How to use this questionnaire
You are about to bring your first clients onto AssistantLabs. Before you switch on a single module or connect a single number, sit down with each client and go through the questions below. Each one says why we ask and what the answer changes, so you finish with a module list, a volume estimate, a channel plan and an order of work.
- Plan the time. A full run takes 60 to 90 minutes. Sections 1 to 5 and 11 are for every client; sections 6 to 9 only for the modules the client picks in section 2, so a client on Agents and Booking skips Sales, Helpdesk and most of section 9.
- Run it for yourselves once. Sections 3 and 10 apply to the agency itself: your pricing and margin, your brand, your plans page, your domain.
- One copy per client. Print the page or copy the Q numbers into your own document. The answer lines are for the call.
- “I don’t know” is an answer. Where the client does not know a number, section 3 gives the way to find it. Write down who will find it and by when.
- End with the build plan. The last section turns the answers into a plan you can start from the same day.
1. The business and its goals
Start with the business, not the software. These answers decide which module matters first, which language the agent is built in, and what “working” means in the first month.
What does the business sell, and to whom?
Why we ask. Decides whether a product catalog, a booking diary or a lead list is the centre of the build, and the tone the agent speaks in.
Answer
If the agent could do one job well from day one, which job?
Why we ask. Answering questions, booking appointments, recovering abandoned carts, qualifying leads or handling support. Start with one job and build it end to end before widening; it keeps the first month measurable.
Answer
What are the ten questions customers ask most, and where do the answers live today?
Why we ask. They become the FAQ. If the answers live in people’s heads, plan a guided interview with the owner as a knowledge source; if they live in documents, plan an upload.
Answer
Who answers customers today, in which hours, and in which languages?
Why we ask. Sets the hand-off: who gets notified, what the helpdesk business hours are, and when a person is expected to step in.
Answer
In which language do customers write: Hebrew, English, both?
Why we ask. The workspace is set up in Hebrew or English and the invite is written in that language; the agent answers each customer in the language they write in, and WhatsApp templates are written per language. Mixed audiences decide which languages the first templates go out in.
Answer
What number would tell you in 30 days that this worked?
Why we ask. Appointments booked, carts recovered, tickets closed without a person, leads called back. One number makes go-live a decision instead of a feeling, and it picks the target a journey is measured on.
Answer
Which topics must always go to a person: complaints, refunds, legal, medical, pricing exceptions?
Why we ask. These become guidelines for the agent, the triggers for notifications to your client’s staff, and the escalation rules in the helpdesk.
Answer
Who decides, and who is the day-to-day contact?
Why we ask. Decides who you invite as admin and who as member, and who has to be present when the WhatsApp number is connected.
Answer
2. Which modules
Every client workspace gets the Agents app. On top of that you switch on any of six modules per client. Which modules a workspace may switch on, and what extra modules add, follow your agency agreement.
| Module | Pick it if… | What it needs first |
|---|---|---|
| Agents (always on) | Customers write on WhatsApp, Instagram, Messenger, email, the website or a shareable link, and an AI agent should answer from the business’s own knowledge: FAQ, catalog, links, guidelines, scenarios, follow-ups, tags and notifications, with training from corrected answers. | Knowledge: a website, documents or an interview. |
| Sales | The business wants to start conversations itself: timed, branching journeys (messages, conditions, waits for an event, offers, cart links, bookings, reminders) that start from a new order, a new customer, an abandoned checkout, a segment, a group or a lead stage, with a goal that is measured; plus a leads board with stages, outcomes and call-back reminders. | A connected agent: Sales sends through the agent’s own channels. Store-driven entries need Shopify; lead sources can be agent labels, CRM record types or CRM companies. |
| CRM | The client needs its own data model, not just a contact list: contacts and companies plus its own record types (units, deals, contracts, devices…) with fields, relations, formulas and rollups, status pipelines, tasks, automations, roles, dashboards, imports and exports, webhooks and an API. | A connected agent: a CRM workspace reads one agent’s contacts. |
| Booking | The business sells its time or its nights: services with lengths, prices and categories, weekly hours and rules, a diary, a public booking page, Google Calendar sync, stays with check-in and iCal, WhatsApp confirmations and reminders, and an agent that books inside the chat. | Services and opening hours. WhatsApp reminders also need approved templates. |
| Helpdesk | Customers write with problems that must be tracked to closure by a team: intake by channel and label, a ticket lifecycle (open, escalated, resolved, stale) with priority, sentiment and topic, nightly auto-resolve, response targets, workflows with approvals and refunds, and a team inbox. | One connected agent. Conversations arrive from WhatsApp, Instagram, Messenger, Telegram, email and the website chat, one ticket per customer. |
| Call Center | Customers should be able to phone the business and have an AI agent answer in a natural voice, then every call is logged with its transcript. | The phone numbers customers will dial, a chosen agent, a voice and a speed. |
| Marketing | The client wants marketing run from one place: a product profile with unit economics, market research, strategies, scheduled social posts, email and WhatsApp campaigns, and Meta ads with generated creative that the marketer optimises within a spending cap. | Business details; for ads, a Meta Ads account; for WhatsApp campaigns, approved templates. |
Do customers ask questions that the business’s own knowledge can answer?
Why we ask. If yes, the Agents app alone is a complete first step, and it is always on. Everything else is added where a second job is real.
Answer
Does the business sell online and want to message people who left, bought or went quiet?
Why we ask. Yes means Sales. If the store is Shopify, journeys start from store events (new order, new customer, abandoned checkout) and segments; without a store they start from groups, uploaded lists, a CRM segment or a lead stage.
