Automate the parts of a WhatsApp conversation that are the same every time: the quick reply, the questions that decide fit, the offer of a real time, and the reminder that asks the person to keep or release it. Leave the parts that change with judgment to a person: the exception, the negotiation, the upset customer, and any sentence that would need a licence, a diagnosis, or a promise you cannot template.
That split is the whole strategy. Teams get into trouble when they invert it. They automate the clever paragraph and still type the reminder by hand. The clever paragraph is where a business sounds wrong. The reminder is where a business loses the chair.
This is the long version of the split. The short note for Indian small businesses is what to automate first. The day-to-day rhythm is the sales day. Use this piece when you are deciding the boundary, especially if a vendor is offering to automate "everything."
If a message would embarrass you when a customer forwards it to a friend, it is not a candidate for automation. If a message is the same sentence you already send twenty times a day, it is.
The split
Every WhatsApp conversation on a small desk has two kinds of lines.
The first kind does not depend on who is working. "You have reached the clinic. Do you want a consultation or a follow-up?" "Here are two openings." "Your visit is tomorrow at 4:30. Reply YES to keep it." These lines are the business being available and consistent. They are also the lines that get skipped when the day is busy, which is exactly when customers are writing.
The second kind depends on a person who can stand behind the sentence. "We can do 8% off if you confirm today, and I will ask my manager." "I am sorry that visit felt rushed. I will move you to Thursday with the same clinician." "Based on what you have told me, this cover is the one I would walk you through." These lines are the business being accountable. An automation can introduce the person who will say them. It should not say them.
Write the split down in those terms, for your own desk, before you pick software. A one-page list with two columns is enough. Left column: lines we already repeat. Right column: lines only a named role may send. If a sentence could sit in either column, it belongs on the right until you have seen it go well fifty times on the left.
Automate the sentence you already send twenty times a day. Leave the sentence you would not want a customer to forward.
The rest of this article fills those columns for a typical local business in India, and then says what to do in the first month so the left column actually runs.
What automation is good at
Automation is good at speed, memory, and not getting bored.
Speed means the 8:50pm message gets a reply before the person opens another chat. Not a perfect reply. A true one: the business name, the next question, and when a person will join if a person is required. Waiting until morning to say hello is a choice to be second.
Memory means the answers land in a row while they are being given. Name, ask, constraint, status, next step, meeting. The sheet guide is the minimum memory. Without it, the night reply is a courtesy and the morning is archaeology.
Not getting bored means the third qualifying question is asked the same way at the end of a long day as at the start. People skip steps when they are tired. The skip is how a hot lead and a not-now lead look identical in the inbox. A script does not skip because it is tired. That is its only virtue, and it is a large one.
Automation is also good at the clock. It can send the day-before reminder because the booking exists, not because someone remembered. It can refuse to offer a time that is not on the calendar. It can stop after one follow-up because the rule said one, even when a colleague feels like "just one more" would be friendly. Friendly, repeated, is how numbers get blocked. The cadence is in follow-ups without spam.
None of that requires the automation to sound like a person pretending to be your best salesperson. A clear script in your real tone outperforms a chatty imitation that cannot keep a promise. Customers forgive a simple bot. They do not forgive a confident wrong time.
What a person should keep
Keep anything where the next sentence could commit the business to money, risk, or care.
Negotiation stays human. A discount, a waiver, a payment plan, a custom package. The chat can collect the constraint that makes the negotiation possible, such as a budget or a date, and then name the colleague who decides. It cannot offer the concession.
Custom work stays human. An interior layout, a travel plan that is not a standard package, a coaching recommendation for a specific student, a car deal with an exchange that needs to be seen. The chat can book the conversation and record the brief. The work product is the person's.
Upset customers stay human, immediately. If the message is a complaint, the script's only job is to acknowledge that a named person will reply, and to make sure that person is actually notified. Do not send a qualification menu to someone who is angry. It reads as indifference, because it is.
Exceptions stay human. The visit type that is not on the list. The locality you do not usually serve but might. The batch that is full except for a seat the counsellor can choose to open. Exceptions are where your judgment is the product. Automating them means you have decided the exception is no longer exceptional, in which case it belongs on the list and is no longer an exception.
