Modern automation platforms make it possible to connect almost anything to anything. That is exactly the problem: when everything can be automated, teams often automate the wrong things first—or build fragile workflows nobody understands six months later.
The hard part isn’t the tooling. It’s choosing the workflow worth automating. Here’s the simple filter we use with clients.
The three-question filter
For every candidate process, ask three questions:
- Friction: how painful is this today? Is it slow, error-prone, or a frequent source of complaints?
- Frequency: how often does it happen? Daily tasks beat quarterly ones.
- Impact: what changes for the business if this runs perfectly—faster responses to leads, fewer billing errors, happier customers?
A process that is painful, frequent and important is a strong candidate. A process that scores low on all three probably isn’t worth the effort yet, however clever the automation would be.
Map the work before you touch a tool
Before building anything, write down how the process actually works today—not how the documentation says it works. Sit with the people who do it. Capture each step, the information that goes in and out, the hand-offs between people and, crucially, the exceptions.
Exceptions are where automations break. If one in ten invoices needs a manual check, design for that from the start rather than discovering it in production.
Score your candidates
Give each process a score from 1 to 5 for friction, frequency and impact, then multiply them. The result isn’t scientific, but it quickly turns a long wish-list into a ranked shortlist—and makes the trade-offs visible to everyone.
From that shortlist, start with something high-scoring but low-risk: a workflow where a mistake is easy to spot and fix. Early wins build confidence and teach your team how automation fits into their day.
Good first automations
Some workflows show up again and again as great starting points:
- New enquiries from website forms flowing straight into your CRM, with the right person notified
- Documents such as proposals or onboarding packs generated from templates
- Recurring reports assembled automatically from the tools you already use
- Customer or employee onboarding checklists that trigger the next step on their own
- Keeping records in sync between two systems that don’t talk to each other
Where AI fits—and where it doesn’t
AI is excellent at the messy middle of many workflows: summarising long messages, classifying requests, extracting details from documents and drafting first responses. It is far less suited to decisions with real consequences that nobody reviews.
Automate the predictable middle—keep people at the edges where judgment matters.
A good pattern is “AI drafts, a person approves”, with every action logged so you can see what happened and why.
Build for the day it breaks
Every integration eventually fails—an API changes, a password expires, a field gets renamed. Reliable automations include monitoring and alerts, sensible retries, clear documentation and a named owner. That’s the difference between a workflow that saves time for years and one that quietly stops working.
Key takeaways
- Prioritise by friction, frequency and impact—not by tool.
- Map the real process, including exceptions, before building.
- Start with high-value, low-risk workflows for quick wins.
- Use AI for drafting and sorting; keep people on key decisions.
- Monitor every automation and give it an owner.
