry taxonomy, territory, lifecycle stage. If two systems disagree on what "enterprise" means, every score downstream inherits the argument. Field mapping standards are boring and load-bearing.
- Write ownership rules and enforce them in code. Orphaned records are where pipeline leaks without anyone noticing. Routing accuracy at Quantexa went from 55% to 88% when ownership moved out of a spreadsheet and into an n8n workflow that ran on every record.
The goal is a CRM where a rep, a model and an agent all read the same account and reach the same conclusion about who it is, how big it is and who owns it. Everything above this layer assumes that is true.
What good looks like: Enrichment runs on a schedule with no human trigger. Schema audits catch drift before it compounds. A new record is deduped, enriched, standardised and owned within minutes of creation, and the log shows which provider did what. The team spends its hours on Modelling and Activation because the hygiene work automated itself out of their calendar.
🤖 Claude Co-Pilot Strategy
Then: estimate the enrichment cost we are wasting on duplicates at our current per-record rate.
04 · layer two
Modelling: intelligence from your own signals
Modelling turns clean data into a prioritised answer to three questions: who should we talk to, about what, and when. A scoring model trained on your own closed-won and closed-lost deals beats any third-party intent tier, because it describes your buyers rather than a vendor's average. You build it once and it runs on every new account.
Four models carry most of the weight.
- ICP scoring on won and lost deals. Start from patterns in the deals you closed and the deals you lost, then weight fit attributes against them. Keep ICP scoring and TAM sizing as separate exercises; conflating them is how a tight score shrinks the market you are allowed to sell into.
- Intent triangulation. No single source is trustworthy. Third-party topic surges, first-party site and product behaviour, job changes, hiring, public complaints. Triangulate at least two before an account crosses a threshold, and apply decay so a signal from March cannot fire outreach in July.
- The customer-truth file. One dated markdown file that reads calls, tickets, churn notes and CRM notes and reports what changed, with a quote, a link or an event count behind every claim. Sales hears one version of the market, support another, product a third. This file is the only place they agree. The mechanism is in win/loss synthesis at scale.
- The feedback loops. Every closed deal recalibrates the ICP weights. Wins outside the assumed ICP widen the TAM. Segments that never convert get cut. A Monday digest tells the team what moved and what to do about it. The full set is in the self-improving revenue system, along with the warning: a loop trained only on closed-won data walls you into your own history.
The vague version of Modelling produces a summary that says "customers want better collaboration." The engineered version says: five calls this week mentioned emergency dispatch, but the three that converted all talked about the follow-up quote that never got sent after the technician left. One of those rewrites the landing page. The other is decoration.
🤖 Claude Co-Pilot Strategy
Build the first customer-truth file from real transcripts.
Read these 20 sales call summaries, 30 support tickets and our last quarter of closed-lost notes [attach]. Write what-the-market-is-telling-us.md: the five pains buyers named most in their own words, which pains appear in deals that converted versus deals that stalled, language buyers use that our website does not, and objections that recur. Every claim needs a verbatim quote or a ticket reference. End with the three things to test this week and the one positioning change the evidence supports.Then: compare the file against our current ICP scoring weights and flag every attribute the evidence says we are over- or under-weighting.
05 · layer three
Activation: execution without a button press
Activation acts on what Modelling produces. A signal crosses a threshold, the account is researched, the angle is drafted, the sequence fires or a human approves it, and the result is written back so the next fire is smarter. No one presses a button. The logic that decides all of this lives in a repository, not in a marketer's head or a SaaS UI, which means it executes on every record and survives the person who wrote it leaving.
Five kinds of activation earn their place in a B2B engine.
- Signal-triggered outbound. Bad outbound starts with a spreadsheet of names. Good outbound starts with timing: who raised, who is hiring for the problem you solve, who complained in public, who fits and has a reason to care this week. Measure signal-to-sequence latency in minutes.
- Inbound routing with context. Every trial or demo request scored, enriched and routed in under two minutes, with a first message pre-written into the rep's CRM record before they open the lead. The routing pattern that does it.
- ABM plays on the engine. 1:Few programmes across a scored account list, contact-level paid on Influ2, syndication through Madison Logic where the committee is wide. The 1:Few programme that influenced $6.7M ran on this stack. Contact-level ABM produced 5.2x the pipeline of cold outreach, and a 1:1 programme cut deal-close time 66%. The ABM playbook covers the programme side.
- Answer-engine content. Search resolves inside the answer now. A structured note that answers the question your best prospects ask on every first call gets cited by ChatGPT and Perplexity at the moment of intent, paid for once. Citation infrastructure is the Activation asset with the longest half-life.
- Agents with job specs. Write each agent's job the way you would write a hire: data source, schedule, what to filter out, what good output looks like, what needs human approval, the metric that matters, where to write the result. Then supervise it like a new hire. Agent management is the discipline; the stalled-deal spotter and the weekly signal monitor are two worked builds.
Two disciplines keep Activation from turning into spam at scale. Error handling, because a workflow that fails silently is worse than one that never ran. And the judgment to leave some work manual, because an agent will make that decision for you if you do not.
