The Proof IT Contact Centre Platform.
A contact centre that runs entirely in a web browser: calls, WhatsApp and email on one screen, an AI assistant that answers when nobody is free, live transcription that guides the operator, and every conversation, customer, task and figure in one place — for one country or several.
What this white paper covers.
The platform model and architecture, the operator workspace, voice from ring to wrap-up, AI on the live call, written channels, customer records and tasks, integrations, administration, reporting, outbound, distributed staff, and security.
The whole contact centre in one browser tab.
An operator signs in with their corporate account, puts on a headset, and from one screen takes and places calls, answers WhatsApp and email, sees who the customer is and works the follow-ups. Alongside them, AI transcribes, answers when nobody is free, guides them through the organisation's own procedures, and writes the record of what was agreed.
Every contact, one workspace
Calls — taken by AI voice agents or by people — messages and follow-up tasks are worked from one screen with the customer's full history beside them.
AI on the live call
Live transcription drives a procedure checklist, looks up identifiers spoken aloud, and writes the post-call summary.
AI when the queue is deep
The assistant establishes what the caller needs and arranges a callback or hands over to an operator who has already read the transcript.
What sets the platform apart
An engineering company built for regulated, high-volume operations.
Proof IT combines deep software engineering, systems integration, data migration and applied AI — with long experience building platforms for insurance, mobility and other document- and process-heavy industries where every conversation and every promise has to be on record.
Core competencies
Operational software at scale
Systems that run products, contracts, claims, documents, finance and partner processes for high-volume operations.
Applied conversational and document AI
Voice agents, live transcription, document intelligence and automated case handling — always under human governance.
Enterprise connectivity
CRM, booking and ERP systems, insurers, telephony carriers, messaging providers and corporate directories via open interfaces.
Configuration & production support
Setup, integration, testing, training, go-live, hosting, helpdesk and continuous improvement after launch.
Built from production components. The Contact Centre Platform is assembled from components Proof IT already runs in production — voice, transcription, WhatsApp, email, summaries, integrations — and configured to the organisation rather than written from scratch.
Contact operations lose control in the gaps between tools.
A telephone line forwarded to a mobile, an outsourced call centre, a shared mailbox and a WhatsApp number on somebody's phone each work on their own. Together they leave the organisation without a record, without figures and without a single view of the customer.
| Area | Typical operational reality | Business consequence |
|---|---|---|
| Forwarded lines | Published numbers ring on desk phones or mobiles at carrier level; nobody sees whether a call was answered or what was said. | No statistics, no recording, no record of the promise made. |
| Queues and peaks | Callers wait, hang up, or leave a voicemail somebody listens to the next morning. | Abandoned calls, lost demand, callbacks that never happen. |
| Several countries | Each market runs its own numbers, staff and tools; a night shift cannot cover more than one. | Duplicate cost, uneven service, no combined view. |
| Language | Callers land on an operator who does not speak their language and are passed around or asked to call again. | Frustration, longer handling, repeated calls. |
| Knowledge | What to ask on a roadside call, a claim or a complaint lives in the heads of experienced staff. | New employees miss questions; the downstream department chases the caller. |
| Written channels | WhatsApp and email are handled outside the telephone system, by different people, with no link to the call history. | Two histories per customer; work falls between the channels. |
| After the call | Summaries are typed by hand, if at all; the CRM and the department that acts on the call learn about it late. | Re-keying, drift between systems, lost follow-up. |
Accounts, number groups, operators and queues.
The account is the boundary of everything; nothing crosses between accounts. Within an account, the number group is the organising principle — a country, a department or a service line — and routing, AI behaviour and reporting follow from it.
| Element | Behaviour |
|---|---|
| Number group | Holds one or more telephone numbers, has operators assigned to it, carries its own queue, its own AI behaviour and its own line in the reports. |
| Telephone number | Belongs to one group and carries a label — what an operator sees when choosing a caller ID for an outbound call. |
| Operator | Belongs to as many groups as needed; sees the queues, conversations and reports of those groups and no others. |
| Queue | Per group, oldest caller first, visible live to the operators of that group and to management. |
| AI behaviour | Switched on or off per group, with its own queue-depth, waiting-time and out-of-hours thresholds. |
| Wrap-up pause | Set in seconds per account, from none at all up to ten minutes, before the next call is offered. |
What the model makes easy
Adding a country or a department
A new group with its own numbers and operators. Two published lines can share one group; the queue still shows which number was dialled.
