Fix the process before you automate it
Automation makes a process faster. It doesn't make it better. Automate a confusing, five-handoff process and you get a fast, confusing, five-handoff process that's harder to change.
The businesses that get real returns from AI do the boring part first: they map how work actually flows, cut the steps that don't matter and only then hand the rest to software.
Step 1: Pick the right process
Good first candidates share four traits:
- High volume. It happens daily or weekly, not twice a year.
- Rules you can write down. Most cases follow the same path.
- Digital inputs. Emails, forms, PDFs, spreadsheets or app data, not handwritten notes.
- Easy to measure. You can count hours, turnaround time or errors before and after.
Typical winners: order entry, invoice processing, quote drafting, scheduling, onboarding paperwork, weekly reporting and first replies to common questions.
Step 2: Map how it really works
Sit with the person who does the work and write down every step, including the workarounds nobody admits to. For each step, note:
- Who does it
- Which system it happens in
- How long it takes
- What goes wrong
You'll almost always find steps that exist only because two systems don't talk to each other, approvals nobody remembers asking for and data typed in twice.
Step 3: Simplify
Before any automation:
- Remove steps that don't change the outcome.
- Standardize inputs with forms or templates instead of free-text emails.
- Clarify decisions so the rules are written down, not in someone's head.
- Pick one system of record for each piece of data.
Sometimes this step alone saves more time than the automation will.
Step 4: Decide what's rules and what's judgment
Split the remaining steps into two groups.
- Rules-based steps (move this data, send this reminder, create this record) are a job for workflow automation: Power Automate, Zapier, Make or n8n.
- Judgment-light steps (read this document, draft this reply, summarize this thread, classify this request) are where AI models help.
- Judgment-heavy steps (approve this exception, handle this angry customer, sign off on this price) stay with a person, and the automation should hand them a clean summary.
Step 5: Build small, then extend
Automate one path end to end before you handle every exception. Run it alongside the old way for a week or two. Check the output. Then switch over and add the next case.
Step 6: Measure and maintain
Compare against your baseline after 30 days. Then assign an owner, because systems change, APIs update and processes drift. An automation nobody owns quietly breaks.
Common mistakes
- Starting with the hardest process because it's the most painful. Start with the one most likely to succeed.
- No baseline. If you didn't measure before, you can't prove it worked.
- No documentation. If only the builder understands it, you're dependent on the builder.
- Ignoring data access. Automations and AI tools need permissions. Grant the minimum they need, not admin rights.
Where to start
Our free automation assessment takes about three minutes. Tick the tasks your team repeats, set rough hours and your hourly cost, and it ranks your best automation candidates by payback.
Take the free automation ROI assessment
When you're ready, our automation sprints are fixed-fee, built on the tools you already own, documented and handed over. See consulting and automation services.