A WhatsApp AI assistant for a Singapore fitness franchise — built to share a phone number it didn't own
Singapore flag
Build · Fitness / AI · Singapore

workspace_premium A WhatsApp AI assistant for a Singapore fitness franchise — built to share a phone number it didn't own

For a Singapore-based FS8 franchisee operating three Pilates and functional-training studios, XMARTECH built a WhatsApp AI assistant that answers fitness and nutrition queries, books and reschedules classes through Mindbody, and hands conversations to a human when it recognises its own uncertainty.

auto_awesome Chatbot / AI Assistant Story

We build for the failure before it happens — here is what that looked like on this project.

This page is structured as buyer questions with answer-first evidence, so clients can quickly understand risk, solution logic, and measurable outcomes.

crisis_alert Problem shape

Context-constrained AI agent on a shared communication channel

insights Lead outcome

Same WhatsApp number · Mindbody booking live · confidence-based human handoff

category Delivery mode

Build

3

Singapore studios on one line

1

shared WhatsApp Business number

2

handoff triggers (user + confidence)

3u21921

integration attempts to working path

menu_book The whole story, in one paragraph

The client asked for a bot. The number was already claimed by three systems.

For a Singapore-based FS8 franchisee operating three Pilates and functional-training studios, XMARTECH built a WhatsApp AI assistant that answers fitness and nutrition queries, books and reschedules classes through Mindbody, and hands conversations to a human when it recognises its own uncertainty. The unusual part: before the bot could go live, XMARTECH had to solve a problem the client didn't know existed — their WhatsApp number was already claimed by a CRM, wired into a booking platform, and used for manual staff replies. Three integration attempts and two failures later, the bot runs on that same number without breaking anything that was already there.

map Story map

gpp_badWhat was breaking or what was the risk?

The client's WhatsApp Business number was already the single communication line for three studios — connected to Mindbody for class bookings, to GROW CRM for lead management, and used by staff for manual replies. They wanted a bot on that number. Not a new number. The same one. A wrong AI answer in a franchise context does not just embarrass one studio — it misrepresents a global brand on a channel customers trust.

build_circleHow does a fitness studio's AI assistant decide what it should and shouldn't answer?

XMARTECH built the FS8 Singapore bot with a defined ruleset that constrains every response to fitness, diet, workout, and nutrition topics — biased toward the client's three studios (FS8 Hillview, FS8 Lower Peirce, and FS8 Upper Thomson) and their class types. The AI references this ruleset before every reply. For new users, the bot provides conversational fitness guidance and recommends a specific studio and class. For existing members with a Mindbody account, it books, schedules, and reschedules classes directly through WhatsApp. The boundary is deliberate: the bot does not create new Mindbody accounts. That step stays human.

precision_manufacturingHow was it built to hold?

Two handoff triggers protect the brand. A user can ask to speak to a person at any point — the bot routes the conversation to a human representative within the same WhatsApp thread. The second trigger is confidence-based: the bot monitors its own certainty against its ruleset, and when it cannot answer accurately, it escalates to a human before the user has to ask. WhatsApp coexistence runs through GROW CRM, which already held the number after Meta direct and YCloud paths failed. Bot and staff share the same thread. Mindbody integration runs live underneath for class transactions for existing members.

How the number was claimed

Three attempts. Two failures. One working path through a system already in use.

Attempted paths

Meta WhatsApp coexistence direct u2014 not supported without a partner. YCloud as partner u2014 number already connected to a business (GROW CRM).

Working path

Bot routed through GROW CRM on the existing number. Bot answers, admin can reply manually, Mindbody books for existing members u2014 nothing already connected is broken.

GPT-5.4 mini
OpenAI API
WhatsApp Business API
Mindbody
GROW CRM
Fitness guardrails

monitoringWhat shipped (showcase — no outcome metrics yet)

  • WhatsApp AI assistant live for FS8 Hillview, Lower Peirce, and Upper Thomson
  • Fitness / diet / workout / nutrition responses constrained by a studio-aware ruleset
  • Mindbody booking, scheduling, and rescheduling for existing members via WhatsApp
  • Dual handoff: user-requested and confidence-triggered escalation to human staff
  • Same Business number shared with GROW CRM, Mindbody, and manual staff replies
  • Account creation left human u2014 bot never invents a Mindbody membership

info Showcase entry: engineering story and foresight only. No messages-handled, booking-volume, or hours-freed figure is claimed until the bot produces defensible metrics after sustained live use.

flare The part they didn't know to ask for

XMARTECH solved a WhatsApp integration problem the client didn't know they had. The client asked for a fitness AI bot. XMARTECH discovered that their phone number was entangled across three systems — GROW CRM, Mindbody, and manual staff replies — and that no standard integration path could add a bot without breaking what was already connected. The three-attempt diagnosis (Meta direct → YCloud → GROW CRM routing) found the working path. The client asked for a bot. XMARTECH delivered the bot and the infrastructure archaeology required to make it possible.

// Three attempts. Two failures. One working path.

quizBuyer questions, answered

How does a fitness studio's AI assistant decide what it should and shouldn't answer?

A defined ruleset constrains every response to fitness, diet, workout, and nutrition u2014 biased toward FS8 Hillview, Lower Peirce, and Upper Thomson and their class types. New users get guidance and studio/class recommendations. Existing Mindbody members can book and reschedule in WhatsApp. New account creation stays human.

What happens when the AI isn't sure?

Two handoff triggers. The user can ask for a person at any time. Separately, the bot monitors its own confidence against the ruleset and escalates before a wrong answer ships. In a franchise under FIT House of Brands, a bad reply misrepresents more than one studio.

Why was connecting the bot the hardest part of building it?

The WhatsApp Business number was already claimed by GROW CRM, wired to Mindbody, and used for staff replies. Meta coexistence needed a partner; YCloud revealed the CRM claim. The working path routed the bot through GROW CRM so bot and humans share one thread without breaking what was already live.

tuneUnder the hood

GPT-5.4 mini via OpenAI API Custom fitness / nutrition / FS8 ruleset guardrails WhatsApp Business coexistence via GROW CRM Mindbody live booking + reschedule for members User-requested human handoff in-thread Confidence-triggered escalation before brand risk No new Mindbody account creation by the bot 3 Singapore studios on one shared number

A bot on a clean number is easy. A bot on a claimed number is the real build.

FS8 asked for a fitness WhatsApp assistant. XMARTECH delivered the assistant and the coexistence path that made it possible on the number they already depended on.

forumDirect conversation model

01

You share business goals, constraints, and deadlines.

02

You talk directly with the engineering team, not only account managers.

03

We turn strategy into a scoped delivery plan and execution timeline.

auto_graph Project-type focus

  • Context-constrained AI responses
  • Human handoff before brand risk
  • Shared-channel coexistence without breakage
arrow_back Back to all work Discuss a similar project arrow_forward