Answer
Does the business take appointments, reservations or stays?
Why we ask. Yes means Booking. It also decides whether the agent books inside the chat or sends a link.
Answer
Do problems need to be tracked to closure by more than one person?
Why we ask. Yes means Helpdesk: tickets with a lifecycle, priority, response targets, workflows with approvals, a team inbox. A business where the owner answers everything usually does not need it yet.
Answer
Is there a customer list to import, or does the client need companies and deals tracked?
Why we ask. Yes means CRM. People and companies come with it; anything else (deals, units, contracts, devices) is a record type the client designs. Contacts import from CSV and vCard.
Answer
Which numbers should customers call, and what should the phone agent do: answer questions, book, take a message, transfer to a person?
Why we ask. Each answer sets the Call Center’s agent, voice and speed, and what a call may end in. A business that already loses calls at lunch or after hours usually starts here.
Answer
How many modules will this client have switched on in total, and does that go beyond what your agency agreement includes per workspace?
Why we ask. Extra modules add to your bill for that workspace. Count them now so the retail price you set covers it.
Answer
Will the client use the apps themselves, or will you operate them for the client?
Why we ask. Decides the invite: the client joins their workspace as admin or member, and you decide how much they can change. You can build the agent and connect CRM, Booking and Helpdesk before they arrive.
Answer
Beyond people and companies, what does the client keep track of: deals, properties, units, contracts, devices, cases?
Why we ask. Each one is a CRM record type with its own fields, links to people and to other types, a status that moves through stages, and its own tab. List them now with their statuses; Section 9 digs into each.
Answer
Which marketing jobs does the client want: profile and economics, market research, a strategy, posts, email campaigns, WhatsApp campaigns, ads?
Why we ask. Marketing is a workspace of its own: each job is a separate part with its own prerequisites (a Meta Ads account for ads, approved templates for WhatsApp campaigns, a mailing audience for email). Say which, so the module is not sold as one lump.
Answer
Which decisions should reach the owner for a one-tap approval: a refund, a discount, a campaign, a stay request?
Why we ask. Approvals from the other apps gather on the Tasks board, where the owner approves or declines in one tap and sees what is waiting on them. List the decisions so they can be wired in.
Answer
3. Usage, volume and cost
Your bill is a fixed base plus a per-workspace and per-module part, plus one number you cannot know in advance: how many automatic conversations your clients’ agents handle. This section turns the client’s answers into that number and into a margin.
What counts as an automatic conversation
A conversation counts when the agent answered and no person on the business side wrote in it, or the first human message came after 48 hours. A conversation where your client’s staff replied within 48 hours is human-involved and is not counted. Test chats in the preview and your conversations with the Logos builder are never counted.
Your sizing worksheet
One row per client, from the answers to the questions below. Add it up, then use the rates in your agency agreement to turn modules and conversations into your cost.
| Client | Modules switched on | Conversations / month (low, typical, peak) | Messages sent per conversation | Your retail price |
|---|---|---|---|---|
| Total |
Margin per client is your retail price minus what that client adds to your bill: its modules and its conversations under your agency agreement, plus its workspace if it takes you past the workspaces your agreement includes. You bill your clients yourself; the plans on your public page show your prices to them.
What Meta charges on top
Meta charges separately for every WhatsApp message a business sends that is delivered. Messages the customer sends are never charged. Israel’s rates, from Meta’s own rate cards (at about ₪3.01 to the dollar):
| Message type | USD | ILS |
|---|---|---|
| Marketing template | $0.0353 | ₪0.106 |
| Utility or authentication template | $0.0053 | ₪0.016 |
| Service message: an ordinary reply, human or AI, from 1 October 2026 | $0.0053 | ₪0.016 |
At ten replies per conversation the Meta cost is about ₪16 for 100 conversations, ₪48 for 300, ₪80 for 500 and ₪160 for 1,000. What moves it is replies per conversation, not conversations: the same 500 conversations at 6 replies cost about ₪48 and at 20 replies about ₪160. A conversation that starts from a click-to-WhatsApp ad or a Facebook Page button, answered within 24 hours, opens a 72-hour window with no delivery charge.
In your retail price: Meta’s per-message charge sits next to your AssistantLabs bill, so decide with each client whether it is built into the plan price or shown as its own line.
In the last 30 days, how many new customer conversations did the business get, per channel?
Why we ask. This is the top of the funnel. Take it from the WhatsApp Business app, Instagram and Messenger inboxes, the website form and the phone log. If there is no count, someone counts one typical week and multiplies by 4.
Answer
Of those, what share does a person answer within 48 hours today?
Why we ask. Only conversations without a human reply inside 48 hours count as automatic. Expect your billable number to be the total minus the ones staff keep answering themselves.
Answer
How long does a typical conversation run, in messages the business sends?
Why we ask. Meta’s per-message charge is driven by it, not by the count of conversations. A chatty agent that splits one answer into three messages bills three times.
Answer
When are the peaks: seasons, sales, launches, holidays?
Why we ask. Your bill is variable. Plan the high month, not the average one, and warn the client before a campaign that will double volume.
Answer
Are paid ads planned that open a WhatsApp chat?
Why we ask. Click-to-WhatsApp leads that you answer within 24 hours are free to deliver for 72 hours. It is the one place where routing traffic differently lowers the Meta cost.
Answer
Will the business send messages first, to people who have not written in 24 hours? How many a month?
Why we ask. Those go out as approved templates, and a marketing template costs about 6.7 times an ordinary reply. Journeys, campaigns and appointment reminders are all in this bucket.
Answer
What will you charge this client per month, and in which currency?
