AI Automation for Sales: What to Set Up First, and What Stays Human

A five-person sales team should automate three things first: lead capture into one place, a drafted first reply that a person approves, and quote follow-up reminders. Keep with a person the pricing decision, any negotiation, a complaint, and the call where the buyer decides. Build in that order, one step a fortnight.

The order matters because each automation feeds the next. A first reply is only as good as the lead record behind it, and a quote follow-up only works if the quote status is recorded somewhere. Teams that start with a clever AI email writer and no tidy lead list end up polishing messages nobody can track.

Follow me on Instagram@sagnikteaches

The five-person team this order is built for

The set-up below assumes an illustrative print shop: the owner, two people who answer enquiries and prepare quotes, a production lead who occasionally joins calls, and a part-time administrator. Roughly 60 enquiries arrive a week from a website form, email, phone messages and walk-ins. About 40 become quotes. About a quarter of those are won. Every figure is an assumption; swap in your own counts from the last month before you build anything.

Connect on LinkedInSagnik Bhattacharya

Count where the hours go first. Ask each person to note, for a week, minutes spent on: copying enquiry details into a sheet, writing a first reply, building a quote, chasing quotes with no answer, and updating status. A common pattern is that building the quote is the smallest of these and the surrounding admin is the largest. That is what the three automations attack.

Subscribe on YouTube@sagnikteaches

Automation one: every lead lands in one place

This is the dull one, and the one that pays for the others. Whatever the source (web form, email address, messaging inbox), the automation adds a row to a single lead list with the same columns: date, name, contact, what they asked for, source, status. Use a shared spreadsheet or the free tier of your CRM. HubSpot's free tools cover up to two users, which is tight once four people share the lead list; Pipedrive's Lite plan is listed at $14 a seat a month billed annually. For a team this size a spreadsheet does the job until reporting matters.

Set-up steps

1. List every route an enquiry can arrive by. Include the ones you dislike, such as a personal phone.
2. Create the lead list with fixed columns and a status dropdown: New, Replied, Quoted, Won, Lost.
3. In your automation tool (Zapier and Make both work), build one automation per source: trigger on a new form entry or a new email in a label or folder, action adds a row.
4. For email, use a dedicated address or label. Zapier's Gmail trigger can be limited to a label or search keywords, which avoids pulling in every message.
5. Test with five real past enquiries, including a messy one with an attachment.

Free tiers constrain what is possible. Zapier's Free plan allows 100 tasks a month and two-step Zaps only, checking for new data every 15 minutes, and leaves out Filters, Formatter, webhooks and AI by Zapier. At 60 enquiries a week (about 260 a month) the shop needs a paid plan, because each new lead consumes at least one task, and automation two needs an AI step anyway. The Professional plan at 750 tasks, listed at $19.99 a month on annual billing ($29.99 month to month), leaves room, but check the current page.

What to check

After a week, compare the count of rows with the count of enquiries your team remembers. Any gap means a source you missed. Do not move on until the counts match within one or two.

Automation two: a drafted first reply that a person approves

Speed of first reply matters in most small-firm sales because buyers send the same enquiry to three suppliers. The aim is a reply within an hour in working hours, drafted by AI from the lead row and approved with one click.

The flow: a new row appears with status New, the automation sends the enquiry text plus your standard facts to an AI step, and the draft lands in the owner's drafts folder or a Slack-style message for approval. Nothing is sent automatically.

A first-reply prompt for the print shop

You are drafting a first reply for a print shop. Write 90 to 130 words,
plain and friendly, no exclamation marks.

Use only these facts:
- Standard turnaround: 5 working days; rush option: 2 working days, +25%
- Minimum order: 50 units
- We quote after seeing artwork or a description; we never give a price
  in the first reply

Customer enquiry:
{{enquiry_text}}

Do three things: confirm what they asked for in one sentence, ask the
2 questions we need to quote (quantity and deadline, unless they gave
them), and say a person will send the quote by end of next working day.
Do not promise anything not listed above. If the enquiry is a complaint
or something other than a new job, write only: FLAG FOR A PERSON.