The ownership test: If the person who built the routing logic left tomorrow, would it still run, and could their replacement read why it does what it does? If the answer is no, the logic is rented from an employee rather than owned by the company, and it will expire with their notice period.
🤖 Claude Co-Pilot Strategy
Write the job spec for your first activation agent.
Write a job spec for a signal-triggered outbound agent, formatted like a hiring brief. Inputs: our ICP definition, the three trigger signals we trust most, our approved and banned language [paste]. Specify: data sources and the schedule it runs on, the filters that drop bad-fit accounts, the research it performs per account, the exact output (angle, first line, evidence for the trigger), what requires human approval before send, the single metric it is judged on (qualified positive replies, not messages sent), and where it writes results so the next run learns. Add the five rules it must never break.Then: list the first ten mistakes you expect it to make and the correction you would add to its memory for each.
06 · positions
Five Positions Worth Defending
Tactics expire. Positions compound. These are the arguments underneath the architecture, each one a claim about where B2B revenue work is going and each one I have had to defend in a budget meeting.
Logic in version control beats logic in a UI
If your routing rules live in a HubSpot workflow nobody has opened since the person who built it left, you do not own your GTM. You rent it from a former employee.
- The argument
- Go-to-market logic is code whether or not you treat it as code. Scoring weights, routing rules, enrichment order, suppression lists, attribution definitions. Held in a SaaS UI they are invisible, unversioned and unreviewable. Held in a repository they have history, a diff, an owner and a reason. A rule written once executes on every lead, and when it changes you change it in one place.
- The implication
- Treat the GTM repo as a first-class asset. Every automation gets a README, a baseline metric and an owner. Every prompt an agent runs on is a file with a commit history. The context engineering work most teams skip is the same work that makes the system survive turnover.
Own the signal or share the edge
By the time your intent vendor flags a surge, three competitors bought the same row. Purchased intent is a floor every competitor stands on.
- The argument
- Signals sort by how exclusively you hold them. Bought third-party topic data sits at the bottom. Borrowed community and partner signal is narrower. Observed first-party behaviour on your site and product is yours alone. Manufactured signal, a benchmark, a free tool, a public index you publish, sits at the top because you create the surge and see it first. Edge climbs as you ascend.
- The implication
- Move budget up the ladder. Fund the free tool and the proprietary index before you renew the intent subscription. Triangulate whatever you buy against whatever you observe, and never let a bought signal fire outreach on its own.
Metric before build
A build that claims to improve everything proves nothing. A build that moves trial-to-paid from 8% to 18% in 90 days on the same inbound volume proves something a CFO can use.
- The argument
- Revenue Engineering ROI is harder to prove than most technical investment because one pipeline touches routing, rep research time, personalisation and data quality, and the credit diffuses across four teams. The teams that prove it define the metric before they build. Log the baseline: current conversion at one stage, current routing accuracy, current median rep research time, current signal-to-first-touch latency. Run for 60 or 90 days. Measure again. The delta is the ROI.
- The implication
- No build ships without a recorded baseline and a named success metric. Pick one stage, not the whole funnel. How to do it and what to count.
The builder drives the car
The most expensive handoff in B2B is between the engineer who built the engine and the marketer who runs programmes on it. Put them in the same seat and the handoff disappears.
- The argument
- Split the job and each side optimises for its own half. The engineer ships a scoring model nobody's campaign uses. The marketer runs an ABM programme on scores they cannot explain to a CRO. Neither owns the number, so neither is accountable when it misses. The marketing-led revenue engineer builds the enrichment pipeline in the morning and presents the positioning change the customer-truth file produced in the afternoon, and answers for the pipeline both of them created.
- The implication
- Hire for both halves or grow one person into both. The interview question that sorts the field: walk me through a demand programme you ran on infrastructure you built yourself. Most candidates can describe one half. The full argument and where the seat sits.
Agents are a commodity. Judgment about what to point them at is the moat.
Every company will have access to the same models. The one that wins learns its market faster and decides better what the machine should build next.
- The argument
- AI collapsed the cost of producing marketing output to near zero, which makes average output worthless and taste expensive. The scarce skills are deciding what should exist, knowing when a personalised first line sounds fake, choosing which of twenty hooks is on-brand rather than clickable, and supervising a worker whose output varies run to run. Those are management skills with an engineer's toolkit attached.
- The implication
- Budget for supervision, not only for agents. Build evals the way you build A/B tests. Decide which outputs get read by a human and which get sampled. The autonomy ladder tells you what your calendar looks like at each rung, and why the person doing this is a leader, not a power user.
07 · the seat
The Seat and the CRO Case
Revenue Engineering is a discipline and a leadership seat. The formal title varies: Marketing Revenue Director, Head of Revenue Engineering, Director of GTM Systems, VP Marketing with a systems remit. GTM engineer is the discipline keyword inside it, the way front-end is a discipline inside product engineering. The test that matters is whether the person who designs the engine also owns the demand number it produces. If they do not, you have hired an ops function with better tooling.
