Contact center · Social housing
1 of 3
Sofratel
Customer relations director
200
housing operators
10,000 calls a day at peak, one agent fleet, zero entity-by-entity rebuild
- Done-for-you
- Multi-entity
- 24/7 emergency line
- Mitel
- French
The voice AI platform for multi-entity networks
Run voice AI across every entity you operate
A club, a practice, an agency, a brand or a client: CallShift deploys, monitors and scales one voice agent, or a whole fleet, with the same consistency across every entity of your network. Live in 1 month on average, 100,000+ calls a day, 80+ languages, EU hosted.
They run their voice AI across their whole network
They also trust us
Built for
Franchise networks (sports clubs, real estate agencies, driving schools)Multi-site groups (medical, dental, accounting practices)Multi-brand groupsBPOs that handle several clients
Multi-entity
One account, one agent, every entity.
An entity is one unit of your network: a club, a practice, an agency, a brand or a client. Head office runs them all from a single CallShift account.
01
One account, one invoice.
The whole network runs from a single CallShift account and is billed on a single invoice.
02
One agent, designed once.
A voice AI agent shared by every entity. Each one keeps its own information: name, address, opening hours, phone numbers and local offers.
03
Improved once, better everywhere.
An improvement validated at head office benefits every entity immediately.
04
Every line stays in its entity.
A phone line cannot be tied by mistake to the agent of another entity, for inbound or outbound calls.
What it gives head office
Steering
- Call logs, reports and consumption can be filtered by entity.
- A consolidated view of the network lets you compare entities with each other.
- An Entities page shows, for each entity, its agents, its lines and its webhooks. The No entity filter shows what is still to be attached.
- Differentiated access rights for your teams: quality control, billing, administration.
Rollout
- A new entity is operational in minutes, with no redeployment of the agent.
- What already exists is attached in one click from the dashboard lists.
- The feature is optional: as long as no entity is created, nothing changes and your existing setup keeps working.
Line security
- A line attached to an entity only works with the agents of that entity or the agents shared across the network. A club's line cannot be tied by mistake to another club's agent.
- The check applies wherever a line is associated: answering inbound calls (telephony settings), outbound call campaigns, the dashboard and the API. It is checked again at the moment of each outbound call.
- On outbound, each call leaves from a line of the entity it is placed for. The network knows which entity calls whom, which helps comply with GDPR and the EU AI Act and clarifies responsibilities.
- Post-call data only goes to the destinations (webhooks) of the same entity.
Integration
- Entities are created and managed through the API, and agents, lines, webhooks and SIP trunks can be filtered by entity. The network's CRM or middleware retrieves everything entity by entity.
- A SIP trunk dedicated to a single location automatically takes the entity of its lines.
Protected operations
- An entity that still contains agents, lines or webhooks cannot be deleted by mistake. Its items are moved in one action to another entity or to the account, never deleted.
Testimonial
We entered our partnership with CallShift.ai with complete peace of mind. They are a quality partner, close to their clients, and show great adaptability in every circumstance. The tailor-made solution they proposed fits our expectations perfectly, as well as those of our own clients.
Julien Dusart
Director of contact center operations, Sofratel
Up to 10,000 calls a day at peak
Integrates with
And 1000+ more, or your custom IT stack or telephony...At a glance
100,000+ calls/dayAverage rollout: 1 monthGDPR CompliancyISO 27001 certifiedHDS health-data hosting availableHQ & Data hosted in 🇪🇺 (Paris)In production since 2024
Calculator
What your network loses on missed calls
Four numbers you already know. The arithmetic is the part nobody runs, because for one entity it looks like noise and across a network it is a line item.
78,000
calls never reached a human, per year
€3,900,000
of value never picked up
Based on 260 working days a year. Everything is computed in your browser, nothing is sent or stored.
Pricing
Pricing built around your volume.
Pilot
10 to 30 entities
from €40 per entity / month
Network
30 to 150 entities
degressive, quoted on volume
Group
150+ entities
custom, volume-committed
Platform and services are two separate lines. The services line drops off the contract when you bring account management in-house.
01
Pay for value, not promises.
The contract is built for multi-entity networks: pricing scales with the number of entities actually deployed. A pilot is priced like a pilot, a network like a network, never a flat licence disconnected from the field.
02
Done-for-you first, autonomous when you choose.
Platform and services are two separate offers. At launch, our team runs everything done-for-you and takes your project live in 1 month on average; once you're ready, you bring account management in-house, and that line simply drops off the contract.
//claude-skill
Run CallShift from Claude.
Drop our skill into Claude and drive your whole environment in plain language: list and duplicate agents, review a campaign's ROI, add contacts, assign SIP numbers, wire up webhooks. It talks to our API for you.
Grab it here
Free · works in Claude and Claude CodeArchitecture
Pipeline orchestration or speech-to-speech? The call is made use case by use case.
Two voice AI architectures, two opposite promises. Pipeline orchestration (STT → LLM → TTS) turns every turn of speech into text first: that is determinism. Native speech-to-speech keeps audio end to end: that is look and feel. Neither wins everywhere, and getting that arbitration right, far more than the model itself, is what decides a rollout.
Pipeline orchestration · STT → LLM → TTS
Favour determinism.
The intermediate text is a control point: what gets transcribed can be inspected, constrained and replayed exactly.
