Overly manual prospecting
Sales teams were losing time searching for information, verifying contact details, following up at the right moment and reconstructing context before each outreach attempt.
Anonymised case, real data
Case Study: An AI Beauty Concierge for 1,700 Salons
This case study is based on a real deployment conducted for a major professional beauty player. Names, brands and sensitive details have been intentionally anonymised. What matters here is the business mechanics — the shift from fragmented prospecting to a manageable AI CRM — and the operational results achieved across a network of over 1,700 salons.
beauty AI agentAI hair salonbeauty AI CRM
1700+
salons covered by the target network
1645
active records consolidated after quality cleansing
9
autonomous agents orchestrated around the CRM
metric
1700+
salons covered by the target network
metric
1645
active records consolidated after quality cleansing
metric
9
autonomous agents orchestrated around the CRM
metric
3783
scoring logs available to prioritise the pipeline
Sector context
Professional beauty is a dense, local and fragmented market
The starting point was clear. A major professional beauty player needed to manage a network of over 1,700 salons spread across diverse territories, with very different commercial realities depending on city, team and the digital maturity of each establishment. In this type of market, data does not behave like it does in a centralised SaaS product. Contacts are scattered, information goes stale quickly, useful signals are local, and commercial decisions often hinge on details that never appear in a generic CRM.
The sector faces three structural difficulties. First, fragmentation. A large share of the market rests on independent or semi-structured establishments, each with its own pace, stakeholders and priorities. Second, the pressure to personalise. A standardised message performs poorly because the goal is not merely to contact a salon — it is to show that you understand its position in its environment, its price tier, its potential and its needs. Third, speed. As the network grows, record volume expands faster than the human capacity to properly qualify it.
That is precisely where a beauty AI agent becomes relevant. Not as a conversational gadget. Not as a chatbot bolted onto a website. But as an operational layer capable of consolidating data, hierarchising priorities, structuring actions and restoring a sustainable rhythm to prospecting. The goal of the project was not to produce another dashboard. The goal was to install a salon AI concierge capable of acting at network scale while preserving fine-grained business logic.
The mandate was therefore simple to state but demanding to execute. Build a beauty AI CRM capable of turning a fragmented database into a manageable commercial engine. Ensure that teams no longer spend their days searching, cleaning and guessing — but deciding, validating and closing.
The challenge
Too much data, not enough actionable context
The real problem was not an absence of prospects. It was the gap between raw data and useful commercial action. Internal audits showed approximately 8 minutes of friction before the first genuinely productive action in the CRM. They also showed 65 percent of records not yet enriched — 1,070 establishments on hold — and only 2.4 percent of email addresses directly usable at the outset. At that scale, manual prospecting becomes mechanically chaotic.
Fragmented and unreliable CRM
The data existed, but it remained partially enriched, inconsistently qualified and difficult to convert into concrete actions in the field.
Personalisation impossible at scale
Without reliable scoring, centralised history and automated sequence preparation, every message was an isolated effort — impossible to industrialise properly.
Why a standard CRM was not enough
A horizontal CRM knows how to store records. It rarely knows how to distinguish what makes a relationship valuable in a field market like professional hairdressing and beauty. Here, each record had to carry far more than a name and a phone number. It needed a qualification level, contact history, compatibility signals, a priority logic, a reading of geographic zones and the ability to guide the next action without requiring the user to reconstruct the context themselves.
The main risk was twofold. On one side, the sales team could exhaust itself handling insufficiently prepared data. On the other, the network could create an illusion of volume without generating a real commercial cadence. In this kind of configuration, chaos does not only show up in bugs. It shows up in forgotten follow-ups, cold records never revisited, under-used channels and the absence of robust prioritisation.
The weak signals blocking performance
- First commercial action too slow — approximately 8 minutes before having actionable context.
- 1,070 records awaiting enrichment at diagnostic, a backlog blocking scoring and outreach.
- Excessive reliance on email when only 2.4 percent of addresses were directly usable at the start.
- Reply rate not reliably tracked, due to the absence of normalised statuses and centralised channel history.
Deployed solution
A salon AI concierge, connected to a unified CRM and 9 autonomous agents
The response was not a simple interface redesign. The system was designed as a complete operational chain. A Next.js frontend for supervision, a Supabase database for data structure, and a layer of AI agents capable of discovering, enriching, scoring, prioritising and preparing outreach actions in the correct order. In other words, a system where intelligence is not limited to generating text, but genuinely organises commercial work.
Map the network
Centralise salons, their contacts, statuses, data sources and completeness level in a single, commercially actionable database.
Enrich automatically
Add operationally useful data — phone, email, website, city, record quality — then correct inconsistencies without relying on permanent manual re-entry.
Score leads
Assign a business score to each salon based on completeness, potential, fit and engagement signals, to prioritise what genuinely warrants action.
Prepare outreach
Structure contact sequences, schedule follow-ups, log replies and make response rates measurable channel by channel.
Manage in a single CRM
Give teams a clear view of the pipeline, next actions, hot leads, deals in preparation and the daily prioritisation decisions required.
