AI Operations Hub: how a medical-aesthetics chain grew its operations into an AI-native platform
One AI-native operations hub replacing a traditional CRM — AI does 80% of the heavy lifting, people make the final call.
A custom AI operations hub for a dual-brand medical-aesthetics chain (11,000+ active clients): one platform replacing CRM + inventory + staff portal + functional medicine (28+ toggleable modules), with AI woven into every repetitive step.
Client: a dual-brand medical-aesthetics chain, multiple clinics, 11,000+ active clients. Delivery: a custom NEXT AGENT AI operations hub that fully replaces a traditional CRM.
The client is anonymized. Only architecture, workflow, and delivery outcomes are shown — no client- or patient-identifiable information.
The pain of a traditional CRM
The chain ran on an industry-standard SaaS. It worked — but they paid for its limits every day:
- Closed, and priced per location. Per-clinic, per-month subscriptions stack up as you grow, and the system is a black box you can't change. A localized workflow means waiting on the vendor — or working around it in Excel. The data is yours; the control isn't.
- Siloed systems. Booking, inventory, staff scheduling, and the functional-medicine line each lived in a separate tool. The same client and the same dollar existed in four systems and never reconciled.
- Everything by hand. Building catalogs, entering packages, scheduling, charting, client notes, daily reconciliation, consent forms — all repetitive manual work. Slow, costly, error-prone, with the most valuable doctor and manager time spent on data entry.
- Reports you couldn't trust. Cash revenue was mixed with membership/package redemptions, so the numbers leadership saw were inflated — decisions built on the wrong basis.
- No AI. Repetitive work couldn't be automated, because the system had no intelligent layer that understood the business.
The solution: an AI-native operations hub
Instead of buying another SaaS, we grew the chain's operating processes themselves into a system, with AI embedded at every repetitive step. One platform replaced four — CRM + inventory + staff portal + functional medicine — as 28+ modules you can toggle on demand. At its core is an “AI operations spine”: processes are code, the LLM fills the blanks, and critical actions pass through a human review queue. AI handles 80% of the heavy lifting; people make the final call — controllable, auditable, and compliant (which is essential when handling medical-privacy data).
How AI works inside the automated workflows
Not a chat box bolted on in the corner — woven into every part of daily operations:
- Spoken catalog setup — a manager just says “create a Mother's Day package: full-face Thermage plus neck, with one hydro-facial, valid for a year,” and AI parses it into a structured catalog, auto-linking services, devices, consumables, and rooms. Days of data entry compressed to minutes.
- AI clinical documentation — the doctor dictates; AI drafts charts, treatment records, and post-care instructions; consent forms are auto-drafted per treatment.
- Smart client notes — high-value clients automatically get a “spend analysis + follow-up suggestions” written back to their profile, so sales follow-ups have a basis.
- Smart task capture — paste a block of meeting notes and AI extracts tasks, owners, and milestones, generating Kanban cards in bulk.
- AI sales insight & copywriting — the pipeline is analyzed automatically; multilingual client messages are generated, reviewed by a human, then sent.
- Speech-to-text — HR interviews and client consultations are auto-transcribed and archived, searchable and traceable.
- AI self-audit & self-test — the clearest sign of how deep the AI goes: multiple AI agents audit each other's code and drive end-to-end tests with real client data; one pass found and fixed 42 defects across cross-clinic data isolation, financial accounting, and concurrency safety. AI doesn't just run the business — it continuously safeguards the system's own correctness and security.
Outcomes
- Four siloed systems → one unified platform (28+ modules)
- An owned platform — no per-location monthly fees; change workflows whenever you want
- Master-data entry: days → minutes (AI spoken setup)
- Strict separation of cash vs. redemption — books that match reality
- Enterprise-grade per-clinic data isolation, hardened by multiple AI audit passes
- 2,000+ automated tests on continuous guard
Why NEXT AGENT
We don't sell generic SaaS, and we don't slap an AI shell on a legacy system. We use AI to grow your operating processes into a system that's yours — AI-native, from the first line of code. If you're also being dragged along by something expensive, closed, and all-manual — this is another way.
Technical breakdown
- Vertically sliced into 28+ modules by business capability: each module owns its tables (module-prefixed), APIs and pages; cross-module table access is banned — modules hold only foreign IDs and talk through public APIs and an event bus, so changing one module never ripples across the system.
- Platform layer separated from business logic: auth/RBAC, multi-tenancy, a transactional event bus (outbox), audit, notifications, file storage and the AI gateway sit in the platform layer, reused by every business module.
- Zero hard-coded client differences: shared modules never contain if(client X) — variance lives in config options and extension points (events/hooks); field-level differences ride on core columns + JSONB custom fields.
- Financial integrity built in: cash-like collections and entitlement redemptions (packages, memberships, points, gift cards) are booked separately, and cash is counted exactly once across the stored-value purchase→redemption cascade — reports reconcile by construction.
- Schema-driven clinical records: treatment types and form fields are pure template configuration (JSONB schema), so new treatments launch with zero code; signing locks the record with immutable snapshots for medical compliance.
- External systems (booking SaaS, EMR, payment gateways) attach through an external-ID mapping table plus optional sync extensions — the core never knows about third parties, so any vendor can be swapped without touching the trunk.
Reusable lessons
- A module boundary = whichever tables change together when one requirement changes. Slicing by that single test produced 28+ modules with clean, cycle-free dependencies.
- Cross-module actions like merging client records become events: the owning module migrates only its own tables and broadcasts; every other module re-points its rows by subscription — a 25-table monolith cascade collapses into a clear contract.
- Inventory multiple projects into a reuse matrix first (any capability appearing ≥2 times is an extraction candidate), then extract versioned shared modules in batches — the resulting library lets the next clinic or education client assemble from modules and start in weeks, not months.
- The right way to embed AI: process as code, LLMs fill the blanks, and critical actions pass a human review queue — AI does 80% of the heavy lifting without sacrificing auditability or compliance.
Modules used
Want results like these?
Tell us about your business and we'll scope a proof of concept.