CROs do not approve automation budgets. They approve pipeline investments, weighed against the alternative they understand: another rep. Lead with the system and you lose the room. Lead with the leak and you get the follow-up meeting. The structure that works:
- Name the leak. "We generate 300 trials a month and convert 8%. The category converts closer to 20%. That is 36 deals a month we should be closing and are not."
- Identify the proximate cause. Skip the root-cause essay. Routing failures. No context at handoff. The same sequence for Tier 1 and Tier 3 accounts.
- Describe the build as an outcome. "Every inbound trial scored and routed in under two minutes, with a first message pre-written in the rep's CRM record." The CRO does not need to hear Clay, n8n and Claude.
- Quantify conservatively, with your numbers. Routing accuracy from 55% to 85% and conversion up 40% is X deals a quarter. Industry averages get discounted; internal baselines get believed.
- Frame the cost of inaction. The leak compounds every month the decision waits. Put a monthly figure on it.
The budget line to fight for: The engine and the programmes in one plan, with a 90-day proof point and a baseline logged before day one. Ask for the systems budget without the programme and you get an ops project. Ask for the programme without the systems and you get a campaign that expires.
🤖 Claude Co-Pilot Strategy
Then: rewrite the same case for a CFO who will ask when it pays back.
08 · execution
The 90-Day Build
Diagnosis before architecture. Most underdelivering GTM projects were misdiagnosed at kickoff; the build was correct and the brief was wrong. Then ship the layers in order, with a metric logged before each one goes live, and end the quarter with a number the CRO recognises.
Days 1 to 30: Diagnose and lay the Foundation
Build trust in the record before anything runs on it.
- Map how revenue flows through the business: where it stalls, what data exists, what is missing. Name the constraining layer.
- Log the baselines: routing accuracy, conversion at one chosen stage, median rep research time, signal-to-first-touch latency.
- Dedupe, then stand up the enrichment waterfall on a schedule. Agree the field standards and ownership rules and put them in code.
- Create the GTM repo. Every rule, prompt and workflow gets a file, an owner and a README.
- Convert the single highest-friction manual workflow the team repeats into a function that runs on every record.
Days 31 to 60: Model, then fire the first loop
One working system beats five half-built ones.
- Ship the first customer-truth file from real calls, tickets and closed-lost notes, with receipts behind every claim.
- Build ICP scoring from your own won and lost deals. Keep TAM sizing separate. Write down the weights and why.
- Pick one activation loop and build it end to end: signal in, account researched, angle drafted, human approval, result written back. Measure its latency in minutes.
- Write the job spec for the agent that runs the loop. Start it on small tasks, read every output, add each correction to its memory.
- Ship the first answer-engine note: the question your best prospects ask on every first call, answered in the first forty words.
Days 61 to 90: Run the programme and prove the number
Drive the car you built.
- Launch the first ABM or demand programme on the scored list, using the angles the customer-truth file surfaced. This is where the engine earns its budget.
- Stand up the growth cockpit: what changed this week, which objection came back, which test won, what to test next, and the decisions taken before the meeting.
- Close the first feedback loop: closed-won and closed-lost from the quarter flow back into the ICP weights.
- Re-measure every baseline from day one. Write the case study with the deltas in it, and the one number that moved most on the first slide.
End of quarter, the story should read like this: we rebuilt the data layer and routing accuracy went from 55% to 88%; the customer-truth file surfaced a pain the deck never mentioned; we built a signal engine around it that shipped 75 targeted messages, got nine qualified replies and booked three meetings; and the programme running on it is the first one with a pipeline number the CRO believes. That is how you get the seat, keep the budget and hire the next person into the discipline.
🤖 Claude Co-Pilot Strategy
Days 1 to 30. Turn the diagnosis into a build order:
Here is our GTM diagnostic [paste]. Produce a 30-day Foundation build order: the manual workflow to convert to code first and why, the dedupe and ownership rules, the enrichment waterfall with provider order, and the baseline metric to log before each item ships.Days 31 to 60. Spec the first loop:
Design one signal-triggered activation loop for [segment]: the two signals that must both fire, the research step, the approval gate, the sequence, the write-back, and the latency SLA. Output it as a runbook an ops engineer can implement in n8n and Clay this month.Days 61 to 90. Write the proof:
Here are our day-1 baselines and day-90 measurements [paste]. Write the one-page results memo for the CRO: the number that moved most first, what caused it, what it is worth in pipeline, and the two builds we are proposing next with their expected deltas.faq
Common questions about Revenue Engineering
01What is Revenue Engineering?
02What is the difference between Revenue Engineering and RevOps?
03What is the difference between Revenue Engineering and GTM engineering?
04What are the three layers of a revenue engine?
05What should a Revenue Engineering team build first?
06How do you make the case for Revenue Engineering to a CRO?
07Is Revenue Engineering a job title?
next
Revenue systems. Not headcount.
I design revenue engines and run the demand programmes on them: enrichment, scoring, signal automation, ABM, and the measurement that survives a CFO's questions. If you want this guide turned into a running engine with a number attached, start a conversation.