Illustrative sample output, and what to fix

Given an enquiry reading "Need 200 flyers for an open day on the 14th, double sided", a typical draft says it can confirm 200 double-sided flyers, asks for the finished size and artwork, mentions standard turnaround, and says a quote follows by end of next working day. Three things to correct when you review early drafts: it may repeat the customer's date back as a promise ("we will deliver by the 14th") that the facts block does not support; it may add a paper stock nobody mentioned; and it may sound more enthusiastic than your shop does. Edit the prompt, not the draft, so the fix is permanent.

The "FLAG FOR A PERSON" line is the most valuable part. It gives the automation a way out when the enquiry is not what it expects. Test it by feeding in a complaint, a supplier pitch and a message in a language you do not serve.

Approval before sending, for the first month

For the first month a person reads every draft. Track, in a column, whether it went out unchanged, edited lightly or rewritten. If over about 80 per cent go unchanged for a month, you can consider sending acknowledgements automatically. Keep quote replies with a person. Running drafts in the background before anything reaches a customer is a technique worth knowing; how to pilot AI in shadow mode explains it.

Automation three: quote follow-up that nobody has to remember

Quotes go quiet. The shop sends 40 a week, and chasing each one by hand is the job people postpone. Automation three watches the status column: when a quote has been in Quoted for three working days with no reply logged, it drafts a short nudge and puts it in front of the person who owns that lead.

1. Add two columns to the lead list: quote sent date and quote value.
2. Build a daily scheduled automation that finds rows where status is Quoted and the quote date is three or more working days ago.
3. For each, draft a nudge using the original enquiry and quote value. Keep it to 60 to 90 words and give the customer an easy exit ("if the timing has changed, tell me and I will close this off").
4. Send the draft to the lead's owner for approval. After a second nudge at day eight, stop and mark Lost or Parked.

Cap the number of nudges at two. A third reads as pressure, and an automation without a cap will happily send ten. If you later move to a CRM with sequences, the same logic carries over; HubSpot against Pipedrive for AI sales automation compares what each offers once a spreadsheet stops being enough.

A before and after in numbers

Before: 40 quotes a week, about 200 minutes of chasing, and roughly half chased late or never. After: drafts appear every morning, approval takes about 5 minutes a day, which is 25 minutes a week. The saving is about 175 minutes a week, nearly three hours, and every quote gets exactly two touches. Whether that wins more jobs depends on the shop; measure the win rate on quotes that were nudged against those that were not before you claim a benefit.

Six sales steps a person keeps

The temptation after three working automations is to push further. Hold the line on these:

• The price. Quotes depend on margin, capacity and the customer's history. AI can fetch the standard rate card and do arithmetic, but chat assistants make arithmetic slips, so check every total. A person sets and signs off the number.
• Negotiation and objections. A buyer who says "your competitor is cheaper" is telling you something that needs listening, not a template.
• Complaints and anything emotional. A drafted reply to an unhappy customer risks sounding hollow. Use the FLAG rule to route these to a person immediately.
• The closing call. The conversation where the customer decides is where trust is won. Automation can book it and prepare notes; a person has it.
• Any promise about delivery or capacity. Only someone who can see the production schedule should make one.
• Relationships with your best customers. A top-ten account should never get a templated nudge.

A useful test for any step: if this goes wrong, who notices, and how fast? Where the answer is "the customer, a week later", keep a person in the loop.

Four more cases from other kinds of small team

A wedding planner: enquiry to consultation

Leads arrive with date, guest count and budget. The first automation logs them; the second drafts a reply that asks for any missing fields and offers two consultation slots; the third nudges after a consultation with no decision. What stays human: the consultation itself and any conversation about budget. See also whether AI can qualify wedding leads by date, budget and guest count.

