Sales automation: what to automate and what not to

Most small sales teams don’t lose deals because they can’t sell. They lose them in the gaps: the lead that waited until Monday, the quote nobody followed up on, the task that never got created after a call.

Sales automation closes those gaps by doing the routine steps for you, every time, without anyone needing to remember. This guide explains how automations are built, gives you ten recipes you can copy, and covers the guardrails and the work you should never hand to a rule.

What sales automation is

Sales automation is software that performs a sales step when something happens. A form comes in, so a text goes out. A deal moves to “Quote sent,” so a follow-up task appears three days later. An invoice goes overdue, so the owner gets a reminder.

The point isn’t to replace salespeople. It’s to give them back time. In Salesforce’s 2026 State of Sales survey of 4,050 sales professionals (fielded August to September 2025), the average seller said they spend just 40% of their time actually selling (Salesforce). That’s vendor research, and your numbers will differ, but anyone who has run a small team will recognize the pattern: data entry, reminders and status updates crowd out conversations.

Automation is good at exactly that kind of work. It’s bad at anything that needs judgment, and we’ll get to that.

Triggers, conditions and actions

Every automation, in any tool, is built from the same three parts.

Part	What it does	Examples
Trigger	The event that starts the automation	A lead is created, a form is submitted, a quote is accepted, an invoice becomes overdue, a call ends
Conditions	Filters that decide whether it should run for this record	Source is “Google Ads,” lead score is above 60, deal value is over $5,000, tag is “commercial,” message contains “price”
Actions	What happens, in order	Send a text or email, create a task, assign an owner, notify someone, move a stage, update a field, add a tag, wait, call a webhook

A useful way to write one down before you build it: “When [trigger], if [conditions], then [actions].” If you can’t fill in that sentence clearly, the automation isn’t ready.

Two details make a big difference in practice:

• Wait steps. “Wait two days, then create a task” is how you build follow-up sequences without a person tracking dates.
• Order of actions. Assign the owner before you create the task, or the task lands on nobody.

What to automate first

A good automation candidate passes three tests:

1. It’s repetitive. It happens many times a week in the same way.
2. It’s time-sensitive. Doing it late costs you something, like a cold lead or a forgotten invoice.
3. It’s identical every time. No judgment is needed about what to do, only that it gets done.

The first reply to a new lead passes all three. So does creating a follow-up task after a quote goes out, assigning new leads to the next rep, and nudging the owner when an invoice is overdue. Writing a custom proposal for a $40,000 remodel fails the third test, so a person should do it.

If you’re not sure where your leaks are, look at three numbers for the last month: how long new leads waited for a first reply, how many tasks went overdue and how many quotes got no follow-up within a week. The worst of the three is your first automation. For more on the first reply, see our speed-to-lead playbook.

10 automation recipes for small teams

These recipes cover the most common leaks for contractors, realtors, lawn care companies, agencies and closers. Adjust the timings to your sales cycle.

#	When	If	Then
1	A lead is created	It has a phone number and SMS consent	Send a short text, then create a high-priority call task for the owner due in one hour
2	A lead is created	It has no owner	Assign an owner by round robin and notify them
3	A form is submitted	Source is “Google Ads”	Add the tag “paid search” and notify the sales manager
4	An inbound message arrives	The message contains “price,” “quote” or “cost”	Add the tag “pricing question” and create a reply task for the owner
5	A deal moves to “Quote sent”	Deal value is over $2,000	Wait three days, then create a follow-up call task
6	A quote is accepted	Any	Create a “schedule kickoff” task and send a thank-you email
7	An invoice becomes overdue	Any	Send a polite reminder email and create a call task for the owner
8	A payment is recorded	The invoice is now paid in full	Send a thank-you message and tag the contact “customer” for review requests later
9	A lead is created	Lead score is above your “hot” threshold	Notify the owner and the manager immediately
10	A deal is marked lost	Lost reason is “timing”	Wait 90 days, then create a check-in task

A few notes on the recipes:

• Recipe 1 is the single highest-value automation for most small businesses. The text buys time; the call task makes sure a person follows through.
• Recipes 5 and 10 use wait steps. They replace the sticky notes and calendar reminders that tend to fail on busy weeks.
• Recipe 9 depends on a lead score you trust. If your scores aren’t explainable, fix that first; our lead scoring guide covers how.
• Recipe 2 works best with clear routing rules. See lead routing and round-robin assignment for the options.