- Complex tool-calling flows, conditional branching, long forms
- Constrained scripts where every mention has to be said verbatim
- Every stage picked and swapped on its own: STT, LLM, voice
- Native transcripts: audit, QA and call review all traceable
Native speech-to-speech · audio → audio
Favour look and feel.
One native audio loop, no transcription layer: the model hears tone and answers with its own.
- Reception, qualification, follow-ups: wherever a natural delivery converts
- Low latency, clean barge-in, emotional nuance heard and returned
- Emails and phone numbers captured with no transcriber to mangle them
- Tone, accent and style steered from the prompt, on the fly
Speech to speech in production
uMinds · 10+ creators
A dozen creators maximizing their funnel. One of the few platforms running Gemini 3.1 Flash Live speech to speech while integrating with custom SIP telephony at scale.
In a head-to-head A/B test run live in production, their Spanish speech-to-speech agents came out about 20% ahead of the Cartesia + GPT-4.1 mini + Deepgram pipeline they ran before, on the metric that pays: performance. Same campaign, same numbers, one architecture swapped.
- Gemini 3.1 Flash Live
- Speech to speech
- Custom SIP at scale
- Funnel conversion
- 50,000
- calls completed in under 3 hours
- 45
- concurrent lines, live in production
- 5 CPS
- on a single number, custom setup
- 3
- sub-accounts run from one place
Figures reflect one customer's production configuration and results, not a contractual commitment.
One of our specialities
The call
Choosing between the two is the job.
We run both architectures in production, on the same networks and sometimes on the same fleet. During onboarding our engineers walk your use cases one by one and make the call with you: determinism here, look and feel there. That arbitration is what takes a rollout live.
01
We map your flows
Use case by use case: what has to be exact, what has to sound human, what has to be both.
02
We make the call with you
Each flow leaves with its architecture, its models, its latency budget, and the reason behind the choice.
03
We re-run the arbitration
A production A/B test settles whatever is still in doubt; switching a flow's architecture never means rebuilding it.
Services
Done-for-you is a rhythm, not a launch.
Technology that ships and is never spoken of again is technology that decays. Every network we run has a cadence behind it, from day one onward.
| Rhythm | What CallShift does |
|---|---|
| Every day | Automated call report by email, per entity and consolidated across the network: volumes, call types, reasons, transfer rate, durations. Quality flags scored on 100% of calls. Configurable alerting on the cases you define, at a threshold that does not drown you. |
| Every week, first three months | Call listening with your own teams, log of friction points, adjustments to the script, the escalation rules and the knowledge base. Fixes in production within 48 h. |
| Every month | Performance review with your teams: what worked, the calls worth your attention, the weak spots, the field feedback. This is where a seasonal spike or a call reason that did not exist before shows up, and where we decide together to evolve the template. |
| Every quarter | Technology review. A better model ships roughly every three months: we test it, benchmark it on your own calls, and migrate when the gain is real. Non-regression validated in staging, no setup fee, no renegotiated rate. |
| Continuous | 2 h response SLA, telephony and connectivity maintenance, upkeep and evolution of the template shared by all your entities, direct WhatsApp channel, three access roles including read-only for your audits and QA. |
The shared template is a common asset: every improvement validated on one entity benefits all the others, with no redeployment and no re-invoicing. The larger your network, the harder this line works for you.
Six pillars
Six things that break voice AI at scale. We already solved them for you.
Cost, Orchestration, Model, Ease of use, Telephony, Scale. Every one of them has sunk a voice AI project somewhere. Our engineers have streamlined all six.
Discover how each Voice AI pillar is managed
The most marketing-washed metric in the industry. Billing increments? System prompt token size impact? Our engineers have streamlined it for you.
Orchestration is tricky: a custom STT with +30% latency means retuning your endpointing by 30% or turn-taking falls apart. Our engineers have streamlined it for you.
The right balance between latency, intelligence and context window; long prompts and long conversations change the equation. Our engineers have streamlined it for you.
Two ways to manage: our online builder, or prompt control from your favorite LLM. Our engineers have streamlined it for you.
Integrating with your telephony provider is a serious challenge (SIP, IPBX, transfers). Our engineers have streamlined it for you.
Delivering at scale with reliability: tuning servers x AI agents x telephony together. Our engineers have streamlined it for you.
Objections
What a head office asks before it starts.
Each of our entities has its own hours and offers. Is that handled?
Yes. Each entity keeps its own information: name, address, opening hours, phone numbers and local offers, while sharing the same agent.
How do we know which entity performs best?
Call logs, reports and consumption are filtered by entity, and a consolidated view of the network lets you compare entities with each other.
What if someone picks the wrong line?
The association is refused, in the telephony settings, in an outbound campaign, in the dashboard and through the API. A line only works with the agents of its own entity or the agents shared across the network.
Who is responsible for an outbound call?
Each call leaves from a line of the entity it is placed for. The network knows which entity calls whom, which helps comply with GDPR and the EU AI Act and clarifies responsibilities.
Everything is already set up on our side. Do we have to redo it?
No, there is nothing to redo. You create the entities, then attach what already exists. Until an entity is created, nothing changes.
Our middleware has to follow. Can it?
Yes. The API filters agents, lines, webhooks and SIP trunks by entity, so your CRM or middleware retrieves everything entity by entity.
Next step
See it on your own network.
Book 30 minutes with a CallShift engineer: we start from your real call flows and show you how a consistent, supervised, multilingual agent rolls out across every entity you operate.