A tutoring centre: trial lesson to enrolment

Parents enquire about a subject and year group. Drafted replies propose trial times from a list of tutor availability. After a trial lesson, an automation drafts a follow-up with the tutor's brief note attached. What stays human: any discussion of a child's needs, fees negotiation and concerns about progress.

A yoga studio: intro offer to membership

A new-student form triggers a welcome message and the intro offer. A nudge goes out after the offer's third class. What stays human: replies from anyone mentioning an injury or health condition, which should be flagged to the teacher.

The shared-mailbox slip in week three

Illustrative: a team connected the first-reply draft to a shared mailbox without a filter. Supplier invoices and newsletters arrived in the same mailbox and the automation dutifully drafted sales replies to them. The fix was two lines: restrict the trigger to the enquiry label, and add the FLAG FOR A PERSON rule. Always filter the trigger first.

Where each automation breaks, and the guard that catches it

Every automation has a typical way of going wrong. Knowing it in advance lets you add one cheap guard at build time instead of one expensive apology later.

Lead capture: duplicates and lost attachments

A customer who fills in the form and then emails the same request creates two rows. Add a duplicate check on the email address before the row is added, or accept duplicates and mark them in a weekly tidy. The second failure is attachments: artwork or briefs sent as files often fail to travel with the row. Store the file in a shared folder and put its link in the lead row, and test it with an actual PDF and an actual image.

First reply: invented details and tone drift

The drafting step will occasionally state something the facts block never said, such as a discount or a stock item. The guard is the "use only these facts" line, plus the weekly habit of reading ten drafts side by side. Tone drift is slower: after a few weeks of edits the prompt no longer matches how the team writes. Keep one example reply in the prompt that a person wrote and liked, and refresh it each quarter.

Quote follow-up: nudging the wrong person

The commonest embarrassment is a nudge to a customer who has already replied by phone or in person. The guard is a rule that anyone can set status to Won, Lost or Parked in two clicks, and a habit of doing so straight after a call. Add a daily list of nudges due, sent to the owner at 9am, so they can strike off any that are stale before approving.

Writing the rules down: a one-page sales automation charter

When a team has three automations, someone will eventually ask "who decided it does that?" Put the answers on a single page. Here is a filled-in illustration for the print shop; keep yours the same length.

SALES AUTOMATION CHARTER (print shop, illustrative)
1. LEAD CAPTURE   Owner: administrator. Sources: web form, enquiries
   mailbox, messaging inbox. Check: Monday count of rows vs enquiries.
2. FIRST REPLY    Owner: senior enquiry handler. AI drafts, a person
   approves. Never sends price. Flags complaints, supplier pitches,
   unknown languages. Prompt reviewed quarterly.
3. QUOTE NUDGE    Owner: owner. Day 3 and day 8 only. Approval needed.
   Stops when status is Won, Lost or Parked.
STAYS HUMAN: pricing, negotiation, complaints, delivery promises,
   top-ten accounts, closing calls.
PAUSE RULE: three errors in a week, or any customer complaint about an
   automated message, pauses that automation until reviewed.
CUSTOMER WORDING: "We use drafting tools; a person checks every reply."

The pause rule is worth copying exactly. It makes stopping an automation a normal, blameless act rather than an admission of failure, which is what keeps people reporting problems.

A seven-day plan for the first automation

If a six-week plan feels like a lot, here is the first week alone for lead capture. It assumes someone gives it about an hour a day.

1. Day 1. List every enquiry route and pull the last month's enquiries into the lead sheet by hand. The columns you end up needing are the columns you build.
2. Day 2. Create the status dropdown and agree what each status means. Write it on the sheet's first tab.
3. Day 3. Build the form-to-sheet automation and test it with three entries.
4. Day 4. Build the email-to-sheet automation using a label or search keywords, and test it with a message with an attachment.
5. Day 5. Turn both on for real, and keep manual entry running alongside.
6. Day 6. Compare counts: automation rows against manual rows.
7. Day 7. Fix gaps, then stop manual entry for sources that matched.