Guardrails that keep automation safe

An automation does exactly what you told it to, at scale, at any hour. That’s the benefit and the risk. Put these guardrails in place before you switch anything on.

Consent on every send. An automated text is still a text. Only message people who have given the right consent for that channel, and honor STOP replies immediately. The safest setup checks consent at the moment of sending, not just when the lead was imported. Our guide to TCPA-friendly texting covers the basics.

Quiet hours. A “new lead” automation that fires at 11:40 p.m. shouldn’t text someone at 11:40 p.m. Messages should wait until the recipient’s morning, based on their time zone.

AI calls need extra care. If an automation places AI voice calls, know that in February 2024 the FCC declared that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act, which brings them under the same consent rules as other robocalls (FCC). This is general information, not legal advice; check with counsel before running outbound AI calls.

Run logs. You should be able to open any automation run and see each step: what triggered it, which conditions passed and what each action did. Without logs, you’re guessing when something goes wrong.

Test with yourself first. Create a test lead with your own phone number and email, and watch the automation run end to end before it touches a real customer.

One owner per automation. Someone on the team should be responsible for each automation, review its logs monthly and turn it off when the process changes.

Stop conditions. Follow-up sequences should stop when the lead replies, books or is marked lost. Few things annoy a customer more than a “just checking in” text an hour after they signed.

What to keep human

Automation handles the when. People handle the what and the why. Keep these in human hands:

• Real conversations. Discovery calls, objections and questions that need context. An automated text can say “thanks, when’s a good time to talk?”; it shouldn’t try to answer a detailed scope question.
• Pricing and exceptions. Discounts, payment plans and custom terms need judgment and accountability.
• Bad news. A delay, a price increase or a mistake on an invoice deserves a personal message.
• Disqualifying leads. A rule can flag a likely poor fit. A person should decide, especially when the lead’s fit is borderline.
• Anything irreversible. Deleting records, closing out accounts or sending money-related messages to a whole list should never happen silently.

A practical test: if you’d be embarrassed for the customer to learn that a rule sent this message without anyone looking, don’t automate it.

Common mistakes

• Automating a broken process. If nobody agrees on what “qualified” means, automating qualification just spreads the confusion faster. Fix the process on paper first.
• Too many automations too soon. Twenty overlapping rules can send three messages to the same lead in a minute. Add one at a time.
• Messages that sound like a machine. Write automated texts the way you’d text a customer: short, specific, signed with a name.
• No review. Processes change. An automation that made sense last spring may be texting leads about a promotion that ended.

How Exa helps

Exa’s automations use the same trigger, condition and action model described above. Conditions can check status, source, score, value, tags, custom fields and message content, and actions can send messages, create tasks, assign owners, notify, move stages, update fields, tag, wait, call webhooks and place AI calls. Every run is logged step by step, quiet hours are respected (messages wait instead of sending), and consent and opt-outs are enforced on every send. Ready-made recipes cover common starting points like speed-to-lead.

You can also ask Exa’s AI assistant to build an automation in plain English, for example “Create a task to call back every lead that replies.” It drafts the automation as an action card, and nothing is created until you confirm it. For the compliance side, see how Exa handles consent and audit logs on the security page.

Final thoughts

Good sales automation is boring. It sends the first reply, creates the task, assigns the owner and reminds you about the invoice, the same way every time, so your team can spend its hours talking to customers. Start with the one leak that costs you the most, put guardrails around it, and keep the human parts human.