Why we ask. Your margin is that price minus what the client adds to your bill (worksheet above). What you charge, and in which currency, is yours to set.
Answer
Is the price flat, or does it move with usage?
Why we ask. A flat price makes you carry the volume risk. A base plus a per-conversation part follows your own bill. This is your commercial decision, outside the platform.
Answer
4. Channels
A channel is where customers meet the agent. Each one has its own prerequisites, and the slow part is usually on the client’s side: who owns the Meta account and whether the number is eligible.
| Channel | How it connects | Ask about |
|---|---|---|
| Meta’s Embedded Signup with a Facebook login. An existing WhatsApp Business app number can stay on the phone and the agent answers on the same number (Coexistence). | Which app the number is in, its age and activity, who owns the Meta business account. | |
| Through Meta’s signup: log in, pick the Facebook Page, connect the linked professional account, approve message permissions. | A professional account linked to a Page; who is admin of the Page. | |
| Messenger | The same Meta signup, on the business’s Facebook Page. | Who is admin of the Page. |
| A private inbound address is minted for the channel and the business adds a forward-all rule from its real inbox. Replies go out automatically or are held as drafts for approval. It suits shared role inboxes such as support@ or info@. | Which mailbox, auto or draft, and whether they will add SPF/DKIM to send as their own domain. | |
| Website chat | A script tag pasted into the site, which opens a floating chat. | Who can edit the site’s code or footer. |
| Shareable link | A hosted chat page anyone can open, with a QR code, landing in its own inbox channel. | Where the link will be printed or posted. |
| Phone | The Call Center: a phone agent customers dial, with a chosen agent, voice and speed, and a log of calls with transcripts. | Which numbers customers dial today, and what happens to missed calls. |
| Connected in the Agents app; the agent answers LinkedIn messages in the same inbox as every other channel. | Whether customers actually start conversations there, and who admins the page. |
Where do customers write to the business today, in order of volume?
Why we ask. The first channel you connect should be the one with the most volume; the rest follow. It also tells you which of Instagram, Messenger and email are worth the setup at all.
Answer
Is the WhatsApp number on the WhatsApp Business app, on regular WhatsApp, or already on another platform?
Why we ask. Coexistence needs the WhatsApp Business app, version 2.24.17 or later. Regular WhatsApp moves to the Business app first, on the same phone and number. A number can be connected to one platform at a time, so an old system has to release it first.
Answer
How long has the number been used for real chats, and how active is it?
Why we ask. Meta decides eligibility by account age and activity and publishes no threshold. A number that is greyed out or refused is a common reason a connection stalls. Check it before connection day, not during it.
Answer
Who owns the Meta business account behind the number, and who can log in to it?
Why we ask. It is often an agency, a relative or a former employee. The connection starts with a Facebook login in Meta’s own window, so that person has to be there. Setups most often stall at this step.
Answer
Will someone open the WhatsApp Business app at least every two weeks?
Why we ask. A Coexistence connection drops if the app is not opened for about 14 days. Name the person who keeps it alive.
Answer
Does the business have an Instagram professional account linked to its Facebook Page, and who admins the Page?
Why we ask. Without the link the signup has nothing to connect. Messenger uses the same Meta signup on the Page.
Answer
Email: which shared inbox, who can add a forwarding rule, and should replies send automatically or wait as drafts?
Why we ask. Forwarding is the connection, so someone with admin rights on the mailbox must be involved. Drafts are the safe start for a business that is nervous about the agent writing in its name.
Answer
Website: who can paste a script into the site, or is the shareable link and a QR code enough?
Why we ask. A script on a Shopify, WordPress or Wix site is minutes of work if someone has the access. If nobody does, the hosted link ships the same day.
Answer
Outside the 24-hour window, what is the first message the business may need to send?
Why we ask. After 24 hours of silence from the customer only an approved WhatsApp template can open the conversation. Templates are approved by Meta, so write them early.
Answer
Which phone numbers should customers call, and when should a call go to a person instead of the agent?
Why we ask. Each number is answered by the chosen agent in the voice you pick. Name the hours, the topics that must reach a person, and where a missed or transferred call goes next.
Answer
5. Onboarding
Onboarding follows the same three weeks as the “Get Your Agent Live in 3 Weeks” playbook. As an agency you can do the first part before the client has even accepted the invite: build the agent inside their workspace from their website and documents, connect CRM, Booking and Helpdesk, and hand it over ready.
| When | What happens | Who |
|---|---|---|
| Before the client arrives | You create the client workspace (Hebrew or English), build the agent from the website, documents and notes, connect the modules and add the booking services. Then you invite the client as admin or member. | You |
| Week 1: identity, knowledge, test | Give the agent a personality, add business details, FAQ, catalog, links and guidelines, and test it in the preview before anyone sees it. | You, with the client’s answers |
| Week 2: connect quietly, then go live | Connect WhatsApp with the agent muted, wait a few days while it listens to real questions, answer what it missed, set up tags and notifications, then turn it on. | Client logs in to Meta; you do the rest |
| Week 3: run it | Learn the inbox, step into conversations when you want, read the customer’s history, and correct answers so the agent learns. | Client and you |
- Hand-over list. The client gives you: the website address, price lists and policy documents, the services and opening hours, the names and WhatsApp numbers of who should be notified, and a person for the Meta login.
- Never collect passwords. Every connection is done by the client signing in themselves in the provider’s own window. If you find yourself holding a client’s password, stop.
What can the client give the agent to learn from?
Why we ask. A website address (scanned), documents (uploaded), a guided interview with the owner, and later the conversations themselves. The richer the set, the less the muted week has to catch.
