In this article
A good WhatsApp pipeline has five to seven stages, each defined by something the customer did in the chat, not by something your team feels. "Sent price" is a stage. "Warm" is a mood. Every card needs one entry rule, one owner, one exit signal and a maximum age, and every stage move should come from one of three sources: a person dragging the card, the AI agent calling a tool, or a flow node.
This guide gives four example stage sets, a table of what moves cards and how, the SLA and ageing rules that stop leads sitting in the chat, and one worked reporting example. It is about design. For the feature overview, see WhatsApp CRM pipeline; for the AI side, see the AI sales agent guide.
How many stages should a WhatsApp pipeline have?
Five to seven, plus Won and Lost. Fewer and you cannot see where leads stall; more and agents stop updating them. Four design rules:
- Name stages after observable events. "Viewing booked" can be checked in the chat. "Interested" cannot.
- One stage, one meaning. If agents argue about which stage a card belongs in, merge or split it.
- Put the handover in the stage list. The point where the AI hands to a human should be its own stage ("Needs human"), so nothing is stuck between them.
- Name your outcomes "Won" and "Lost". WhatsGrow AI treats stages named Won, Closed Won, Lost or Closed Lost as outcomes: dragging a card into one sets the deal status, stamps the won or lost date and feeds win-rate reporting and the Meta purchase event, if you connected Meta's Conversions API. A lost reason can be recorded on the deal.
The board shows each stage's pipeline value, the deal value on every card, who it is assigned to, open and overdue task counts, and average response time. You can add, insert and drag-reorder stages. One warning: deleting a stage asks whether to delete all its leads too. To retire a stage, move its cards first, or rename it instead.
What stage sets work for real estate, clinics, education and B2B services?
Example sets (starting points; adjust the names to your process):
| Industry | Stages | Entry rule for the first two |
|---|---|---|
| Real estate | New enquiry, Qualified, Viewing booked, Viewed, Offer made, Won, Lost | Qualified = budget, area and timeline captured |
| Clinic | New enquiry, Qualified, Appointment booked, Attended, Treatment accepted, Won, Lost | Qualified = service needed and preferred time known |
| Education | New enquiry, Qualified, Counselling booked, Application started, Fee paid, Won, Lost | Qualified = course, intake and eligibility confirmed |
| B2B services | New enquiry, Qualified, Call booked, Proposal sent, Negotiation, Won, Lost | Qualified = need, budget range and decision maker known |
Notice the pattern. Stage two is always "Qualified" with explicit fields behind it, stage three is always a booked event with a date, and the last working stage is a commitment (offer, treatment accepted, fee, negotiation). Because the booking is a calendar fact, "booked" cards can be checked against reality instead of opinion.
The old habit of tracking "Follow up 1, Follow up 2, Follow up 3" as stages is worth avoiding. Follow-ups are tasks on a card, not places the card lives.
What moves a card between stages?
| Mover | How | Best for | Watch out |
|---|---|---|---|
| A person | Drag the card, or edit the lead | Judgement moves: offer made, negotiation, lost | Updates slip when agents are busy |
| AI agent | Calls the move-pipeline-stage tool with a stage name | Moves backed by chat facts: qualified, booking requested | Needs the tool switched on, exact stage names |
| Flow | A move-stage node, by stage name or id | Rule-based moves: form submitted, template button tapped | Creates the lead if the contact has none |
Three practical points:
- Stage names must match exactly. The AI looks up the stage by the name in its instructions. The AI page checks for this: if your instructions mention a stage that does not exist, it tells you and suggests the closest name. Renaming a stage means updating the instructions and any flows.
- Let the AI move forward, humans move backward and outward. AI moving "New enquiry" to "Qualified" when the fields are filled is reliable. A human should decide "Lost".
- Every automatic move should be explainable from the chat. If you cannot point to the message that justified it, tighten the rule.
Leads also arrive on the board from connected lead sources and your WhatsApp enquiry flow, so decide up front which stage a new lead lands in: normally "New enquiry", never "Qualified".
