Calculated properties don't notice time
HubSpot recalculates a calculated property when one of its inputs changes, never because time passed. Anything that should decay, expire or age needs a timer outside the formula.
HUBSPOT, ENGINEERED
Custom objects, cards on the record, webhooks into your own code, and syncs that keep every record current. Engineered for how you work, inside the limits of your edition.
HubSpot is our specialty. Every system we ship, every test suite and every lesson in our field notes is built on it. So we know it from its API responses, not its feature pages: which writes it silently discards, which values never recalculate, what your edition includes, and what to build outside it when it can't do the job.
Every hour goes into HubSpot's APIs, limits and developer tools.
We check what your subscription includes before the design depends on it.
What production teaches us goes into tested code and runbooks, not into someone's memory.
Most HubSpot projects stop at configuration. Ours keep going, into code, wherever the business needs it.
| Topic | Standard setup | What we build |
|---|---|---|
| Data in | Standard setupCSV imports and whatever the native app installs | What we buildScheduled syncs and verified webhooks, keyed so nothing writes twice |
| Duplicates | Standard setupMerged by hand when someone notices | What we buildMerge plans from normalized names, addresses, phones and geocodes, reviewed, then applied through the merge API |
| Record view | Standard setupDefault properties in the sidebar | What we buildUI extension cards with data from outside HubSpot, served by signed Workers |
| Associations | Standard setupWhatever the person creating the deal remembered to link | What we buildEvery contact of the company linked to each new deal within seconds, labeled where the relationship needs a name |
| Values that age | Standard setupCalculated properties that recalculate only when an input changes | What we buildValues recomputed on events, on timers and every morning, so a quiet deal shows as quiet |
| Automation | Standard setupWorkflows that stop where the editor stops | What we buildCustom workflow actions and webhooks into code: queues, throttles, external sends, results written back |
| Change control | Standard setupEdits in the UI, no history | What we buildDeveloper projects in git, workflows backed up before every edit, uploads from the CLI |
Nine kinds of work we build on HubSpot. Combined, they turn a portal into one organized system.
Objects for what your business actually tracks, created through the schemas API, with association labels for relationships that need a name: principal, guarantor, billing contact. Custom objects need Enterprise. On Professional we model with companies typed by kind, deals and labels, and write down where that design strains.
React cards built with @hubspot/ui-extensions, in record tabs and the preview panel. They run in a sandboxed web worker with no DOM and reach outside only through hubspot.fetch, which allows listed origins, caps payloads at 1 MB and times out after 15 seconds by default. We build the Workers behind them to fit those limits.
Apps, cards, app pages, webhooks and workflow actions defined as files in a developer project, versioned in git and uploaded with the HubSpot CLI. We pin the CLI version per project: an older CLI can validate a newer project and upload it without its page component.
Subscriptions to record creation, property changes and association changes, delivered to a Worker that verifies HubSpot's v3 signature, answers at once and does the work after. We prove each subscription by triggering a real event and timing the delivery.
Cloudflare Workers for the jobs HubSpot cannot run: Durable Object alarms for values that change with time, D1 queues, KV caches, scheduled backstops. HubSpot's own serverless functions need Enterprise; the Workers in our field notes run behind a Professional portal.
Two kinds. App-defined actions that call your code from any workflow: hold a record and release it on a schedule through the callback API, send through your own mail server after a guard checks opt-outs. And custom code actions, which run your JavaScript inside the workflow on Data Hub Professional.
Batch reads and writes of up to 100 records per call, search for filtered sets, full exports paged with cursors, and backoff on 429. Every batch response is checked for partial failures, because a partly failed batch still reports COMPLETE.
Two-way syncs with ERP, billing, email and data providers. Email is logged through the CRM v3 emails API with the right direction and threading, and a unique key per message, so HubSpot rejects a duplicate itself.
Companies and contacts matched on normalized names, shared phone numbers and two-stage geocoding, merged through the merge API from a reviewed plan, with a progress log a run can resume from. Primary-company fixes go the same way: dry run, then execute.