Answer
Is there a product catalog, and where does it live?
Why we ask. A store connection, a sheet, a PDF or nothing. Decides whether the catalog is read live, imported, or typed.
Answer
Who will test the agent before it goes live, and with which ten questions?
Why we ask. Testing in the preview is free and never billed. A client who has asked the agent their own hardest questions trusts it at go-live.
Answer
Who watches the muted days and answers what the agent missed?
Why we ask. The agent starts muted and listens first. Someone has to read its misses and answer them so it learns. If no one owns it, the quiet week is wasted.
Answer
Who gets a WhatsApp notification, for which triggers, and in which hours?
Why we ask. Notifications go to the numbers you list when a tag fires, such as complaint, asks for a person or high value. Too many and people tune them out, so start with three.
Answer
Is there a date the business must be live by, such as a season, a sale or an ad campaign?
Why we ask. A hard date works backwards: Meta steps are the long ones, so schedule the WhatsApp connection first and do the rest around it.
Answer
Who is invited, with which role, and when?
Why we ask. You can have the workspace fully prepared before the invite goes out. Admin can change settings; member works in the inbox. Decide per person.
Answer
Is there history to learn from: past conversations, past tickets?
Why we ask. WhatsApp history syncs on connection if the client shares it (about six months), and the agent can learn from conversations. A helpdesk imports up to the 20 most recent conversations from the last 14 days as tickets when it is connected.
Answer
6. Customer journeys
A journey is a timed, branching path that a business starts instead of waiting for the customer, which can watch what the customer does, wait for a real-world event, make an offer, hand over a cart, book a diary slot, remind a person, and stop itself at the right moment, with a goal it is judged against. This section captures all of it.
| Building block | What it does |
|---|---|
| Entry | What puts a person in: a new order, a new customer, an abandoned checkout, membership of a segment or group, or a lead moving into a stage or pressing a button. An extra filter can narrow it, such as a cart over a given value or a returning customer. |
| Message | AI-written, fixed text, an approved template, a product card or a cart. Sent on WhatsApp, email, Instagram, Messenger or the website chat. Outside the 24-hour window only a template goes out on WhatsApp; email has no window. |
| Wait | A delay in minutes, hours or days, or a clock time such as “the next morning at 07:00”, in the workspace time zone. A due message waits for business hours unless the step says to send anyway. |
| Condition | Branches yes or no on what happened: replied, clicked a link, purchased, unsubscribed, booked. Opens and clicks are read where the channel reports them, such as email; on WhatsApp branch on replies, purchases and bookings. |
| Event | Parks the person until a real milestone arrives: an order paid, a checkout abandoned, the customer replied. |
| Offer | A personal discount code minted through Shopify, with a strategy and tiers. |
| Cart | A one-tap checkout link that carries the items already in the cart, so the customer pays without rebuilding it. |
| Booking | Sends the booking page, or lets the agent book the person in. The diary then feeds the “booked” condition. |
| Reminder | Writes a to-do on the lead for a person: call back, send a quote. |
| Exit rules | Guards checked before every step: stop on purchase, on a reply hand to the agent or just stop. An unsubscribe always stops the journey. |
| Goal and target | The goal steers the AI-written text. The target is what is measured: a purchase, a purchase with a discount, a signup, a course signup or a happy customer, inside an attribution window (7 days by default). |
Which moments make the business lose people or leave money on the table: an abandoned cart, a first purchase with no second, a lapsed customer, a quote that went quiet, a no-show, a lead nobody called back?
Why we ask. Each answer is one journey. Rank them by money lost; the top one is the first build.
Answer
For each one, what starts it: a new order, a new customer, an abandoned checkout, a segment, a group, a lead entering a stage, or a button a person presses?
Why we ask. Entry is the first decision of the builder. An abandoned checkout waits a short while (15 minutes by default) and enrolls only checkouts that no order claimed; segments and groups enroll the people in them.
Answer
Is the store on Shopify, and may AssistantLabs read its customers, orders and abandoned checkouts?
Why we ask. Store-driven entries come from Shopify. After connecting, recommended segments (VIPs, lapsed buyers and similar) are offered and created in the store.
Answer
Should anyone be filtered out at entry: a cart under a certain value, first-time customers only, people who already got a message this week?
Why we ask. An entry filter is an extra gate on top of the trigger. Without it, everyone who matches the trigger enters.
Answer
Draw the path: after the first message, what happens if they reply, click, buy, book, or say nothing for two days?
Why we ask. Every “what if” is a condition or an event wait. Writing the paths on paper first turns a vague “follow up” into a branching journey you can actually build and test.
Answer
How long between steps: minutes after a cart, a day, or a fixed clock time such as “the next morning at 07:00”?
Why we ask. Timing is a delay or a clock time in the workspace time zone. A message due outside business hours waits for the next open time unless that step is set to send anyway.
Answer
How should each message be written: by the AI, fixed wording, an approved template, a product card, or the customer’s own cart?
Why we ask. Inside the 24-hour window all of these are available; outside it only an approved template goes out, and when a journey starts from an abandoned checkout no step has an open window, so every message must be a template. Templates are written per language.
Answer
May a journey offer a discount, and what is the ceiling?
Why we ask. An offer step mints a personal code through Shopify and can follow a strategy with tiers. Set the ceiling with the client now so nobody discovers it in a campaign.
Answer
When the customer replies, who takes over (the agent, a person, or nobody), and what else must stop the journey?
Why we ask. Exit rules decide: stop on purchase, on a reply hand to the agent or just stop. An unsubscribe always stops it. This is the line between a nudge and a nuisance.