What the 9 agents actually did
The core of the system rested on clear specialisation. Each agent had a precise business role, a defined cadence, a stable scope and output that the CRM could consume. This logic avoids the false multi-agent pattern where everyone does a bit of everything. Here, each agent prepared the next step.
1
Discovery of new establishments
2
Daily network refresh
3
Record enrichment
4
Scoring and prioritisation
5
Outreach action preparation
6
Follow-up and reply tracking
7
Commercial matching against business criteria
8
Data quality control
9
Operational reporting for human oversight
The business core of the solution
The CRM became the command centre. A salon record no longer merely stored information — it became a unit of action. Score, status, contact, sequence, follow-up, compatibility, history: everything was brought together in one place.
Lead scoring was connected to field reality. It combined record completeness, engagement signals and commercial priority logic. As a result, the team no longer worked from an alphabetical list, but from a decision queue.
Outreach was no longer a blind spot. Sent, replied, no-reply and scheduled statuses were tracked in the data. Reply rate became calculable by channel — and above all, actionable in daily management.
Results
Real metrics, finally actionable commercially
The first result was restored visibility. The system enabled management of a network of over 1,700 salons with a consolidated base of 1,645 active records at quality-check date. It relied on 3,783 scoring logs already available for use and 9 specialised autonomous agents. These are concrete metrics, not marketing promises.
1700+
salons covered by the target network
1645
active records consolidated after quality cleansing
9
autonomous agents orchestrated around the CRM
3783
scoring logs available to prioritise the pipeline
What changed operationally
Before deployment, the team faced a classic paradox. The network existed, the data existed, the commercial need existed — but productive time was being consumed by context reconstruction. The internal diagnostic measured approximately 8 minutes before the first genuinely useful action. In a prospecting environment, that kills cadence.
With the AI concierge, time is no longer absorbed by information retrieval. It shifts to validation and decision-making. Records are enriched continuously, priorities surface automatically, follow-ups are logged, response signals are traceable and next actions are visible the moment the CRM is opened. The time saving therefore does not reduce to an isolated figure. It translates into a transformation of daily cognitive load.
The second major shift concerns reply rate. At the outset, the real problem was not a bad percentage displayed on screen. It was the inability to measure that percentage properly. Once statuses were unified and history structured, reply rate became a manageable indicator by channel. That is a decisive turning point — moving from intuitive to instrumented prospecting.
Before / after, in business terms
before
1,070 records on hold — 65 percent of the stock still unenriched at diagnostic.
after
Backlog handled by a daily enrichment and prioritisation mechanism, rather than opportunistic manual catch-up.
before
Only 2.4 percent of email addresses directly usable at the outset.
after
Deliberate multi-channel strategy, with reply rate and follow-up tracking by channel.
before
First commercial action too slow — around 8 minutes of friction.
after
Context, score, history and next actions consolidated in a single interface.
before
The salon volume existed, but remained difficult to convert into a concrete pipeline.
after
1,700+ salons manageable, 1,645 active records consolidated, 3,783 scoring logs to guide execution.
Tech stack
A lean, robust stack built for business data
The project relied on a deliberately pragmatic architecture. No unnecessary abstraction. No gratuitous complexity. A Next.js frontend for the operator experience, Supabase for transactional and real-time data, and a layer of AI agents connected to the CRM to automate what needed automating.
Next.js
Supervision interface, business pages, CRM views, dashboards and management components. The goal was not visual impressiveness but surfacing priorities without friction.
Supabase
PostgreSQL database, action history, statuses, scoring, contacts, outreach and real-time synchronisation. The backbone of the beauty AI CRM.
AI Agents
Discovery, enrichment, scoring, follow-up preparation, quality control, reporting. The agents did not replace the CRM — they made it operational at scale.
Lessons learned
What this case reminds us about AI in real commercial environments
Raw volume is worthless without enrichment
A network of over 1,700 salons looks impressive on paper. In practice, as long as records remain incomplete, the team is not managing a network — they are handling a pile of cold prospects.
A vertical CRM must reflect field reality
In professional beauty, the salon name is not enough. You also need contact history, qualification level, commercial compatibility, location and the operational reality of the record.
Email alone cannot carry the strategy
The initial diagnostic showed only 2.4 percent of email addresses were directly usable. The right answer was not to hope for more emails, but to orchestrate genuinely multi-channel prospecting.
Scoring only has value if it triggers an action
A score tucked in a corner of the CRM helps no one. A score that reorders the work queue, suggests a follow-up and prepares a message becomes a business lever.
Humans remain central — just no longer in the middle of everything
Agents prepare, monitor, qualify and structure. Human teams validate, prioritise and close. That division is what makes the system sustainable.
CTA
You manage a fragmented network and your CRM can no longer keep pace
If you manage retail locations, franchisees, field partners or a distributed sales network, the issue may not be generating more leads. The issue is often turning existing data into an action system. That is exactly what a well-designed AI concierge can do.
Let us discuss your situation. We can show you how to structure an AI CRM around your business priorities — without exposing your data and without copying a generic model that does not understand your field.