Customer Intelligence for multi-site networks
顧客とのあらゆる会話を、
turned into a case with an owner
We run the channels your customers talk on and read every conversation into cases your network can act on — each carrying the sentence behind it. Automotive dealer networks are where it runs today.
30分のウォークスルーを予約市場機会
Your network already knows
what is about to go wrong.
Every day your customers tell someone on the front line what they want, what went wrong, what delayed their decision and why they may not come back.
What leadership cannot answer today
- Which customers need attention right now.
- Who missed the follow-up, and at which location.
- Where the revenue starts leaking away.
- 00 No layer
- 01 取り込む
- 02 理解する
- 03 振り分ける
- 04 Act
Customer side
AI voice bot Inbound · outbound
Direct call Inbound · outbound
Other signals
Email · Web forms · Feedback tools · Photos · Dealer notes · Social · DMS records · Car software
- Hidden delivered, nobody opened it
- Disconnected the call hangs on nothing case — owner — vehicle —
- Delayed three days, then the customer drives over
- Unstructured 38 seconds nobody can query
Network side
音声メモ Structured record Case 8842 0:38
Third week waiting for the part on order 4471 . Nobody calls, nobody tells me anything, and I need the car this week .
- issue
- Spare Parts
- about
- order 4471
- urgency
- high
- outlet
- Turku
- owner
- R. Salo
- state
- escalated
- due
- today
- attached to
- customer · vehicle
- sentiment
- negative
- psf
- 3 of 10
- nps
- 4 detractor
Readings · this conversation
Airfra network intelligence
Network sees every outlet · a dealer sees only its own
Follow-up friction Turku · repeated across cases
Client systems case written to 2 of 4
- CxMS
- ERP · SAP
- CRM
- DMS
CRM — customer file
J. Virtanen
- 12 Aug service booked
- 28 Aug part ordered · supplier delay
- Today case 8842 · escalated · sentiment negative
The future of network performance is not more communication. It is control over what communication already reveals.
What it answers
The pattern no single case can show.
Every case closes with the customer it belongs to. What it leaves behind is a fact about the outlet that produced it — and the layer holds those facts for every outlet at once.
Outlets head office view · an outlet sees only its own
- Helsinki
- Tampere
- Espoo
- Vantaa
- Lahti
- Kuopio
- Oulu
- Pori
- Joensuu
- Vaasa
- Jyväskylä
- Turku
- Kotka
- Rovaniemi
- Repeat, not incident Turku
- The same category coming back from one outlet is a supply line or a process, not three unlucky customers. One case cannot say which; the outlet's cases together can.
- Follow-up that never came Tampere
- Who answered, who did not, and how long the customer waited before asking again. The layer runs the channel, so the silence is on the record with everything else.
- One outlet pulling away Oulu
- Sentiment or time to resolve drifting from the rest of the network, while every individual case there still looks ordinary enough to close.
- Where it starts costing
- The step the detractors have in common. Revenue leaks at a point in the process before it shows up as a person or a quarter.
What we deliver
We supply the conversations,
not only what we read out of them.
-
01 Channels we run
Running today
Outbound as well as inbound, on lines we operate — not a listener attached to yours.
-
out Templates go out
in Replies come back — voice notes, photos and documents
- AI voice bot
-
out Places the call
in Answers it, in the language the customer chose
- Direct call
-
out One click from the open case to the customer
in A way back to your team from the web
-
02 Airfra
Running today
The layer that reads them. Every conversation is transcribed, summarised, classified by issue type and urgency, and closed with the next action to take — in whatever language it was held. Two of those readings carry the most weight.
- Sentiment
- negative · neutral · positive How the customer actually felt, read from what was said — not from a survey they never filled in.
- PSF & NPS
- 0–10 · detractor 0–6 Whether they come back. A weak score is an early signal of churn — and of the location that keeps producing them.
-
03 The system it lands in
Assembled with you
Cases, escalations, team tasks and analytics become measurable and governed.
What your systems receive
- 案件
- Escalations
- Team tasks
- Callbacks
- Location scorecards
- BI dashboards
- CRM / DMS updates
- Audit trail
- 現在
- Our own case system: cases, escalations, team tasks and analytics in one place.
- On request
- If you have no system of your own, or the one you have does not fit, we assemble it — what already runs, plus what you ask for. CRM and ERP follow the same path.
How you check it
Every verdict comes with
the sentence that caused it.
Example — one conversation from a dealer network
-
Airfra bot
Your service booking is confirmed for Friday, 14:00. We will message you when the car is ready.
-
customer · voice note
“They told me the car would be ready on Friday. Nobody called. I found out on Monday, when I drove over myself.”
The sentence the classifier quoted back as its evidence.
- Issue
- Delivery Delay one of 42
- Sentiment
- Negative
- Urgency
- High
- PSF
- 3 of 10
- NPS
- 4 detractor, 0–6
- Flags
- Callback requested, action required
- Stored with
- The quote it relied on, a confidence score, and the version of the model that produced it.
- Attached to
- The customer, the outlet, the salesperson and the vehicle the case is about.
- Logged after
- Every write that follows — who made it, from which value to which, and when.
The categories are the trade's own words, not abstractions — Delivery Delay, Late Handover, PDI Issue, Variant Mismatch, Registration Issue, Spare Parts, Warranty. Forty-two in all, and the classifier picks exactly one.