Answer
What is the goal in words, and what single outcome is the measured target?
Why we ask. The goal steers the AI text. The target is one of: a purchase, a purchase with a discount, a signup, a course signup (for chosen products), a happy customer. Success is counted inside the attribution window, and the strength of the evidence is kept per conversion (a redeemed code, a cart link, or only a purchase inside the window).
Answer
Where do leads come from, and who is NOT a lead?
Why we ask. A lead can be an agent contact picked by labels (with labels that exclude suppliers, partners and your own team), a person who actually wrote in, a new record of chosen CRM record types, or a CRM company that churned. Paying customers are left out by default.
Answer
Who are the groups: uploaded lists, a label, a Shopify segment, a CRM segment, a webhook feed?
Why we ask. A group is a saved audience that feeds one journey, and a live match count is visible while you build it in the app.
Answer
If the business works leads (B2B, quotes): which stages and outcomes, who owns a lead, and what happens on each button?
Why we ask. Stages and outcome buttons are configurable and each can carry effects: change stage, add a label, set an owner, set a reminder, notify, send a message, start a booking, start or stop a journey, call a webhook. A closing outcome stops automation by default, and a stage can fire when a lead sits untouched too long. A daily “who do I owe a call” reminder can go to WhatsApp or Slack.
Answer
Who are the test users who receive real messages before launch?
Why we ask. Test users get the real journey on their own phones, and a test run lets you walk the whole path. The client’s own people are the right audience for the first run.
Answer
Should the agent also follow up on chats that went quiet, outside of Sales journeys, and when should it stop?
Why we ask. Follow-ups in the Agents app nudge after a delay you choose and stop when the customer replies. Pair them with tags so a human sees what matters.
Answer
7. Booking services
Booking is one diary with four ways in: a public booking page, the agent in the chat, Claude, and the app itself. All four go through the same rules, so the answers below are the rules.
List every service: name, length, price, category, a short description, and what the customer should bring.
Why we ask. Each service is a row. An empty price means the agent says it depends and asks. “What to bring” is copied onto the booking and sent in the confirmation.
Answer
Opening hours for each weekday, including breaks, and the time zone.
Why we ask. Two windows in one day make a break. The starting defaults are Sunday to Thursday 09:00 to 19:00 and Friday 09:00 to 14:00 in Israel time; change what does not fit.
Answer
The rules: how often a start time is offered, the gap after each appointment, minimum notice, how far ahead, how many at once.
Why we ask. The starting defaults are a 30-minute grid, 60 minutes notice, 60 days ahead and one appointment at a time.
Answer
Which services are held online, and is there one meeting link per service?
Why we ask. An online service carries one meeting link, set once, and it goes out with every confirmation.
Answer
Does the business sell stays by the night?
Why we ask. A stay has check-in and check-out, minimum and maximum nights, units, maximum guests, and either instant booking or a request the owner confirms. Airbnb and Booking.com calendars can block nights through iCal feeds, and each stay service exports its own feed.
Answer
Is there a Google Calendar to sync, and in which direction?
Why we ask. Sync runs both ways and each direction can be switched. The calendar can be shared (no sign-in) or connected by signing in with Google. Event titles never leave the owner’s screen.
Answer
May the agent book directly, send a link, or both?
Why we ask. With “send a link” the agent sends the booking page; with “book directly” it books inside the chat; with both it picks per conversation. The default is both.
Answer
Which WhatsApp messages around a booking: invite, confirmation, reminder, cancellation, and how long before?
Why we ask. Nothing goes out until its template is approved. Reminder timing is set per service, and one message can be muted for one service. Confirmations and reminders are utility templates, an invite is marketing.
Answer
Do staff need to log walk-ins or book outside the customer rules?
Why we ask. The business is not bound by the customer’s notice and advance rules, but it still respects hours and capacity.
Answer
The booking page: address slug, accent colour, show prices and length, a WhatsApp button, a welcome line.
Why we ask. The public page lives at booking.assistantlabs.io/book/<slug>. A visitor from abroad reads it in English on their own clock, with a time-zone picker.
Answer
Which categories do the services fall in, and who closes each appointment as visited or no-show?
Why we ask. Categories colour the diary. Past appointments nobody closed show up under “Needs you”, and visited and no-show feed the customer’s record and the numbers above the diary. Name the person who closes them.
Answer
Should bookings also appear in the CRM and in Sales?
Why we ask. Every booking is mirrored onto the customer’s contact (requested, booked, moved, cancelled, visited, no-show, and who booked it). With Sales, a journey can include a booking step, branch on “booked”, and a lead can be booked in from its page.
Answer
In which languages do customers read the messages and the booking page?
Why we ask. Messages are written per language and the workspace has a language of its own; one-step message setup writes them in the workspace language and submits them for WhatsApp approval, which usually takes up to a day.
Answer
Does the owner want to run the diary from Claude: add a service, move hours, send an invite?
Why we ask. With the AssistantLabs plugin installed in Claude, the owner adds services, changes hours, restyles the booking page, connects the calendar and invites someone to book, all in plain words. Decide who on the client side gets it.
Answer
Which appointments need a deposit, who should be the named staff member for each service, and should a freed slot be offered to a waitlist?
Why we ask. A deposit is set on the services that need one, a customer can ask for the same person every time, and a cancelled slot is offered to the waitlist automatically. List them per service.
Answer
8. Helpdesk
The helpdesk sits on top of the agent’s conversations: every conversation that arrives can become a ticket that is open, escalated, resolved or stale, with a priority from 1 to 5, a topic and a sentiment. The agent still does the talking; the helpdesk adds triage, ownership and closure.