Trust stays human when the customer asks for it directly. "Are you a real person?" deserves a straight answer. If a person is about to join, say so and say when. If the night hours are automated, say that too. Pretending otherwise is a short-term trick that makes the handoff feel like a breach.
Month one, and what can wait
Month one has four movements, in order. Do not start at movement four because it looks more like "automation" in a demo.
The first reply, within seconds, in the words your desk already uses. It names the business and asks the first question. It does not attach a catalogue.
Three questions. What they want, the constraint, and when they can talk or visit. Then a status: hot, warm, or not now. The qualification guide is the method. Three is the maximum until you can show that a fourth question changes a decision you actually make.
The booking, when they are hot and the calendar has a real opening. Confirmation in one message. The same time on Google Calendar. The long version is how to book appointments on WhatsApp. If you do not book visits, the equivalent is a scheduled call with a named owner. A "we will call you" with no time is not a booking.
The reminder, day-before, asking them to confirm or release. Template written in advance, utility rather than marketing, opt-in already true because they booked on this chat. Details are in message templates and the reminder guide.
What waits: review requests, loyalty messages, a second WhatsApp number, CRM integrations, and any report you do not already look at every morning. Broadcasts and custom AI training are on the Professional plan, but they are not the first week. Month one is the four movements, on Growth or Professional, until those are dull.
Also wait on automating the follow-up beyond a single, specific nudge to warm rows. Add a second nudge only after you have watched the first one for a few weeks and can say who replied, who asked to stop, and whether the sheet was updated when they did. If you cannot say that, you are not ready to send more.
Industry lines you do not cross
The four movements are safe only if the script stays inside the work of scheduling and qualification. A few desks need the boundary stated in the script itself, not assumed.
Clinics and dentists book listed visits. They do not interpret symptoms, rank urgency, suggest a medicine, or promise that a visit will fix something. If a person describes an emergency, the script points to the human path the clinic already uses. It does not become the path. Read the clinic and dentist playbooks before you write a single clinical-sounding line. If a sentence would be odd coming from the receptionist, it does not belong in the automation either. The receptionist is the ceiling, not the floor.
Insurance advisors book a conversation about a product the person asked for. They collect the facts the advisor already asks before a call, such as what the person said they want to cover and when they can talk. They do not recommend a policy, compare premiums as advice, or tell someone they are approved. The insurance playbook is deliberately narrow for that reason. Narrow is what keeps the chat useful.
Coaching institutes can ask which course, which batch timing, and who the class is for, and they can book a counselling slot or a demo. They should not promise a score, a rank, or that a particular student will be admitted. Gyms can book a trial and a tour. They should not prescribe a workout or a diet in the first reply. Salons can book a listed service with a named stylist. They should not diagnose. Property, car, travel, and interior desks can qualify and book a viewing or a call. They should not invent a price that the human has not authorised.
If you sell in a regulated space and you are unsure whether a line is advice, leave it on the human side. The cost of a slower chat is smaller than the cost of an automated claim you did not mean to make.
Follow-ups that stay welcome
A follow-up is welcome when it continues a job the person started. They asked about a course and did not pick a time. They booked and need to confirm. They asked for a brochure you said you would send. The message refers to that job, offers one next action, and has a way to say stop.
A follow-up is unwelcome when it starts a new job they did not ask for, repeats itself, or arrives after they have gone quiet more than once. The short guide is the rulebook. The practical test, before you automate a nudge, is whether a thoughtful colleague would send this exact text to this exact row. If the answer is "only if we add a personal line," it is not ready to automate. Either make the text specific enough that the personal line is unnecessary, or let the owner send it by hand.
Warm rows get one prepared follow-up. Hot rows with a meeting get the reminder, which is not really a follow-up. It is part of the booking. Not-now rows get nothing until they write again, or until a genuinely new fact exists, such as a batch date they asked you to tell them about. "We have not heard from you" is not a new fact.
Honour stop on the same day, in the sheet, not only in someone’s memory. The next template send should read that mark. A stop that depends on a colleague remembering is a stop you will eventually miss.
How you know it is working
Do not wait for a dashboard to tell you. For a small team the evidence is in the morning, and it is qualitative on purpose. This article will not offer a percentage to chase.
Open the sheet before the inbox. If yesterday’s enquiries already have an ask, a constraint, and a status, the night reply did its job. If you are reconstructing those from the chat, it did not, even if the chat reads politely.