Who sees what
Everyone works in the same system.
Nobody sees the same thing.
The centre compares the whole network. A single location sees its own work and nothing more — including nothing of its neighbours' scores. In an automotive network that is the OEM and its dealers.
Head office sees
- Every case in the network, from every channel
- Locations side by side — sentiment, escalations, follow-up quality
- Where the same issue repeats across regions
- Who owns each escalation, by name
One location sees
- Only the cases assigned to it
- Only its own scores
- Its own escalations, tasks and callbacks
The boundary is not a switch in the interface. It is added to the query on the server for every role on the location side, so a view that is not allowed cannot be reached by editing the address bar.
独自の強み
ボットは簡単に真似できます。
ワークフローの蓄積はそうではありません。
- Transcription, not a summary
- One row per turn, with the speaker named — bot, customer, agent, system — and a single store behind all three channels, so a call and a chat read the same way.
- Voice notes are input, not attachments
- A voice message is transcribed and classified like any call, instead of sitting in the thread for someone to listen to later.
- Escalation is a state, not a flag
- None → In progress → Escalated → Resolved → Closed, with a named manager on each one. The layer raises the flag; a person decides and the decision is logged.
- The location side is its own build
- Not one interface with rows hidden from some accounts — a separate branch, with the scope written into the query behind it.
- Language belongs to the customer
- Not to a message or a setting. Every step after capture — summary, classification, reply — stays in the language that customer uses.
- Reach into systems without an API
- Where an interface is the only way in, we drive it — controlled automation rather than leaving the system outside the rollout.
- ワークフローの蓄積
- Each deployment leaves reusable playbooks, classification logic and escalation paths behind. That is the part a competitor cannot start with.
ユースケース
What it catches
in an ordinary week.
Eight scenarios
Drag or scroll
-
Delivery slips
WhatsApp
The date moves and nobody calls. The next message says so, and a case opens with an owner on it.
-
Handover complaints
WhatsApp
What went wrong at collection, classified and routed before a survey would have gone out.
-
Silent customers
WhatsApp · no reply
Someone who stops replying is a signal, not an absence.
-
Calls both ways
Direct call
Outbound from the case, inbound from the web — the same transcript either way.
-
Detractors, same day
NPS / PSF
A weak score reaches the person who can still act, while the customer is in the loop.
-
Repeat faults
Case history
The same fault, the same customer, a second visit — read as a pattern, not as two unrelated tickets.
-
Escalation with a name
エスカレーション規則
Not a shared inbox: a state, a named manager and a time on every step.
-
Location drift
Network comparison
One site pulling away from the rest on sentiment, follow-up or time to resolve.
Where this goes
The same loop,
run the other way.
Today the layer speaks up when a customer is about to leave — a detractor, a delay, a promise nobody kept. It already runs the channels, and it already knows what the last conversation was about.
Pointed forward, that is a reason to call first. A service coming due, a policy about to expire, a customer nobody has spoken to since delivery.
That is the direction, not a switch in today’s build: what runs now is the reading and the channels it runs on. The outbound side gets built with the first network that wants it.
想定するお客様
Built for leaders answerable for
customers other people serve.
If the people talking to your customers are not on your payroll — dealers, franchisees, importers, partner outlets — the conversation is where your visibility ends. Closing that gap should not cost those teams another portal to log into.
It runs today in automotive dealer networks. The same shape fits importers and national distributors, franchise and multi-site service operations, enterprise networks with field or branch teams, and the CRM, DMS and CX platforms that serve them.
お話ししたい方々
- 最高顧客責任者（CCO）
- Head of Network Operations
- カスタマーエクスペリエンス責任者
- アフターセールス責任者
- セールスオペレーション責任者
- デジタルトランスフォーメーション責任者
- CRM / DMS Product Owner
- Regional Performance Director
- エンタープライズ・イノベーション／AI責任者
Security & control
Who can reach what,
and where that gets decided.
- Access
- Three roles — head office, location admin, location agent. Accounts are invited from the head office rather than self-registered, and the first sign-in forces a password change.
- Sessions
- The session token carries the role and the location it belongs to, and every request is checked against it on the server before a row is returned.
- Channel keys
- Credentials for WhatsApp and the calling platform stay server-side. Nothing that can send a message in your name is ever held in a browser.
- Hosting
- EU region on Microsoft Azure: a static front end, a serverless API and a managed SQL database.
- Compliance
- GDPR-aligned by design, EU data residency, contractual DPA.
よくある質問
Before the first call
Is this a chatbot?
Do we have to replace our DMS or CRM?
What happens to conversations that are not in English?
Can we check why the AI decided something?
パイロットパッケージ
1つの地域から始める。
Scale on what it finds.
-
Day 0
One market, one dealer group or one aftersales workflow — on the channels those locations already use.
-
Every week
A report on what the layer is finding.
-
Day 30–60
A review: what the layer found, the classification and escalation rules it settled on, the integration plan for your DMS, CRM and BI, and a rollout plan for the rest of the network.
See what your customers are already telling you
1つの業務、1つの地域、1つのチャネルをお預けください — 顧客との会話の中に、どれだけのインテリジェンスがすでに眠っているかをお見せします。
- Built for multi-site networks
- CRM / DMS ecosystems
- EU域内でのデータ保管
- GDPR準拠