Which conversations should become tickets: every channel, or only some, and all of them or only those with a label such as “support”?
Why we ask. Intake takes WhatsApp, Instagram, Messenger, Telegram, email and website chat, and can be narrowed to conversations carrying chosen labels. It matters for a business whose agent also sells and books and wants only the support part here.
Answer
Who is on the support team, and who is admin and who member?
Why we ask. Admins change settings; members work the inbox. List names and emails now so seats are ready when the client arrives.
Answer
Business hours, time zone and closed days.
Why we ask. They decide whether a closure day is marked as an after-hours resolution, and whether response targets stop counting while the business is closed. Israeli holidays can be added from a suggestions list, and out of hours the helpdesk can auto-reply, auto-tag, pause the response clock or let the agent answer; choose which.
Answer
What first-reply and resolution times does the business promise, and do they differ by priority?
Why we ask. Response targets are switched off until you turn them on, so old tickets are not suddenly overdue. There is a default target and a target per priority; the AI’s reply counts as the first response, and with business hours on the clock counts open minutes only. A target is minutes: “one day” is 1,440 open minutes, not a working day.
Answer
May tickets close on their own when the agent replied last and the customer did not come back?
Why we ask. A nightly pass (2 AM UTC) closes tickets as “assumed resolved” and escalates the rest. It is on by default and can be switched off. Say plainly to the client that it infers, it does not ask.
Answer
Should the agent answer order-status questions, and how much proof of identity is enough?
Why we ask. Answers are off by default. When on, a policy decides: verified (the order must match the customer’s verified identity), by order number, or off. Choose which order fields may be shared.
Answer
What must always reach a person, and how are they alerted?
Why we ask. The agent hands these to a person. Alerts go by WhatsApp notification, or to a Slack channel through the Slack integration, so decide where the team actually looks.
Answer
What vocabulary should tickets and contacts be tagged with?
Why we ask. Give a starting list of ticket tags, contact tags and topics. The nightly pass matches against it instead of inventing new words, and grows it.
Answer
Which email addresses, website chat and other channels should feed the helpdesk, and should one customer keep one ticket across them?
Why we ask. Telegram, email and website chat conversations become tickets just like WhatsApp, Instagram and Messenger ones, and the same customer writing on two channels stays one ticket. List the addresses and the pages where the chat is embedded.
Answer
Who owns a ticket, and when should the AI be paused on it?
Why we ask. A ticket can be assigned to a team member, carry internal notes, be escalated or resolved by hand, and have the AI paused so a person takes the conversation. Decide the working rule: who picks up what, and who may pause the agent.
Answer
Which processes need steps and approvals: a refund, an exchange, a manager sign-off, a call to another system?
Why we ask. Workflows are chains of steps: message the customer, send a template, ask for approval, create a task, branch on a condition, call an HTTP API, notify the team, refund through Shopify, act on the ticket. List each process and who approves it.
Answer
How should new tickets reach the right person or team: by topic, by language, by priority, by time of day?
Why we ask. Routing rules assign each new ticket on arrival. A rule can assign one person or a team that takes turns. Give the rules in order, with who gets what and who is the fallback.
Answer
Which parts of the helpdesk should be switched on for this client?
Why we ask. Features switch separately: human hand-off, the instructions the helpdesk adds to the agent’s prompt, ticketing, the nightly auto-resolve, and the history import (chosen once, when the agent is connected). The first two change what the client’s own agent does; the rest only change what the helpdesk does.
Answer
9. CRM, marketing, integrations and data
Three things live here: what the client wants to keep in the CRM, what marketing it wants run, and where the agent reads facts from and what it may write back. The rule of thumb for integrations: the agent reads live from the client’s own systems, and writes back only where you allow it.
The CRM
| CRM part | What the client can have |
|---|---|
| Record types | Its own kinds of record beyond contacts and companies, each with an icon, a colour, a tab, a title, list columns and filters. |
| Fields | Text, long text, number, currency, date, single and multiple choice, phone, email, link, checkbox, team member, files, priced line items (a quote with VAT), and relations to other records. |
| Formulas and rollups | Computed fields: dates and days left, totals with VAT, commission, and rollups that sum what is linked here (a contract’s total paid). |
| Pipelines | A single-choice status can limit which value may follow which, and colours show progress, so a deal or a case moves through stages. |
| Tasks | Today, overdue, upcoming and done, mine or everyone’s, as a list or a calendar. |
| Automations | When, only if, then rules per record type. They are saved switched off and act once per record and event. |
| Roles | Team roles that limit what each member sees and changes. Members with no role keep full access. |
| Dashboards and reports | Reports over record types, saved as dashboard widgets. |
| Import and export | CSV and vCard for contacts; Excel or CSV for a record type, up to 5,000 rows, with a check step before anything is saved; export to CSV. |
| Broadcasts and signals | Send or schedule a message to an audience from the CRM; signals such as going quiet or due back raise tasks. |
| Webhooks and API | Events out to other systems and a developer API in. |
| Page design | Record pages and lists are designed with blocks, presets and a “design my CRM” proposal that is reviewed before it is published. |
List every kind of record the client tracks, and for each one the fields that matter, including which are required or must be unique.
Why we ask. This is the record-type design. Create the types others link to first. People and companies already exist as contacts and companies, so do not duplicate them.
Answer
How do the records relate: a contract to a company, a payment to a contract, a unit to a building?
Why we ask. Each link is a relation field; the linked record’s page lists everything pointing at it. A link needs its target type to exist first.