One night shift for several countries
An operator who speaks several languages is attached to every group and answers everything — the pattern for covering several markets with a small team out of hours.
Different rules per market
The assistant answers first in one country, only out of hours in another, and not at all in a third — a threshold setting, not a different product.
Interface languages. The operator interface is available in English and the local languages of each market, with further languages added per deployment; the choice is stored on the platform and follows the person to any machine. The language of the conversation is handled separately (pages 11, 13).
Every channel meets in one coordinator; every call passes one media anchor.
Audio does not travel directly between caller and operator: both legs terminate on the platform, which mixes them. Recording, live transcription, an AI voice mid-conversation, hold audio, warm transfers and in-call translation all follow from that one property.
One screen: transcript, guidance, customer.
A fixed navigation carries Workspace, Call collaboration, Queue and AI, Tasks, Customers, Inbox, Reports and Admin, beside a top bar with live queue depth and service level, a dialpad reachable from anywhere, the post-call countdown and the operator's availability. During a call, three panels sit side by side.
Flat tyre, Liepāja highway km 17, vehicle LV-1234, caller safe on hard shoulder. Awaiting decision: tyre change or tow.
Illustrative layout, not a screenshot of shipping software. Names and numbers are fictitious.
The call, from ring to wrap-up.
A caller dials a published number. The platform answers immediately with the organisation's comfort audio, records which number was dialled, matches the calling number against customer records, reads the customer's assets and previous contacts from the organisation's own system, and places the call in the group's queue.
Operator answers in the browser
Recording, live transcription and procedure guidance start with the conversation. Call controls sit above three panels — transcript, summary, customer.
Call ends — the record is written
A summary is drafted automatically and edited by the operator, the recording is playable, a follow-up task is created if needed, and the summary is sent to the organisation's own system.
Wrap-up pause, then available again
A configured number of seconds with the countdown visible and endable early — time to finish the note, breathe, and take the next call.
Transfer, conference, bring anyone in — without leaving the platform.
Outbound calls are placed from the same screen that answers inbound ones, from any caller ID the operator's groups permit, and get the same recording, transcript, guidance and summary. Outbound capacity begins at ten simultaneous calls and rises to whatever the operation needs.
| Control | Behaviour |
|---|---|
| Answer · decline · end | The basic controls, in the browser, with hold and resume as often as needed — the caller hears the organisation's own hold audio. |
| Warm transfer | The operator consults an available colleague while the caller stays connected; the call moves across only when the colleague answers, carrying its recording, transcript and history. |
| Blind transfer, off-platform transfer | Passed straight through without consultation — to a colleague, or to any telephone number: a regional partner or a specialist who is not a platform user. |
| Conference | A colleague is added to a live call; all three are on one conversation, the colleague leaves without ending it, recording and transcript continue uninterrupted. |
| Bring in an external mobile | The operator presses a button, the platform calls any mobile number and joins that person in. The call never leaves the platform: recording, transcript, timing and statistics continue, and nothing is installed on the recipient's side. |
| Ownership | Exactly one operator is in charge at any moment, responsible for the summary and the follow-up. Ownership is recorded and passes to a remaining participant if the owner drops out — who is told so. |
| Internal calls | Operator to operator, without going out to the public network. |
| Recording & playback | Every call is recorded on the platform, not the operator's machine — which is why it survives transfers, conferences and bridged mobiles. Played in the browser from the call, the timeline or a task; downloaded as audio under the same permissions. |
Why the call is anchored. Because both legs terminate on the platform, an AI assistant can join a live conversation, a colleague can be conferenced in, and a technician on a mobile can be bridged in — and the one recording and the one transcript cover all of it.
Menus, out-of-hours, language routing and triage.
What happens before an operator answers is a set of configurable front-door behaviours: a keypad menu, an out-of-hours assistant, voicemail where a recorded message is what the organisation wants, and three ways of putting a caller in front of somebody who speaks their language.
