HubSpot-first, no RevOps
“I've hired for GTM, but there's no system”
Outbound has filled HubSpot with records that are not people. Nobody trusts the contact count, the ICP lives on a slide, and each rep applies it differently.
Audiences & Signals
Know who to talk to, and why now — with the source attached.
A GTM engineer builds your audience layer: a CRM where a contact is a person and a company is an account, your ICP and personas as rules that grade every record the same way, prospecting and enrichment data spent where it matters, and the signals, from your own data, web research and social listening, that say an account is worth working this quarter, with the source attached.
Who this is for
You buy data. You still can't say which accounts to work.
Ecosystem sellers and agencies
“We run this for others”
The fact you need is not for sale: which agency is likely working that event, which brand is likely re-platforming. By the time anyone announces it, the window to act has closed.
What we build
A clean CRM first. Then rules. Then signals, with their sources.
CRM hygiene
Every contact and company is classed by where it came from, duplicates and orphans go into review lists your team clears in the CRM, email health is flagged, and merges and deletes run from rules your team set.
Your ICP, written as rules
Your ICP is derived from the deals you won and lost and written as rules that grade every company into a tier and every contact into a persona, checked against your team's own labels before they go live.
Enrichment that needs a reason
Enrichment falls back across the data providers you have connected, on your own keys, behind spend rules that buy data only where your ICP says it matters, up to the cap you set.
Research with its sources
Account research is a set of typed questions, each answered with a confidence and at least one source, written to the record as a dated signal that expires.
Hard-to-get signals
A signal no vendor sells becomes a likelihood question, scored from evidence that starts with the prior year so there is lead time to act: we say likely, and we show the evidence.
Social listening
A daily queue of public conversations about the problem you solve is triaged by problem and intent, with a reply drafted in the voice of a named person on your team, who posts it.
What you're left with
The rules are files. The reasons are on the record.
In your repo
The ICP rules, the hygiene settings, the research question set and the listening taxonomy are files in your repo, each with a version. A change is a pull request, and the history shows who changed what.
In your CRM
Record class, tier, persona, email health and signals sit on the record, each with what was decided, why, the evidence and when. The review lists your team clears are CRM lists.
GrowFlex-operated, on this line
Likelihood engine
Involved only for a hard-to-get signal. The question, the scored list and your labels are yours, and are kept in your accounts.
What you receive at exit
Fixed-weight combiner and your labelsA fixed-weight combiner you run yourself, plus your truth labels. What stays behind is the calibration learned across clients.
Warehouse
Involved only if you have no warehouse of your own and want reports and history kept in one.
What you receive at exit
pg_dump and CSVsA full export, or the warehouse re-pointed at your own database and backfilled.
Where judgement sits
The rules are code. Judgement sits on top.
- The same every run. Record class, dedup tiers, ICP grading, spend caps and the research question set are code in your repository, so a record is graded the same way on every sweep.
- Where judgement adds something. Deriving the ICP from your deals, reading a page to answer a research question, sorting a listening thread and drafting a reply are reasoning, and each result is recorded with what the model was shown.
- What your team decides. Your team signs the ICP, labels a sample of records, sets the spend reasons and caps, and posts every reply from their own accounts.
Entry offers
One piece at a time.
A written scope, a typical timeline, and a deliverable you can inspect on day one. Timelines are typical, not guaranteed. The proposal sets the one for your stack. Every engagement starts with the Stack Audit: 1 week, and its readout proposes the next piece for your stack, priced in the proposal.
Typical timeline: 2–3 weeks
ICP & CRM Hygiene Sprint
For the team that can't trust the CRM: junk, warmup artifacts, duplicates, orphans, stale emails; no reliable ICP.
- What we do
- We clean your CRM as a plan you approve, covering record class, duplicates, orphans, decay and email health, then compile your ICP into rules checked against your team's labels before they go live, with review lists and a nightly sweep.
- Day-one deliverable
- The hygiene plan: every record id, action and reason, with counts by category (orphaned contacts, companies and deals; what to automate, what to review).
Typical timeline: 2 weeks
Signal Sprint
For the team that buys data but still can't tell which accounts to work this quarter.
- What we do
- We write a small set of typed, sourced research questions, label and evaluate them on a sample of accounts, write the answers to your CRM with their evidence, and report coverage.
- Day-one deliverable
- A list of target accounts with answers and the URLs behind them; the question set you now own.
Typical timeline: 3–4 weeks
Hard-to-get Signal
For the team that needs a signal no vendor sells ("likely to be at that event", "likely re-platforming") with enough lead time to act.
- What we do
- We take one likelihood question end to end: evidence design including your internal data, prior-year evidence first, a schedule, a scored list with evidence lines, the triggered email, and calibration once the outcomes are known.
- Day-one deliverable
- First scored list, each account banded act or review, with the evidence per account.
Typical timeline: 2 weeks
Listening Program Setup
For the team that knows buyers talk about the problem in public but has no safe way to join.
- What we do
- We audit the taxonomy and venues, set up the sweep and the LinkedIn capture tool, write the style cards, run daily triage with your team for the first weeks, and deliver the first synthesis memo.
- Day-one deliverable
- The taxonomy, the queue, the first week's ranked engage items with drafts.
Proof
What already runs.
Used in GTM systems we've built for B2B SaaS clients running in production every day.
- Every company in the CRM is kept graded under a versioned rule set, re-checked by a nightly sweep.
- Every new contact is classified and graded shortly after it is created.
Questions
Questions we hear first.
HubSpot already dedupes. What is left to do?
HubSpot dedupes on exact email, and it cannot tell a warmup mailbox from a prospect. The duplicates you can see in its own tool are the smallest part of the cleanup. The larger part is classing every record by where it came from, which is what the Hygiene module does first, so that a warmup artifact is never merged into a real person.
We already have an ICP. Why write it again?
Most ICPs live on a slide, and each rep applies them a little differently. Written as rules, yours grades the backlog and every new contact the same way, puts the reason on each record, and can be checked against your reps' own judgment. The grading is deterministic code, so it is the same every time and you can audit it. Reasoning is used where it helps: deriving the definition from your deals, and reviewing where the rules and your reps disagree.
Apollo already enriches inside HubSpot. Why put a gate in front of it?
It does: on every record, with no reason and no cap. That is how credits get spent on records nobody is going to work. The gate spends only on the records your ICP says matter, writes down why it refused the rest, and reports spend by reason, so you can see what the cap bought.
We tried agent prompts per row and the answers didn't hold up. Why would this?
So did we. An open prompt per row is expensive, hard to check, and wrong without telling you. Here research is a closed set of questions. Each has a typed answer, a source and a cap on what one answer may cost, and a question goes live only after it has been checked against records your team answered by hand. Reading the pages and answering is the reasoning half; the question set, the cost cap, the expiry and the write to the record are code. Unknown is an allowed answer.
ICP & CRM Hygiene Sprint. 2–3 weeks, fixed scope.
Every engagement starts with the Stack Audit: 1 week. Its readout says which piece to build first for your stack and prices it as a fixed fee.
Vendor spend stays yours: reported, never marked up.