Guide // AI Automation
What to automate first in an operations-heavy business (and what to keep human)
A four-question test for picking the first workflow, plus the work that should stay with people.
By Danylo ZavhorodniiPublished 7 min read
Short answer
Automate first the work that is frequent, follows clear rules, and mostly moves information between systems: enquiry intake, CRM updates, quote and order handoffs, reminders, and recurring reports. Keep people on pricing, exceptions, sensitive conversations, and anything published in your name. Start with one high-volume workflow, measure it, automate it end to end, and add failure alerts before building the next.
Why the first automation matters
Most teams do not struggle to find things to automate. They struggle to choose. A shared inbox, a CRM nobody keeps up to date, a weekly report rebuilt by hand, an automation account full of scenarios nobody remembers setting up: everything looks like a candidate. The first workflow you automate sets the pattern for the rest, so it is worth choosing on purpose rather than by whoever complains loudest.
The automation fit test
Score every candidate workflow against four questions. A workflow that is strong on the first three and low-risk on the fourth is a good first candidate.
| Question | Strong candidate | Weak candidate |
|---|---|---|
| How often does it happen? | Daily, or many times a week | A few times a year |
| Could you write the rules down? | The same steps every time, with clear inputs | Every case depends on context and judgment |
| Is it mostly moving information? | Copying details from emails or forms into a CRM, ERP, or spreadsheet | Deciding what to do, or creating something new |
| What does a mistake cost? | Easy to spot and cheap to fix | Expensive, public, legal, or safety-related |
Frequency matters because the time saved multiplies with volume. Rules matter because software needs them. Moving information matters because that is where manual work adds the least value and causes the most errors: nobody learns anything by retyping a phone number. The cost of a mistake does not rule a workflow out. It decides how much human review the workflow needs.
What usually belongs on the automate-first list
- Enquiry intake and routing. Form submissions, emails, and booking requests land in one place, get an owner, and get an acknowledgement, so nothing depends on someone noticing an email.
- CRM data entry and sync. Contacts, companies, and deals are created and de-duplicated automatically, and key fields stay in sync with the other tools your team uses.
- Quote, order, and RFQ handoffs. Requests arrive as structured records with the details your team needs, instead of free text spread across inboxes. Our guide to automating RFQ intake walks through this one in detail.
- Scheduling and reminders. Confirmations, follow-ups, and internal reminders fire from the dates and stages already in your systems.
- Recurring reports. The weekly spreadsheet rebuilt from three exports becomes a report that updates itself.
- First drafts of routine replies. Answers to common customer questions, or replies to reviews, drafted by AI from your own information and approved by a person before they go out.
Each of these is frequent, rule-based, and mostly about moving information. They are also where leads, orders, and follow-ups most often fall through the cracks, which makes the gain easy to see.
A worked example: enquiry intake
For a business that takes enquiries through a website form, email, and the phone, the first workflow often looks like this.
| Step | Before | After |
|---|---|---|
| Capture | Form submissions arrive as emails in a shared inbox; phone enquiries live in notebooks | Form, email, and logged phone enquiries all create a record in the CRM |
| Assignment | Whoever sees it first, or nobody | An owner is assigned by service, region, or customer type |
| Acknowledgement | Sent by hand, when someone remembers | Sent automatically, with a realistic next step |
| Follow-up | Depends on memory | A reminder fires if the enquiry is still open after an agreed time |
| Reporting | Counted by hand at the end of the month | A live view of volume, response time, and outcome by source |
None of this requires a new CRM. It requires the existing tools connected, clear rules for who owns what, and one person who is responsible for the workflow.
What to keep human
Automation is good at doing the same thing correctly a thousand times. It is poor at deciding what the right thing is when the situation is new. Keep people in charge of:
- Pricing, discounts, and exceptions. Rules can propose a price. A person should own anything outside the rules.
- Complaints and sensitive conversations. An upset customer should reach a person quickly, not a loop of automated replies.
- Anything published in your name. Review replies, social posts, and public statements can be drafted automatically, but a person approves them.
- Decisions about people. Hiring, performance, and HR matters stay with humans.
- Anything with legal, financial, or safety exposure. When a wrong answer is expensive, automation should prepare the decision, not make it.
- Relationships. The call that keeps a key account, and the conversation that opens a new one.
The middle ground: AI drafts, a person approves
Many useful workflows sit between fully automatic and fully manual. AI reads the incoming message, pulls the relevant facts from your systems, and prepares a draft: a reply, a quote summary, or a suggested next step. A person reviews it and sends it. The team keeps the judgment, and the automation removes the typing and the searching.
Two rules keep this pattern safe. First, ground the AI in your own information (product data, policies, past answers) instead of general knowledge, so it says it does not know rather than inventing an answer. We use the same approach for AI knowledge assistants. Second, make approval a real step, with a named owner and a time limit, not an optional checkbox.
How to avoid a pile of disconnected automations
The usual failure is a pile of small automations that nobody fully understands, each built to fix a single annoyance. When one breaks, nobody notices until a customer does. Design the system before building the pieces:
- Map the workflow end to end before automating any step: where the request starts, every system it touches, and where it ends.
- Choose one source of truth for each kind of record, usually the CRM for customers and deals, and make every automation read from and write to it.
- Give every workflow an owner who is told when it fails.
- Add monitoring and error alerts so failures surface in Slack, Teams, or email instead of failing silently.
- Document each workflow in plain language, next to the workflow itself.
- Keep the accounts in your company's name, with your own team as administrators, so the system can be handed over cleanly and you are never locked in to whoever built it.
How to start in five steps
- List the recurring workflows that touch customers, orders, or reporting.
- Measure what they cost today. For a week, note roughly how often each one happens and how long it takes. This is your baseline.
- Score each one with the fit test above and pick the single highest-scoring workflow.
- Automate it end to end, including the acknowledgement, the handoff, and the failure alert, rather than automating one step and leaving the rest manual.
- Review after a month. Compare against the baseline, fix whatever the team works around, and only then choose the next workflow.
If you would like a second opinion on where to start, tell us which tasks eat the most hours. We will map the workflow and scope the smallest automation that saves real time.
FAQ
Common questions
- What should a small business automate first?
- Usually enquiry intake: getting every form submission, email, and booking request into one place, assigned to an owner, with an automatic acknowledgement. It happens often, follows clear rules, and every missed enquiry is lost revenue.
- Should we use Zapier, Make, or n8n?
- All three can run the workflows in this guide. Zapier is the quickest to set up, Make handles more complex branching in a visual editor, and n8n can be self-hosted and suits workflows that need custom logic or tighter control over data. Choose on your team's comfort, your data requirements, and your volume rather than on feature lists.
- Can AI answer customer questions or reviews automatically?
- AI can draft good replies when it works from your own information. For anything public or sensitive, keep a person approving each reply before it is sent.
- How much time will automation save us?
- It depends on how often the workflow runs and how long it takes by hand today, which is why measuring the current process comes first. That baseline, not a vendor's estimate, tells you whether an automation worked.
- Who should own our automations after they are built?
- Your business. The accounts, workflows, and documentation should be in your company's name with your team as administrators, so you are never dependent on whoever built them.
Next step
Not sure where your first automation should go?
Send a short brief. We reply within one business day, map the workflow, and scope the smallest build that saves real time.