What SLA and ageing rules keep leads from stalling?
Two clocks matter: how fast the first reply goes out, and how long a card can sit still.
First response. WhatsGrow AI's speed-to-lead SLA sets a response deadline per new lead (default 15 minutes, configurable from 1 minute to 24 hours), alerts managers and admins when it is missed, and can reassign the lead to another agent up to a set number of times (default 2). This SLA engine is part of the real-estate pack today. For other industries, the AI agent is your first-response SLA: reply in seconds, then hand over at the right moment.
Ageing. The board does not show a built-in "days in stage" clock, so run the rule through tasks, which show on each card with an overdue count. Starting values to adjust:
| Stage | Max age without activity | Action when exceeded |
|---|---|---|
| New enquiry | 15 minutes to first reply | Escalate to a manager |
| Qualified | 2 days | Task: send options or booking link |
| Booked (viewing, appointment, call) | Day before and day of | Reminder template; task for owner |
| Offer or proposal | 3 days | Task: call the customer |
| Negotiation | 7 days | Manager review, then Lost or discount decision |
Add WhatsApp's own limit to the thinking: inside 24 hours of the customer's last message you can send free-form replies; after that you need an approved template, and since 1 October 2026 a marketing template to an Indian number costs ₹0.8631 (cost breakdown). A card that waits three days for a reply now costs real money to reopen, which is another reason to use the ageing table.
How do you report on a WhatsApp pipeline?
Each week, count the cards that entered each stage and divide by the previous stage. Example (hypothetical numbers) for a clinic with 200 new enquiries in a month:
| Stage | Cards entering | Conversion from previous stage |
|---|---|---|
| New enquiry | 200 | |
| Qualified | 120 | 60% |
| Appointment booked | 70 | 58% |
| Attended | 52 | 74% |
| Treatment accepted | 40 | 77% |
Overall, 40 of 200 enquiries (20%) became patients. The weakest step is Qualified to Appointment booked at 58%, and it is the only one a chat change can fix quickly: send the booking link in the same message as the qualification summary, and offer two time slots. Compare it against the attendance step, which reminders already handle. This is how to use the board: fix the single worst conversion each month, not everything.
Other numbers worth a monthly look:
- Pipeline value by stage, shown on the board, to see whether tomorrow's revenue is thin.
- Lost reasons. If half of the losses say "price", work on qualification, not discounts.
- Average response time, shown on the board, against the SLA you set.
- Won by source (ad, referral, walk-in), so you know where to spend.
If your leads come from Meta ads, map key stages to Meta events so the ad system learns which enquiries convert, not just which clicked. See automation for the flow side and pricing for plan limits on pipelines. Screens are on the pipeline feature page.
What are the most common WhatsApp pipeline mistakes?
- Stages that describe mood. "Hot", "Warm" and "Cold" invite argument. Replace them with events.
- Too many stages. Ten columns means cards live in nine of them for an hour each. Merge until every stage has a distinct next action.
- No owner on a card. Unassigned cards are nobody's problem. Decide assignment at entry: by source, by language or round robin.
- No lost reason. Without it you cannot tell a price problem from a response-time problem.
- Deleting a stage to tidy up. The board warns that deleting a stage also deletes its leads. Move or rename instead.
- Letting the AI and humans move the same cards by different rules. Write down which stages the AI may move to, and leave the rest to people.
- Renaming stages without updating instructions. The AI and any flows look stages up by name.
An afternoon is enough to set the board up: write the stage list with entry rules, create the qualification fields, add the stages, then update the AI instructions and one flow to use the exact names. Run it for two weeks before changing anything.
Sources (checked 13 Oct 2026)
- WhatsGrow AI pipeline, AI tool, flow and SLA behaviour: product documentation and current release
- Meta: Pricing on the WhatsApp Business Platform (India marketing rate in the INR rate card effective 1 October 2026; 24-hour window rules)
- The clinic numbers are an invented example, not a benchmark.