Running manual and automated in parallel for a couple of days sounds wasteful, but it is the cheapest test you will ever run. Only the comparison tells you whether the automation is trustworthy.

Questions to ask before adding any fourth automation

After the first three, the pressure to add more is constant. Run each new idea through four questions. Does it use data we already capture reliably? Would a mistake be visible to a customer? Can we describe the rule in two sentences? Is there a named owner? A "no" to the first or last means fix the foundation before building. A "yes" to the second means a human approval step is mandatory. Plenty of good fourth automations exist, such as internal alerts when a large quote goes quiet, a weekly pipeline summary for the owner or a prompt to ask satisfied customers for a review. They share a trait: low risk, internal first, and built on data you trust.

Cost and time for the whole set-up

A reasonable plan for the five-person team is one automation per fortnight: six weeks in total, with roughly four to eight hours of someone's time per automation, including testing. Software costs depend on tool choices; a Zapier or Make subscription at the entry paid tier, plus an AI step, is a few tens of dollars a month. Make bills in credits, and its pricing FAQ says most actions use one credit while some AI features use more, so read it before you size a plan. Zapier charges AI by Zapier per AI step and per tool call, priced by model tier, so test with real volume before you pick a task level.

If you would rather have help with the design, my AI implementation consultation covers choosing the first automation and handing over something the team can run. For general guidance on which steps need AI at all and which just need a rule, AI versus rule-based automation is the page to read next.

Four numbers that show the set-up is paying off

Track four numbers monthly: the median time from enquiry to first reply, the share of quotes that received a follow-up, the win rate, and the weekly hours spent on admin by the team. Expect the first two to improve straight away. The win rate takes a quarter to mean anything. If admin hours do not fall, find the step where people are still copying data by hand and automate or remove it.

Finally, set a review date. At six weeks, ask the team what the automations got wrong, which drafts they always rewrite and which step they would like next. That list, not a vendor's feature page, decides your next automation. It also keeps the team involved; getting staff to actually use AI tools covers how to do that without forcing it.

Sales automation questions from small teams

Should AI send follow-ups without approval?

Not at the start. Let AI draft and a person approve for the first month, then review how often the draft was sent unchanged. If it is nearly always unchanged for one simple message type, such as an acknowledgement, consider automatic sending for that type only.

Do I need a paid CRM to automate sales steps?

No. A shared spreadsheet or the free tier of a CRM can hold leads and statuses while an automation tool moves data. A paid CRM earns its place when more than two people touch the same leads or you need reporting you trust.

How many leads do I need before this is worth it?

Think in hours rather than leads. If the team spends more than about three hours a week on first replies, data entry and chasing, a basic set-up usually pays back within a couple of months. Below that, a saved reply template may be enough.

What if a customer realises they are talking to automation?

Tell them. Say plainly in the first message that a draft or assistant is involved where that is true, and offer a person on request. Honest framing costs little; discovered concealment costs trust.

Further reads

• Signs Your Business Is Ready to Automate Sales Follow-Ups With AI — Check the signs your follow-ups are ready to automate.
• AI Automation Setup for Small Businesses: DIY or Hire It Done? — Decide whether to build these yourself or hire help.
• How to Run Your First AI Pilot Project in a Small Business — Structure the first month as a proper pilot.
• Automation Audit: Find the Zaps and Scenarios Nobody Owns — Keep track of who owns each automation you build.
• Do I Need an AI Consultant? 8 Signs It's Time to Get Help — Signs that outside help will speed this up.
• AI Business Development Software: Do You Need It or a CRM? — A decision rule for small B2B firms: when a CRM plus a chat assistant is enough, and when dedicated AI prospecting software earns its fee.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Zapier pricing page and Gmail integration page, Make pricing FAQ, HubSpot CRM pricing page and Pipedrive pricing page, all checked November 2026.