Answer
What should the system compute: an end date, days to renewal, a total with VAT, a commission, a balance owed?
Why we ask. These become formulas and rollups. If a person does the arithmetic today, it is a formula tomorrow.
Answer
What are the stages of each process (a deal, a case, a rental), and may a record skip or go back?
Why we ask. A status field can list the allowed next values, which is how a pipeline is kept honest. Also decide which colours mean good, waiting and a problem.
Answer
What follow-ups must never be forgotten, and who owns them?
Why we ask. These are tasks, optionally created by an automation (when a record changes, then create a task or change a field). Name an owner for each.
Answer
Who on the team should see and change what: sales, finance, admin, external?
Why we ask. Roles limit visibility and edits per member. Structure changes (record types, roles, rules, connecting the agent) are for workspace admins only.
Answer
Which five numbers should the owner see on a dashboard on Monday morning?
Why we ask. Each is a report over a record type, saved as a widget. If the number needs a field that does not exist yet, it is added to Q9.1 now.
Answer
What existing data has to come in: spreadsheets, a phone’s contacts, another system’s export?
Why we ask. Contacts and each record type import separately, with duplicate handling and a check step before saving (up to 5,000 rows at a time). Importing is easier when the sheet’s columns already match the fields from Q9.1.
Answer
Which other systems must be told when a record changes, or must write into the CRM?
Why we ask. Webhooks send events out and the developer API writes in. Name the system and the event.
Answer
What currency and VAT does the workspace use, and what language are record names in?
Why we ask. The workspace has one currency that a currency field shows when it has no unit of its own, and record types are named in the workspace language.
Answer
Marketing
What does the product earn per sale and cost to acquire: price, cost, margin, what a customer is worth, what a lead may cost?
Why we ask. The product profile holds six inputs and works out the derived numbers from them, so research and strategy rest on the client’s real economics rather than on guesses.
Answer
Who are the competitors and the alternatives, and which markets and audiences matter?
Why we ask. The research is built around the profile: it looks at the market and the trends around the product. Name what the client already knows so the research starts from there.
Answer
What is the goal for the next quarter, and is there one strategy or several (products, audiences, markets)?
Why we ask. A strategy turns the profile and the research into goals, audiences, messages and campaigns. A client with distinct lines of business may need more than one.
Answer
Which social channels will post, how often, who approves a post, and in which language?
Why we ask. Posts are composed per connected channel and scheduled for a date and time. Decide the cadence and the approver so the calendar fills the same week the channels connect.
Answer
Which email and WhatsApp campaigns: to whom, how often, from which sender, and what consent do the recipients have?
Why we ask. Campaigns have an audience, a design and a send. WhatsApp campaigns go out as approved templates, and Meta charges per marketing template delivered, so the audience size is a cost decision. Consent must be settled before the first send.
Answer
Will the client run Meta ads, with what budget and spending cap, and who owns the ad account?
Why we ask. The marketer generates on-brand creative and copy in several angles, in Hebrew and English, moves budget to winners, cuts losers and never crosses the cap; bigger moves arrive as one-tap approvals. Report on outcomes against goals and cost per lead.
Answer
Where should leads from ads and campaigns land: the chat, the CRM, a Sales journey, the booking page?
Why we ask. A click-to-WhatsApp ad lands in the agent’s chat and, answered within 24 hours, is free to deliver for 72 hours. Decide the first hop so no lead lands where nobody looks.
Answer
Integrations and data
| Connects | What the agent gets | What the client does |
|---|---|---|
| Shopify | Products, variants, availability, order status (opt-in), and the customers and checkouts that Sales journeys use. One shared install serves Sales and Agents. | Installs the AssistantLabs app from the Shopify admin. |
| WooCommerce | Products, categories and the customer’s orders through read-only keys. | Creates read-only REST keys and, if wanted, pastes the chat snippet. |
| Google Sheets | Plain-language search over rows, with an edit reaching answers within minutes. A conversation never writes to the sheet. | Shares the sheet as “anyone with the link: viewer”. |
| Google Calendar, Calendly | Free and busy times for the agent’s booking tools. | Connects the calendar they actually use. |
| Wix, WordPress, eShop, Monday, Flashy, Make, a form webhook, send-message | Store data from Wix and eShop, the chat widget on WordPress, contacts and conversations synced with a Monday board, WhatsApp templates fired from Flashy, and events in and out through Make, a form webhook or the send-message API. | Only if the client already uses it. |
| Any system with an API | A custom integration turns an external HTTP API into a tool the agent calls while it answers, for example order tracking or stock. | Provides the endpoint and a key; you define the call. |
Which platform runs the store: Shopify, WooCommerce, Wix, or something else?
Why we ask. Decides the integration and which kinds of questions can be answered from live data (stock, order status).
Answer
Where do the prices and stock actually live?
Why we ask. A store, a sheet, an ERP, a PDF. A sheet is the quickest honest answer for a small business: edit a row and the answer changes within minutes.
Answer
May the agent answer order questions, and for which orders?
Why we ask. Order lookups return only the writing customer’s own order and are opt-in. Settle it together with the helpdesk policy in section 8.
Answer
Is there an existing customer list, and in what format?
Why we ask. CSV and vCard import into the CRM with duplicate checking. A Monday board syncs contacts and conversations.
Answer
Is there a system with an API the agent should call while it answers?
Why we ask. This is a custom integration. It needs an endpoint, an authentication method, and a plain-language description of when to call it.
Answer
Should anything be pushed out when something happens: a lead arrives, an order closes, a booking is made?
Why we ask. Sales leads can fire a signed https webhook; forms can post in through a form webhook; Make and the send-message API cover the rest. Decide the events before the build.
