Fake business data generator: a fictional company for agent tests

fake-business is an open-source generator of realistic, fully fictional small-business datasets (CRM, quotes, jobs, invoices, emails, calls and an event timeline) for testing AI agents and integrations.

Try fake-business

Choose an industry, a seed and a size, then generate. The generator runs in your browser and the same inputs always give the same business.

The interactive tool needs JavaScript. The same core runs locally: npx @teamshift/fake-business

Who it is for

Anyone who needs business data that makes sense end to end: agent developers writing evals, integration builders filling a CRM or accounting sandbox, and teams demoing a product without real customer records.

How it works

It simulates one business over a period of history (12 months by default, ending 2026-09-30). Leads become deals, deals get quotes, accepted quotes become jobs, jobs are invoiced and paid, and every step leaves messages, calls, tasks and timeline events that cite real record numbers, dates and amounts. Staff activity follows the company's business hours and timezone.

Eleven kinds of realistic mess are injected and labeled: duplicate contacts, quotes never followed up, missed calls never returned, stale deals, overdue invoices, an invoice on the wrong job, a reschedule that did not reach the technician, missing contact details, records that contradict each other, unmatched payments, and leads nobody contacted.

Emails use the reserved .example domain and phone numbers sit in the fictional 555-0100 to 555-0199 range, so a dataset is safe to publish.

Example

Home services, seed 42, small produces "Copper Kettle Plumbing & Air" with $19,010.00 invoiced, $298.00 outstanding and 11 labeled anomalies. On the command line the same business comes from npx @teamshift/fake-business --industry home-services --seed 42 --size small, and --format sql-sqlite, csv, hubspot or quickbooks export it for a database or an import wizard.

Limitations

Three industries: home services, a dental clinic and a marketing agency.
A fixed small-business schema. It does not generate arbitrary tables.
Output is stable for a given generator release; a new release can change it, so pin the version for long-lived evals.
This page offers JSON, NDJSON events and SQL downloads. CSV, HubSpot and QuickBooks exports are in the CLI.

FAQ

How do I test an AI agent without real customer data?

Generate a fictional business, load it into the tool your agent uses (a database, a CRM sandbox or the sandbox-mcp server), give the agent a task such as following up every stale quote, and compare what it did with the labeled anomalies. The same seed recreates the same scenario after every change.

Is the generated data really fictional?

Yes. Every dataset is marked synthetic, emails use the .example domain, phone numbers are in the 555-0100 to 555-0199 range, and company, town and street names are invented. People's names come from a list of common US names, so any match to a real person is coincidental.

Is the output reproducible?

Yes. The generator ships its own seeded random number generator and word lists, so the same industry, seed, size and dates produce byte-identical output on any machine for a given release.

How is this different from Faker?

Faker generates random values you assemble yourself. fake-business generates one complete business whose records link together, follow a causal timeline, and carry labeled problems you can score an agent against.