By Nudge team
A readiness guide for owners and operations teams
Updated
Start with one customer journey, not “automate WhatsApp”
“Automate WhatsApp” is too broad to build or judge. Pick one useful journey with a clear beginning and end. A good first workflow might answer service questions, collect two qualifying details, check a real calendar and offer a handoff when the request needs judgment.
Write the current human process first. Where does the enquiry come from? What questions does the team answer? Which details are truly needed? What system records the outcome? Where do delays and mistakes happen? Automation should improve that path, not conceal a process nobody understands.
Use the free WhatsApp automation readiness checker to identify the first missing foundations. The score is a planning aid. It does not promise approval, revenue or a safe production launch.
Prepare the official Meta Cloud API path
Production automation should use the official Meta Cloud API and a correctly configured WhatsApp Business Account. Avoid unofficial browser automation. It is not a reliable base for customer service, consent controls, message status or a business-critical workflow.
Decide who owns the Meta business assets, phone number, templates and access. Keep credentials on the server. Plan for webhook verification, signed events, retries and duplicate messages. If an agency or provider helps, the business should still understand who controls the account and how it can leave.
Review Meta’s current WhatsApp Business Platform documentation before launch. Current platform documentation is the source of truth for setup and supported behavior.
Create one approved source of business knowledge
Gather services, prices, hours, locations, policies and the answers staff already give. Remove contradictions. Add an owner and review date. If two pages show different prices, an AI cannot decide which one represents the business.
Record the limits too: questions the system must not answer, facts that change often, and requests that need a person. A useful agent says “I’ll ask the team” when the approved knowledge is missing. It should not become a general-purpose chatbot or invent an answer to keep the conversation moving.
CustomerCan you guarantee this service will solve my problem?
BusinessI can explain what the service includes, but I can’t guarantee an individual result. I’ll ask a teammate to help with your specific question.
Save the customer’s question and route it to the named handoff owner instead of guessing.
Define consent, service-window and follow-up rules
Keep service replies and marketing permission separate. A customer asking a question does not automatically agree to unrelated promotions. Record opt-in evidence for marketing and make opt-outs permanent. STOP must win over a campaign, scheduled job or later contact import.
Track the customer’s most recent inbound message. Free-form replies belong inside the supported service window. Business-initiated contact outside it should use the appropriate approved template. Put these checks in the send path and repeat them immediately before queued work sends.
The WhatsApp 24-hour rule guide explains the operating decision in detail.
Choose one real business action with a clear result
A useful AI Front Desk does more than answer. Choose one real business action for the pilot: check calendar availability, create a provisional booking, write a qualified lead to the CRM or create a payment link. The action should have defined inputs, permissions and a result the system can verify.
Do not give the model broad database access. The application should validate the business account, input values and user permissions, run a narrow action and record its outcome. If booking fails, the customer should not receive a confident confirmation.
For the technical sequence, read how to build WhatsApp AI automation with the official Cloud API.
Design human handoff before the first live conversation
Name the person or queue that receives handoffs. Define triggers: the customer asks for a person, the answer is missing, the request is sensitive, an action fails, a complaint arrives or the workflow reaches a limit. Tell the customer what will happen next.
Send enough context for the teammate to act: the customer’s intent, approved facts already shared, details collected, actions attempted and the unresolved question. Pause automation while a person is in control. Resuming should be an explicit operating decision.
Run a measured pilot before expanding
Test the entire journey in a simulation or sandbox first. Include correct questions, missing knowledge, conflicting information, expired service windows, opt-outs, action failures and human takeover. Then pilot with a narrow live scope and a person watching.
Judge readiness by evidence, not how fluent the demo sounds. Review conversations, action records, handoff times and policy exceptions. Fix the process before adding more channels or more complex actions.
- The AI answers only from approved business knowledge.
- Unknown or sensitive requests reach the correct person with context.
- One real action completes correctly and is recorded in its source system.
- Free-form sends stop when the service window closes.
- Marketing sends require opt-in and STOP remains permanent.
- The team can explain who owns every open conversation and next action.
Once these foundations hold, expand one workflow at a time. That is a much stronger base than publishing a broad chatbot and hoping the team can repair the operating gaps later.