Skip to content

The problem

A standard setup leaves the business in pieces.

Most CRM projects stop at the standup: a pipeline, some properties, an import. The rest of the business stays where it was. Orders live in the ERP, invoices in billing, conversations in the inbox, and the numbers leadership reads in a spreadsheet someone rebuilds every Monday.

  1. Five versions of one customer

    Each system spells the account its own way, so nothing joins and duplicates come back with every import.

  2. Activity the CRM never sees

    Replies, orders and payments happen elsewhere. The record says the account went quiet; the inbox says otherwise.

  3. Automation that fails quietly

    A workflow can run on every new deal for months and do nothing. Nobody notices, because a wrong value looks like a real one.

FIG. 02BEFOREMON 08:10Sample: one customer as five disconnected systems see it, with different spellings, owners and last-activity dates.

The idea

One record per relationship.

Every system writes to it, on a schedule or the moment something happens. Duplicates merge by rules you can read. Agents research, enrich and flag on that same record, and people approve what matters. Leadership reads the business from one place, because there is only one place.

What we build

The parts a standard setup skips.

Data pipelines

Scheduled jobs that move millions of rows from public records, ERPs and legacy systems into a database and into HubSpot. Every run logs itself, and a bulk load is proved against what is live before it publishes.

  • Nightly and intraday jobs with heartbeats, locks and logs
  • Staged loads, proved, then swapped in one transaction
  • Matching on addresses, names and legal entities
Data & integration automation
FIG. 03NIGHTLY CLERK SYNC19:30 CTSample run log of the nightly county clerk sync: file downloaded, filings loaded, heartbeat written, trustee roster refreshed an hour later.

Integrations and sync

Webhooks and APIs connect HubSpot to the systems around it: email, billing, ERP, forms and data providers. Every inbound request is verified, and every write carries a key, so a retry never makes a duplicate.

  • HubSpot signature checks on every webhook; write endpoints fail closed
  • Idempotency keys, so a redelivered event writes once
  • Association logic that links contacts, companies and deals correctly
Data & integration automation
FIG. 04WEBHOOK TO RECORD14:02 CTSample data flow: three events pass through a verifying integration worker and land on HubSpot records.

AI agents in the record

Buy where it fits, build where it doesn't, integrate either way. Packaged agents for research, qualification, call notes and support triage, wired into your HubSpot data; custom agents on Claude where nothing off the shelf does the job.

  • Packaged agents selected, deployed and connected to your records
  • Custom agents on Claude that call the HubSpot APIs
  • Dry runs, deny-lists and approval steps before any write
AI agents
FIG. 05AGENT INBOX07:30 CTSample agent work queue: two items awaiting approval, one flagged deal, one research task in progress and two completed overnight.

Custom HubSpot apps and UI

When the standard objects stop fitting, we model the business as it is: custom objects and associations, cards on the record that show data from outside HubSpot, and workflow actions that run your own code.

  • UI extension cards on deal, company and contact records
  • Custom objects, association labels and permission design
  • Custom workflow actions backed by serverless Workers
CRM architecture & development
FIG. 06PROPERTY CARDS10:15 CTSample HubSpot deal record for a retail property with four sidebar cards: property overview, valuation and tax, sale history and area snapshot.

Built around how your industry works.

Same engineering, different data model. Each page sets out the objects, integrations and agents that business needs.

Private equity & roll-ups

Several acquired companies, several CRMs, no portfolio view. One data model in HubSpot, roll-up reporting, and a migration that keeps the history.

Mid-market B2B

You outgrew the starter setup. Reps work from spreadsheets and nobody trusts the forecast. We rebuild HubSpot around your real sales motion.

Real estate & investment

Properties, owners, entities, capital partners and deals as first-class records, fed by public-records data every night.

Manufacturing

Equipment makers, parts suppliers and contract manufacturers selling direct and through dealers. RFQ to order, ERP pricing and orders in HubSpot, and the installed base.

Food & beverage brands

Distributors, brokers, retail chains and foodservice in one account hierarchy, with orders, reorders and depletions synced from your systems.

Field Notes

How the hard parts got built.

Anonymized notes from HubSpot systems running in production: what was broken, what we built, what it does now.

Deal momentum on Cloudflare Workers

A HubSpot field that showed stalled deals as healthy, rebuilt as a Worker that recomputes on every activity, schedules its own decay and links contacts to new deals in two seconds.

Read the field note
FIG. 07 · DEAL MOMENTUM. Sample board of deals grouped by momentum: Advancing, Engaged, Slipping, Cold and Never contacted, with the time since last contact on each.

Figures show illustrative sample data. Names, companies and values in them are invented.

By the numbers

Counted in the systems, not the pitch.

Every figure below is counted from systems we built and run in production: their own load reports, logs and test suites.

county filings indexedA clerk's real-property index back to 1960, extended every evening at 19:30 Central.
35.7M
property records in one databaseAppraisal rolls from four counties in three states, loaded and proved by one loader.
3.8M+
Cloudflare Workers extending HubSpotWebhooks, enrichment, email logging, send scheduling and intake.
11
custom cards on HubSpot recordsProperty, valuation, sale history, area and map data on deals and companies.
15
from new deal to contacts linkedMeasured live. The native workflow path waits two minutes by design.
2.0 s
test assertions on one WorkerThe property-data proxy's suite runs in plain Node, with no network.
1,500+

Counts as of November 2026, taken from systems in production.

What you get from us.

Architecture

We design the data model before the first property: objects, associations, owners, and the source that feeds each field. You see the map before we build.

Engineering

Real code in your accounts: Workers, loaders, cards, tests and runbooks. HubSpot specialists, down to the API behavior the documentation leaves out.

Partnership

We stay through launch and after. When the business changes, the system changes with it, and the documentation keeps up.

From first call to a running system.

How engagements run
  1. Discover

    We map how the business actually runs: every system, every spreadsheet, every handoff, and where the data breaks.

  2. Architect

    We design the data model, the integrations and the agents, and agree the order to build them in.

  3. Build

    We build in small pieces. Each one is tested, documented and live before the next starts.

  4. Run and evolve

    We watch the jobs, fix what drifts, and add what the business needs next.

Show us the system you wish you had.

Bring the current setup and the spreadsheet everyone actually uses. We'll sketch the target system with you in one working session.