A menu before the queue
A spoken list of options; the caller presses a digit and reaches the group behind it. Keypad tones are captured throughout the call, so a menu also collects a satisfaction score at the end.
An assistant instead of a voicemail box
A conversation producing a task with a name, a reason and a callback time. Voicemail remains available: the recording lands as a task with its audio attached.
Audio uploaded per group
A spoken message, a queue announcement or music — so a roadside line greets callers differently from a reservations line.
Emergency and roadside
Answered immediately by the assistant, which establishes what happened, where and what is needed, then connects the caller onward to the right regional partner. Intake, connection and outcome stay on one record.
Language routing — three mechanisms, combined as preferred
| Mechanism | How the language is established | What the caller experiences |
|---|---|---|
| A number per language | Before anyone answers — the number dialled decides the group. | They dial the English or the local-language line and reach people who speak it. |
| A first-line question | At the front door, asked by a keypad menu or by the AI assistant. | They state or select their language and are routed accordingly. |
| Transfer to a colleague | During the call, by the operator — or live translation on the line (page 14). | A warm handover, without having to call again. |
Keep your numbers. Stay independent.
The recommended way onto the platform is not to port published numbers. One carrier number is taken per country, and the existing published numbers are permanently forwarded onto it. Nothing changes in the organisation's published contact details, and the numbers are pointed somewhere else by changing a forwarding rule rather than by migrating.
One per country
Taken on the platform's carrier, attached to the country's group.
Published numbers forwarded
Permanently, at the existing operator. Callers dial what they always dialled.
Caller ID confirmed
An automated call back with a code, once per number, against the organisation's own carrier.
Outbound shows your number
Operators and campaigns dial from the platform, the customer sees the number they know.
Port only if you choose
Porting is a consolidation decision taken after the platform is in use, not before.
Capacity and carrier independence
Effectively unlimited
What a caller meets when every line is busy is a policy set per country: a busy tone, continued ringing, a recorded announcement, or the AI assistant.
Starts at ten simultaneous calls
Raised on request to whatever the operation needs. Bandwidth per call is small enough that an ordinary office connection is not the limit.
Interchangeable connections
Each carrier connection declares which capabilities it supports; changing carrier or running more than one is a configuration exercise, not a rebuild.
No live line needed
A browser-only connection and simulated inbound calls exercise routing, queueing and assistant behaviour before a real number is attached.
Live transcription, and an assistant that answers when nobody is free.
The conversation is transcribed as it happens, each line attributed to the caller, the operator or the assistant — the foundation of guidance, lookup, summary and translation on the next page. Assistant and human turns appear in one thread, so an operator taking over reads what was already said.
When the assistant engages — per group, per country
More than N callers waiting
Takes the next caller instead of leaving them waiting.
A caller has waited longer than N seconds
Apologises for the wait, asks the reason for the call.
No operators on shift
Every call answered; every one produces a task, not a voicemail.
What it does, and what the caller decides
| The caller | What follows |
|---|---|
| Accepts a callback | The assistant confirms a convenient time. One task appears in the worklist: the caller's own words as title, the assistant's notes, explicit consent, the due time, a link to the recording and transcript. |
| Prefers to wait | They return to the queue with their original position and waiting time preserved. |
| Asks for a person | At any point. It gives their queue position or asks when to call back — nobody is kept talking to a machine against their wishes. |
| Is offered an operator who just came free | It asks whether to put them through and transfers only on agreement; the operator arrives reading the transcript. |
What it knows, how it sounds. It is given the organisation's own material per group — services, conditions, procedures, opening hours, what to do in each situation — so it speaks about that organisation, not in generalities — in whatever language the caller uses, without being told which — and always says it is an AI. Four pacing dimensions are tuned on real calls: speaking rate, interruption readiness, end-of-speech sensitivity and answer length.
The operator is guided, not left to remember.
