How we keep our CRM clean
Leave a CRM alone and it gets messy all by itself. Web forms and imports create the same person twice. New records arrive with half the fields blank. Deals sit in the pipeline for weeks after everyone has stopped talking. Soon the pipeline report describes a business you don't quite recognise, and the sales team quietly stops believing it.
Our fix is an agent on Machine that tidies HubSpot every night. It merges duplicate contacts, fills in blank fields from enrichment data, flags deals that have gone cold, and posts a short report on the state of the data. Any bulk edit above a set size waits until a person approves it.
The problem
Keeping the CRM tidy never makes it onto anyone's calendar. When one company is split across two contact records, its history is split too, and the account owner only sees half of it. Blank job titles, industries and company sizes make every segment a guess. Deals that have not budged in weeks still inflate the forecast.
The common remedies wear off. A big one-off clean-up lasts until the next import messes it up again. A paid de-duplication add-on catches the doubles but does nothing for blank fields or cold deals. Asking reps to tidy their own records means asking them to choose admin over selling, and selling wins every time. The upkeep has to happen by itself, every night, with nobody standing over it.
What we built
Every night a cron trigger points an agent at HubSpot. Each run gets a sealed-off session of its own and narrow access to HubSpot and our enrichment provider. The agent finds duplicate contacts and merges them, fills in what is missing from enrichment, marks deals that have stalled, and posts a report of everything it touched. A bulk edit that crosses a set size pauses for a person before anything is saved.
How it works
Schedule a nightly run
A cron trigger starts the project once each night. Every start creates a brand-new session: a fresh cloud computer (sandbox) used only for that run, holding just the CRM and enrichment access a clean-up needs. When the job is done, the computer is thrown away.
Write down the house rules
How we model our data is stored in skills and memory, the playbook loaded at the start of every run. It covers which fields must be filled, how we tell that two contacts are really one person, when a deal counts as cold, and how merges should be done. Whenever an odd case turns up, we add it to the rules.
Plug in HubSpot and enrichment
Connectors are scoped connections to your tools, and the keys stay on our servers, out of the model's reach. Through them the agent can:
- Work in HubSpot: read contacts and deals, merge the doubles and fix fields.
- Ask the enrichment provider: fill in blank job titles, company sizes and industries using outside data.
- Post in Slack: a brief report of each run in the RevOps channel.
Put a limit on big changes
Merging duplicates and filling single blanks happen without supervision. A bulk field update above a set size is different: it halts at a human approval gate before saving, so any change that sweeps across hundreds of records gets a look first. Credentials are stored encrypted in the Secrets Manager and only handed to the agents you grant them to, at the moment a run starts.
Read the morning report
Each run merges duplicates, enriches the blanks, flags the deals that went quiet, and then reports back: how many contacts it merged, which fields it filled, which deals it flagged. If a bulk edit is waiting for approval, the report spells out what it is and why it is waiting.
The pattern
Start the clean-up with a cron trigger. Give the agent scoped connectors into HubSpot and your enrichment source. Write your data rules down as skills and memory, and put a person in front of any big bulk change. From then on, the nightly run tidies up and reports back by itself.
Guardrails
An agent that can write to your CRM has to protect the data, not just tidy it. On Machine that means:
- Isolation. Every run lives in a sandbox of its own, and the HubSpot and enrichment edits it is explicitly allowed to make are the only things that leave it.
- Scoped secrets. The HubSpot and enrichment credentials are stored encrypted in the Secrets Manager and only handed to the agents you grant them to, injected into the sandbox when the run starts.
- Human approval gates. A bulk field update above the set size cannot save until a person approves it.
- Everything is code. The duplicate rules, the must-have fields and the meaning of "cold deal" are files in the repo. They have a version history and change through a reviewed change request, not a toggle in a dashboard.
The outcome
Our pipeline reports now look a lot more like real life, because the records underneath get looked after every night, not in the odd spring clean. When a change is big enough to matter, a person sees it before it happens. And the RevOps channel keeps a running diary of what changed and when.
More use cases
How we answer GDPR data-subject requests before the deadline
Our Machine agent checks each GDPR access or deletion request, finds the person's data across our product database, builds the report well within the SLA and hands it to legal before anything is sent.
How we keep our docs matching what the code does
Once a day a Machine agent reads every commit merged since its last pass, finds the doc pages those changes made wrong, rewrites them and sends you one pull request to review.
How we test every pull request without lifting a finger
Our Machine QA agent works with GitHub and our test environment. For each pull request it checks out the branch, runs the tests, tries the change for real, and posts what it found.
Put your first agentto work today.
- SSO
- RBAC
- Governed agents
- Full audit trail