Answer
Does the owner want to run the business from Claude or ChatGPT?
Why we ask. In Claude, the owner installs the AssistantLabs plugin from Claude’s plugin directory and signs in once; it carries every app the workspace has, and sending, posting and ad spend stay off until the owner allows them. In ChatGPT it connects as an app on a Business, Enterprise or Edu workspace. Decide who uses it and what it may do.
Answer
Who administers each third-party account, and can they be present on the day?
Why we ask. Each connection is made by the account’s own administrator signing in themselves. The person who can click “install” is the real dependency.
Answer
10. Brand and white label
This section is mostly about you, not the client. It decides how your clients and their customers see the product, and what they can reach you through.
Your brand: name, logo, primary colour, and whether to hide “powered by”.
Why we ask. The colour is applied across the apps your clients use. A logo is PNG, JPEG or WebP up to 2 MB.
Answer
How is each client workspace named, and in which language does it open?
Why we ask. A workspace is Hebrew or English: the agent’s language, the invite and the account all follow it. Name them consistently so your own switcher stays readable at ten clients.
Answer
Your retail plans: names, prices, billing interval, modules included, bullet points, which one is highlighted.
Why we ask. They show on your public plans page, and you bill clients yourself. Keep each plan’s modules honest against the module count in section 2.
Answer
The public page: address slug, headline, short “about”, contact email, and who answers leads.
Why we ask. The page lives at agency.assistantlabs.io/<slug> once switched on. Leads are stored and emailed to the contact address, so decide who answers and how fast.
Answer
Do you want your own domain, and who controls its DNS?
Why we ask. Ownership is verified with a TXT record at _assistantlabs.<your domain>. Each app then gets its own subdomain on it (app., crm., booking.…). Once verified, your clients reach every app on your own domain.
Answer
Should invites reach clients from your own email address?
Why we ask. With an SPF and DKIM domain verified, invites come from an address on your domain instead of ours.
Answer
Who is the agency owner who pays, and by which card?
Why we ask. Only the owner can pay the base fee at the checkout, since their account carries every client’s usage.
Answer
11. Risks and compliance
This is not legal advice. It is the list of questions that have to be asked before an AI agent speaks to a business’s customers on its behalf. Under Israeli privacy law the client is the database owner and AssistantLabs is a holder, acting on the client’s instructions; accountability to individuals stays with the client.
Does the client handle sensitive information (health, finance, minors) or is it a public body?
Why we ask. It raises the bar and changes what the agent may answer. Hand the client’s privacy officer the privacy pack: data-flow description, sub-processor list, security measures, DPA and the Israel Privacy Addendum.
Answer
Who is the privacy officer, and does the privacy policy mention messaging and AI?
Why we ask. A policy that does not mention the messaging channel or the use of AI does not cover this. The policy is the client’s, on their site.
Answer
Who approves the opening disclosure text?
Why we ask. The first message of a new conversation says it is an automated system, what the data is for, who receives it, and how to reach a person. It is fixed text, not something the model writes. We supply a draft; the client approves it.
Answer
How long is data kept, and what may the agent never answer?
Why we ask. The retention period, the response scope and the blocked field types are the client’s decisions, written down in the configuration record.
Answer
Are any customers outside Israel?
Why we ask. Language, dates and currency follow the visitor’s region: Israel sees Hebrew and shekels, everywhere else English and dollars. Public pages such as the booking page switch to English for a visitor from abroad.
Answer
Does the client want the agent to have a human first name or to type slowly to look human?
Why we ask. No to both. Both undercut the disclosure and both are prohibited by the WhatsApp Business messaging policy, independent of privacy law.
Answer
Who handles a request from a customer to see, correct or delete their data, and who reports an incident?
Why we ask. The client replies and reports; we provide the data and the erase function, and we tell the client about an incident quickly enough for them to report it.
Answer
From answers to a build plan
When the call ends, you should be able to fill this in the same afternoon. Copy it per client.
| Step | Take from | Output |
|---|---|---|
| 1. Module list | Q1.2, Q2.1 to Q2.11 | Which modules are on; how many beyond your agreement. |
| 2. Estimate | Q3.1 to Q3.8 | Conversations a month (low, typical, peak); cost under your agreement; Meta cost; your margin. |
| 3. Channel plan | Q4.1 to Q4.10 | First channel, owner of its Meta account, eligibility risk, templates to write. |
| 4. Knowledge and access | Q1.3, Q5.1 to Q5.3, Q9.18 to Q9.25 | Sources to add; documents and logins to request; who signs in. |
| 5. Journeys, bookings, helpdesk, CRM, marketing | Q6 to Q8, Q9.1 to Q9.17 | The first journey, the services list, helpdesk intake and response targets, the CRM record types, the marketing plan. |
| 6. Order of work and date | Q5.4 to Q5.8 | Who does what in which week; go-live date; who watches the muted days. |
| 7. Open risks | Q11.1 to Q11.7 | What has to be approved, by whom, before the first customer is answered. |
Before you leave the call
- Every question has an answer, or a name and a date for who will find it.
- The module list and the number beyond your agreement are agreed with the client.
- The first channel and the person who signs in to its Meta account are named.
- The first journey, the first services and the first helpdesk rules are written down.
- The date for go-live and the person watching the muted days are fixed.
- The privacy pack has gone to the client’s privacy officer and the opening disclosure has an approver.
One last thing: Keep the filled-in copy. In three months it is the document that tells you whether the client’s volume, journeys and hours turned out the way you planned, and it is the first thing we will ask for if something looks wrong on a bill.