Business process automation: if it repeats, it can be automated.
Business process automation is the practice of handing repetitive, rule-bound work — hand-offs, copy-paste between tools, month-end spreadsheets, CRM fields that go stale — to software that runs it on a schedule, logs every step, and asks a person only on the exceptions. Not every job needs a full AI system; most of it is orchestration, built on the tools you already run.
The eight things that get automated
Workflow automation
The steps a person clicks through every day, orchestrated: queueing, state, retries, notifications, a watchdog for anything stuck, and an approval gate where it matters.
Integrations
APIs, webhooks, and sync jobs that move data between your tools correctly, every time, with reconciliation built in. Linear, GitHub, Teams, Zendesk, HubSpot, and your own systems.
HubSpot & CRM automation
Enrichment, deduplication, routing, and follow-ups running on HubSpot automatically. High-confidence writes only, an audit trail for every field, a person on everything uncertain.
Dealership automation
Which scheduler or DMS does each rooftop run? Resolved rooftop by rooftop from manufacturer APIs, vendor fingerprints, and archives, written to the CRM only at high confidence.
Data enrichment
Pipelines that collect, verify, and enrich records from public sources and APIs, with precision gates before anything is written. Research that used to need a team, running on a schedule.
Finance & back-office ops
Reconciliation, invoice matching, scheduled reporting, and alerts that run themselves, with a human signing off on the exceptions. Month-end stops being a week.
n8n consultant
Production pipelines on n8n: queues, state, callbacks, watchdogs, and approval gates, self-hosted on your infrastructure. The orchestration behind the AI dev agent and bug fixer.
Test automation consultant
End-to-end suites in Playwright, Cypress, WebdriverIO, and Appium that run on every release, across web, TV, and mobile, from frameworks built for Eurisko and Visit Saudi.
AI agents on AWS Bedrock
Agents whose model calls go from your own AWS account to Amazon Bedrock, in your region, with no training on your data. The reference is the support agent running in production, resolving about 60% of tickets with no human touch.
AI automation for small business
The same systems at small-company scale: your tools, your accounts, one accountable engineer, a fixed price, and model usage on your own bill with a ceiling you set.
Which processes get automated first, and why
Volume times minutes gives hours a week. Weight it by what a mistake costs. Then draw one line: everything above it that is reversible runs on its own; everything irreversible keeps a person on the button. The Automation Map template is that rule on one page.
| PROCESS TYPE | TYPICAL SHAPE | WHERE THE PERSON STAYS | TOOLING |
|---|---|---|---|
| Hand-offs between tools | A form, a ticket, or a deal moves from one system to another and someone re-types it | Nowhere, once the mapping is tested; exceptions are logged | n8n, Zapier, or Make, chosen per job |
| Enrichment and research | Looking things up across sources and writing the answer into the CRM | On anything below the confidence gate | Pipelines with precision gates, written to HubSpot only when confident |
| Reconciliation and month-end | Matching, checking, reporting, chasing | Signing off the exceptions and anything that moves money | Scheduled jobs with an approval step |
| Orchestrated work with state | Queues, callbacks, retries, things that get stuck | On the irreversible step, such as a merge or a payment | n8n, self-hosted, with a watchdog |
| Judgment calls | Anything that needs context a machine does not have | Owns it; the map marks it "not yet" | None, on purpose |
n8n, Zapier, or Make: how the orchestrator gets chosen
The tool follows the job. Simple, stateless hand-offs owned by an ops team stay on Zapier or Make, where they are fast to build and easy to hand over. Anything with a queue, a callback, a loop, code, or sensitive data runs on n8n, self-hosted on the client's infrastructure, which is how the AI dev agent and bug fixer are orchestrated at the marina-software client. The reasoning is written out in n8n vs Zapier vs Make, and the production patterns on the n8n consultant page.
What "a human on the exceptions" means in practice
Every automation here has a fixed list of actions it may take, a confidence gate on anything uncertain, and an audit trail of what it did and why. When the gate is not met, the work lands with a person as a prepared case, not a blank ticket: the inputs found, the draft decision, and the reason the machine stopped. The person spends their time on judgment, and the machine on the repeats. The pattern is set out in the human-in-the-loop guide.
Mapped first, then built, then run
Automation fails when it starts from a tool instead of a job. Every engagement starts with the Automation Map: the repetitive work inside your company listed, ranked by hours saved and by risk, with the honest "not yet" written down where a machine would underperform.
Then the first workflow is built at fixed scope and fixed price, with logging, retries, a watchdog, and an approval gate agreed before anything runs. It goes live on your accounts and infrastructure. The second workflow is usually obvious by then.
Map the grind
Every repetitive job, who does it, how long it takes, what breaks when it is skipped.
Rank by return and risk
High volume and low risk runs on its own. Irreversible steps keep a human on the button.
Build the first one
Fixed scope, fixed price. Orchestration, state, notifications, and the approval gate agreed up front.
Run it, then the next
Live on your infrastructure with an audit trail. Hours saved are measured, not estimated.
Questions about business process automation
What is business process automation with AI?
Business process automation with AI is software that runs the repeatable steps of a business process on its own, using AI models where a step needs reading, classifying, or drafting, and a fixed rule for when a person must decide. It ranges from a form landing in the CRM correctly to an agent that resolves support tickets end to end.
How is this different from buying Zapier?
Zapier is a tool; this is the job done. The engagement starts by mapping every repetitive process in your company and ranking it by hours saved and risk, then builds the ones worth building, on Zapier where it fits and on n8n or code where it does not, with logging, retries, a watchdog, and an approval gate agreed before anything runs.
What does an automation engagement cost?
A fixed price for a fixed scope, written into a proposal before any build, after the free Automation Audit has produced the map. There is no price list because the same workflow costs very different amounts depending on integrations and volumes; the pricing page explains the model and what moves the number.
Which processes should never be fully automated?
Anything irreversible or judgment-based: moving money, changing a customer's account, deleting data, deciding a dispute. Those steps keep a person on the button by design, with the machine preparing the case. The Automation Map marks them "not yet" with the reason, which is often the most useful column on it.
Do you also automate customer support and QA?
Yes. Support tickets, release regression, engineering tickets, bug fixing, and documentation are handled by the AI workforce systems, six of which run in production at the marina-software client. This page covers the workflow, data, CRM, and finance automation around them.
Name the grind. We'll map the machine.
The free Automation Audit ranks every repetitive job in your company by return, in 30 minutes, with a written Automation Map. Yours to keep either way.