Open the calendar next to a hot row. If the times match the confirmation messages, the booking is real. If they do not, stop adding features and repair the write. A reminder on top of a wrong time is a louder mistake.
Look at what the morning person does for the first half hour. If they are calling hot rows and sending the one warm follow-up, the priority is visible. If they are greeting new people while yesterday’s hot rows sit untouched, the sheet is not the start of the day yet. Fix the habit before you fix the script.
Listen for the question you hope not to hear: "Sorry, what did you need again?" Once in a while is a human day. Every day means the handoff is empty. The shared number piece is the repair.
Count only what you will change your behaviour over. For example, how many released slots were offered to a waitlist this week, or how many not-now rows were correctly left alone. Skip counts that flatter, such as messages sent. A large number of messages sent can mean the follow-up rule is being ignored.
A 30-day rollout
Week one is the script and the sheet, with a person still watching every send. Write the first reply and the three questions. Run them on live chats. Each evening, fix one awkward line. Do not add a question. Do connect the row so status is not a mental note. By Friday you should be able to sort the sheet.
Week two is the booking. Offer real openings. Send the one-message confirmation. Check the calendar twice a day for disagreements and fix the cause, not just the event. Write the reminder template even if you are not sending it automatically yet. Say the words out loud. If they sound like a poster, rewrite them until they sound like the desk.
Week three turns the reminder on for bookings made in the chat, and turns on the single warm follow-up. Watch who replies, who ignores, and who asks you to stop. Update the sheet when they do. If the reminder is not getting replies, the slot-release rule is probably missing or not believed. Add the consequence only if you will carry it out.
Week four is the week you stop babysitting and start reviewing. The owner rule is in force if more than one person sends. Unowned rows are cleared each morning. You look at the two columns again and move a line left only if it was repeated, without incident, all month. You do not celebrate by launching a broadcast. You celebrate by having a quieter inbox and a diary you trust.
If a week goes badly, repeat it. The rollout is not a deadline. A team that reaches week four with a true sheet has done the hard part. A team that reaches week four with a catalogue bot and an empty calendar has bought activity.
What it costs to start
ConverseWith’s Growth plan is ₹2,999 a month. It includes one WhatsApp number, 500 AI replies a month, intent capture, and a one-time basic AI training. Professional is ₹7,999 a month: three numbers, unlimited AI replies, lead scoring, customer analysis, auto follow-ups, AI template generation, broadcasts, knowledge feeding, and custom AI training with more context and dynamic updates. Enterprise is a custom scope for high-volume teams that need CRM integrations and a dedicated manager. The first 14 days are free, with no credit card required to start, and most teams are live in about ten minutes for the starter setup.
You still pay WhatsApp’s own conversation charges where they apply. Those rates change, so read them when you turn templates on, and do not treat a software price as an all-in message price.
The cheaper mistake is not picking Growth when you later need Professional. The expensive mistake is automating the right column first. Start with the four movements. Keep judgment in a person’s hands. The guides are the short versions of each movement, and the playbooks are the same movements with the nouns of a specific trade.
What should a business automate first on WhatsApp?+
The first reply, three qualifying questions, the booking, and the reminder. Those four remove waiting and forgetting. Catalogues, broadcasts, and a custom CRM can wait until the four are steady.
What should a person still do in a WhatsApp sales chat?+
Anything that needs judgment. Discounts, custom quotes, upset customers, and promises only a named colleague can keep. Clinics should not let a chat interpret symptoms. Insurance chats should not recommend a policy.
How do you know WhatsApp automation is working without a vanity metric?+
The sheet has a status by the time you open it, the calendar matches the chat, and the morning is spent on hot rows rather than on reconstructing who wanted what. If those three are true for a week, it is working.
How much does it cost to start?+
ConverseWith’s Growth plan is ₹2,999 a month for one WhatsApp number, 500 AI replies, and a one-time basic AI training. Professional is ₹7,999 a month for three numbers, unlimited AI replies, lead scoring, auto follow-ups, and custom AI training with knowledge feeding and dynamic updates. The first 14 days are free.
Automate the four lines you already repeat
Reply, questions, booking, reminder. The first 14 days are free.
Start free