Small and mid-size companies rarely suffer from a shortage of software. They suffer from people retyping the same facts into three systems, chasing approvals in email, and finding out on Friday that a lead sat in a shared inbox since Tuesday. Business process automation is the work of taking one of those loops and making the happy path happen without a human courier.
The fifteen workflows below are a menu, grouped so an owner or an operations lead can recognize their own week. They are a starting set for a first conversation. Your first project should come from the scoring section, using your hours and your risk, rather than from whichever example sounds most impressive in a demo.
If you want the shorter version of where to look for manual work, read Automation for Operations: Where to Start. If you are deciding between a connector tool and a screen your staff will live in, read Workflow Automation vs. Internal Tools.
Write the path down before you connect anything
If you cannot describe the happy path in about ten lines, including who is allowed to make an exception, you are still defining the process. Connecting tools at that point freezes the confusion in place and gives it a log file. Spend an afternoon with the person who actually does the work. Watch the exceptions. The exceptions are the product.
A useful write-up names the trigger, the fields that must exist, the system of record, the person who still has to approve, and what “done” looks like. It also names the failure you can tolerate. A duplicate CRM lead you can merge on Monday is a different failure from a payment you cannot claw back.
1. Sales lead routing
Sales. A form submission or a monitored inbox creates one CRM record and assigns it to someone who can work it today, by territory, product line, or a round-robin of reps who are actually in. The record carries the source page and the fields the person filled in. If it sits untouched for a window you set—often a few business hours—it moves to the next person, and every hop is logged. The loop you are removing is the lead that spent a weekend in a shared mailbox.
2. Quote generation
Sales. An approved rate card plus a short form—line items, a discount inside a band, and standard terms—produces a PDF and a CRM opportunity that show the same numbers. A discount above the band routes to a manager before the PDF goes out. The point is that the quote finance will invoice matches the quote the customer signed. Two calculators, one in a spreadsheet and one in the CRM, are how margin disappears without anyone making a decision.
3. Client onboarding
Delivery. A signed agreement creates the workspace from a template, sends a welcome note that states the real next step, and opens internal tasks with owners: access, a billing profile, a kickoff agenda, the files you always ask for. The client sees a checklist. Your team sees the same checklist. Nothing starts from a blank email that says “excited to kick off” and then waits three days for someone to remember the folder structure.
4. Invoice processing
Finance. A vendor invoice arrives by email or upload. The workflow pulls vendor, invoice number, date, total, and a purchase-order number when one is present, then matches it to an open PO or flags the miss. It creates a draft bill in the accounting system and notifies the person who should look. It does not pay the vendor. Payment stays a separate, human action until the match rate has been boringly right for a while.
5. Accounts payable approvals
Finance. A bill routes by amount and department. The approver gets the packet—the invoice, the match result, and the coding—rather than a forwarded email with three attachments. A reminder goes out on the second business day. An escalation goes out on the fifth if you have a policy that says so. The decision and the timestamp are stored on the record. The person who releases money is still a person.
6. Support triage
Support. A new ticket is classified, attached to the customer record, and placed in a queue a human owns. A suggested reply or a macro can sit in the draft field. A person sends it. Auto-close belongs only to a narrow class you can define exactly, such as a password-reset confirmation the product already verified. The win is a shorter time-to-first-response and a queue that is sorted before the morning standup, with a human still accountable for what the customer reads.
7. Appointment reminders
Operations. Reminders go out on a schedule you choose—often two days before and a couple of hours before—by the channel the customer agreed to. The message includes a reschedule or cancel link. Cancelled and already-confirmed appointments drop off the list. Times use the location's zone. The reminder that fires after someone cancelled is the one that trains customers to ignore the next one.
8. Scheduled reporting
Leadership. A short Monday snapshot goes to the people who will actually read it: pipeline created, invoices past due, tickets past their response target, inventory under the reorder point. One page, pulled from the systems of record, with links back to the records. The email is a prompt. The dashboard or the ledger remains the source of truth. A twelve-tab workbook emailed “just in case” is how reporting becomes another chore.
9. HR onboarding
People. A signed offer opens a checklist with owners and due dates: email and core accounts, payroll setup, equipment, a day-one agenda, policy acknowledgements. The automation creates the tasks and nags the owners. IT still reviews access, because a template that grants every new hire the same admin rights is an incident waiting on a start date. Offboarding deserves the same pattern in reverse: a departure date that closes accounts on a list, not a hope that someone remembers Slack.
10. Inventory alerts
Operations. When on-hand quantity crosses a reorder point you set from real usage, purchasing gets the SKU, the recent run rate, the current on-hand count, and the preferred vendor. That is the alert. An automatic purchase order can wait until the alert has been right for a few months and the vendor will accept an electronic PO you can void. Alerting on a stale reorder point is how you fill a shelf with the wrong part.
11. CRM data hygiene
Sales operations. A nightly job finds likely duplicates on email and phone, leads with no owner, and records with no activity inside a window you choose. It writes a queue or a task. A person merges. Automatic merges feel efficient until they collapse two real companies that share a parent domain, or erase the history on the record you meant to keep. Hygiene is a queue with a weekly owner, not a silent cleanup.