Several HubSpot features depend on your subscription. We check yours before the design depends on any of them, and build around the gaps.
| Feature | Edition | How we design |
|---|---|---|
| FeatureCustom objects | EditionEnterprise | How we designOn Professional: companies typed by kind, deals and association labels, with the trade-offs written down |
| FeatureServerless functions in developer projects | EditionEnterprise | How we designCloudflare Workers, called through signed fetches and webhooks |
| FeatureCustom code and send-webhook workflow actions | EditionData Hub Professional | How we designApp-defined workflow actions that call a Worker |
| FeatureWorkflows that recur daily, weekly or monthly | EditionData Hub Professional | How we designA scheduled Worker, or a timer that fires at the exact moment a value should change |
| FeatureUI extension cards in a private app | EditionShipped on Sales Hub Professional | How we designCards for outside data on deal and company records, served by Workers |
Editions and packaging change. We confirm against your portal's own limits endpoints before we design, not against a pricing page.
Lessons from systems running against a live portal. Each one shaped code we ship.
HubSpot recalculates a calculated property when one of its inputs changes, never because time passed. Anything that should decay, expire or age needs a timer outside the formula.
Read-only rollup properties accept an API write, return success and stay empty. We read each property's definition before writing, and write to the object the rollup mirrors.
Batch endpoints report status COMPLETE alongside a count of errors. We check the error count and the errors array on every response, never the status alone.
Updating an enumeration property's options replaces the whole list; sending only the new option deletes the rest. We read the list, add to it and send it back whole.
A record created a moment ago may not show up in search yet. Duplicate checks search again just before the create, and a unique property backs them up.
A new deal's primary company is a rollup that fills in seconds after creation. A workflow that matches on it at the trigger finds nothing; ours waits for it.
Property-change webhooks fire for HubSpot's own activity rollups, such as last contacted, but an immediate re-read can return the old value. We trust the value in the payload.
Every script, Worker and card on the same private-app token shares one rate limit, so we batch reads and writes and back off when HubSpot says to.
A private app's webhook subscriptions can't be read through the API with its own token. We prove each one by triggering a real event and watching it arrive.
Legacy CRMs, spreadsheets, a second portal that came with an acquisition: we move records with their history (activities, associations, owners, and the IDs that link them back to the old system). Nothing is cut over until counts and samples match.
Objects, fields, workflows and integrations in every source, with record counts.
Each source field to its HubSpot property, with the rule that wins on conflict and the old system's ID kept on the record.
A full load into a developer test account, compared count by count and sampled record by record.
A final delta load, workflows switched on in a planned order so bulk writes don't flood enrollments, and the old system set to read-only.
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 noteFifteen custom cards that put property, valuation, sale history, area and map data on HubSpot deal and company records, with the property data behind a signed, cached Worker.
Read the field noteEvery send and reply logged to the right HubSpot contact, every positive reply turned into a deal for its owner, and follow-ups drafted by Claude inside a workflow action with a send guard.
Read the field noteFAQ
Depth. Every system we build, every test suite and every lesson in our field notes is on HubSpot, so we know its APIs, limits and developer tools from production, not from a feature list.
It depends on the build. Custom objects and HubSpot-hosted serverless functions need Enterprise. Custom code actions, send-webhook actions and workflows that recur daily, weekly or monthly need Data Hub Professional. A lot of what we build runs on Professional through Workers and app-defined workflow actions. We check your portal's limits before we design.
Yes. We consolidate legacy CRMs and spreadsheets into one HubSpot portal and keep the history, so nothing your team built up is lost in the move.
Yes. We read each workflow first, back it up before editing, and test the fix on real records. We have found a workflow that ran on every new deal for months and changed nothing.
Every inbound request is checked against HubSpot's signature. Endpoints that write refuse to run without their secret. Tokens live in the Worker's secret store, never in code or logs, and each private app gets only the scopes its job needs.
It can: a bulk write enrolls records in every workflow whose criteria they now meet. We check enrollment criteria before any bulk change and test on a handful of records first.
Figures show illustrative sample data. Names, companies and values in them are invented.
We'll read how it's built, tell you what we'd change first, and what your edition makes possible.