12. Document extraction
Operations or finance. A model reads a PDF or a photo—an invoice, a receipt, an application—and proposes fields: vendor, date, total, and whatever else the downstream system requires. The screen shows the source document beside the fields. Below a confidence threshold you choose, the record cannot proceed until a person edits or confirms it. You store the original file and log who confirmed the fields. Unattended posting to the general ledger is a later decision, after you have counted the corrections and the corrections are rare.
13. Renewal and contract reminders
Sales or finance. At 90, 60, and 30 days before a renewal, the owner gets the contract value, the renewal date, and a pointer to whatever usage or satisfaction signal you actually have. A reminder with no owner is calendar noise. A reminder that includes the last invoice and the open issues gives the owner a reason to call this week instead of the week the contract auto-renews into a complaint.
14. Expense approvals
Finance. A receipt photo becomes a draft expense: merchant, date, amount, and a project or trip code when the employee supplies one. In-policy amounts route to the manager the policy names. Out-of-policy amounts stop for a reason code. The reimbursement export runs after approval. Extracting the merchant from a crumpled photo is a good job for a model. Deciding that the dinner was allowable is still a manager's job.
15. Customer status updates
Customer operations. When a job, order, or request changes state in the system of record, the customer gets a plain update: the new state, and the next step you can actually promise. The text comes from the state, written by you in advance. A model improvising a delivery date or a decision you have not made will sound confident and be wrong. Status mail reduces the “any update?” tickets. It has to be dull and accurate to do that.
How to pick the first one
Score candidates as hours saved each time, multiplied by how often the work happens in a month. Then apply risk as a gate, not as a vibe. If a wrong run charges a customer, pays a vendor, deletes or merges a record, or tells a customer something you cannot take back, it fails the gate. Automate the preparation—the packet, the draft, the reminder—and leave the commit with a person. High hours times high frequency times a failure you can undo is the first project.
Use your own numbers. Here is the arithmetic, with round figures you should replace. Re-keying web leads into a CRM: 10 minutes, 30 times a month, about five hours a month. A bad run creates a duplicate you can merge. That clears the gate. Releasing vendor payments: 15 minutes, 20 times a month, also about five hours, and a bad run pays the wrong amount. Same hours, failed gate. Assemble the packet and keep the payment click human. An annual narrative report: eight hours, once a year. Low frequency. Leave it until the weekly work is quiet.
Pick one workflow, run it for a month, and count two things in a shared note: hours your staff says they got back, and exceptions a person had to fix. Expand when both numbers are boring. A second workflow started in week two, before the first one is trusted, is how automation projects become a graveyard of half-finished zaps.
Zapier or Make, versus a custom build
A connector product such as Zapier or Make fits when the happy path is “a record in A creates or updates a record in B,” you are joining two or three products, someone on your staff can open the run history and read the payload, your security expectations allow that vendor to see the data, the volume fits the plan, and a missed run is obvious the same day. Lead routing into a CRM, a simple reminder, and a Monday email pulled from a spreadsheet often live happily there.
A custom workflow or a small internal tool fits when you need permissions, a queue of exceptions, an audit log an operator can read without exporting JSON, branching that has outgrown a scenario nobody will edit, or a volume and a per-task price that have become the cost of the process. It also fits when the person doing the work needs a screen—approve, reject, correct the extracted total—rather than a notification.
Plenty of companies use both. The connector handles a side path. The screen handles the job someone does all afternoon. That split is the whole of the internal-tools question, and it is why a first automation sometimes wants a modest custom slice instead of a larger zap. When the workflow is the operation, look at workflow automation and, if the work is local to us, custom software development in San Diego.
Where AI helps, and where a rule or a person should stay
AI helps when the input is messy and the output is a draft a person will check. That is document extraction on invoices and receipts, a suggested support category, a draft reply sitting unsent, and a duplicate suggestion in the CRM. The pattern is the same in each case: propose, show the source, require a confirmation when confidence is low, and keep the original.
A rule is the better tool when the input is already structured. A ZIP code and a territory table should route the lead. An amount threshold should pick the approver. A reorder point should send the inventory alert. A reminder schedule should send the appointment text. Putting a model in the middle of a table lookup adds cost and a new way to be wrong.
Keep a person on the commit whenever the action moves money, destroys or merges history, or makes a customer promise—a date, a price, a coverage-style decision—that you cannot unwind with an apology. AI is a poor system of record. The CRM, the ledger, and the ticket system remain the record. The model proposes fields and drafts. Staff confirm. If you want a longer treatment of that boundary, the practical test is whether you can show an auditor or a customer the source and the name of the person who accepted it.
Bring one workflow, not a transformation program
Nightcoders builds workflow automation and AI automation from San Diego for companies that can point at a loop and describe the exception. Start with workflow automation, the regional page for AI automation in San Diego, or the broader AI automation practice, depending on whether you need routing and approvals, a model in the loop, or both.
When you are ready, contact Nightcoders with the trigger, the systems involved, and what a wrong run would do. We will tell you if a connector is enough, if the first slice should be custom, and if the workflow you named should wait until a safer one is running.
Put automation into production
Nightcoders builds AI & automation, integrations, and internal tools from San Diego for Southern California and remote teams. See delivery context in case studies and the full portfolio.
Bring a concrete workflow or product surface to contact—we will be direct about fit, risk, and a sensible first slice.