While the conversation runs, the platform recognises what the call is about, finds the matching procedure in the organisation's own library and puts it in front of the operator as a checklist. As each point is covered, the live transcript ticks it off; whatever is still open stays visible until the call ends.
| Call type | What the checklist establishes |
|---|---|
| A collision | Vehicle registration, exact location, injuries, whether the vehicle is drivable, whether the police were called, photographs of the scene, the other party's details. |
| Roadside assistance | Registration, precise location including road and kilometre marker, whether the caller is safe, warning triangle and vest, tyre change or tow, where the customer needs to get to, what is promised about arrival. |
Recognition is reliable across roughly ten well-defined call types rather than a hundred: setting it up means naming the call types that matter and the checklist behind each.
Identifiers spoken aloud
A registration number, a contract number, a name — picked out of the running transcript and looked up in the organisation's system while the call is still going. The asset, the customer and the previous contacts appear without anyone typing.
Both directions, on the line
Between any two languages the operation works in — Spanish and English, German and French — with roughly half a second of delay. The caller speaks English, the operator hears Spanish, and back. It covers the vocabulary of ordinary service conversations.
A few lines, owned by the operator
Who called, about what, what was agreed, what happens next — written in the call's language. The operator corrects, deletes or adds, and only then is it stored and sent onward.
Templates per call type. A roadside summary is required to state the registration, the location and the promise made — "a tow truck was promised" — so the department receiving it reads the same structure every time. The full transcript and the recording remain for anyone who needs the detail.
Two speech-recognition modes. One recognises whichever language is actually spoken, whatever it is; the other is pinned to a known language for higher accuracy on names, addresses and numbers.
One inbox, the same people, the same records.
WhatsApp and email arrive in a shared inbox, worked by the same operators against the same customer records, tasks and reports as the telephone. A call on Monday and a message on Tuesday are one history, not two systems.
| Element | Behaviour |
|---|---|
| Filters | Needs reply · Waiting on customer · Mine · Resolved · All — plus a channel filter across all, email and WhatsApp. |
| Claiming | "This one is mine" — an operator takes a conversation so colleagues see it is handled; it is assigned to a named colleague or released back. |
| Thread state | Explicit, so a supervisor sees what is genuinely outstanding as against what waits on the customer. |
| Actions | Reply in the thread, create a task from it, resolve, reopen when the customer writes again. Delivery state and timestamps per message. |
| Duplicates & visibility | The same inbound message delivered twice appears once. An operator sees the conversations of their own groups, as they see their own queues. |
Seconds in, seconds back
Inbound messages appear within seconds; a reply lands on the customer's phone. WhatsApp allows a free-text reply only within 24 hours of the customer's last message — the platform shows that window plainly, so an operator knows before writing. One number, or one per country, organised into groups.
Your own Microsoft 365 mailboxes
Incoming mail arrives continuously and replies are sent from the same mailbox, so the customer sees the address they wrote to. One support mailbox or one per country, attached to the corresponding groups.
AI pre-reading and prepared replies. Each inbound message is classified for topic and urgency and summarised before an operator opens it. Straightforward cases become a one-click reply the operator reviews and sends; anything needing a person arrives with structured context — who the customer is, what they asked, what has already happened. Replies leave the platform only when a person sends them.
One customer, one timeline, one worklist.
A customer holds several identities — the mobile they call from, a work email, a WhatsApp number. Whichever they use, the platform resolves it to the same customer, shows the match before the operator answers, and asks the operator to choose only when a match is genuinely ambiguous.
Long histories stay readable. Summaries in place of full entries for older interactions, grouping by day, hiding anything older than a chosen period, and filtering by channel — configurable, and combinable.
Tasks — the unit of follow-up
| Property | Behaviour |
|---|---|
| Where they come from | The assistant after an intake call, an operator during or after a call, a message thread, a customer record, or a campaign call where the person asked to be called back. |
| What is attached | The customer and their number, the assistant's notes, callback consent, the originating call with its recording and transcript, a due date and time. |
| Ownership | Claim next takes the oldest due task in the operator's groups; it then belongs to them and leaves everyone else's queue. Two operators cannot claim the same task. |
| Actions & views | Open the recording, call back in one action, reschedule, resolve, return to queue. Overdue tasks are flagged to operator and supervisor. Views: Queued · Mine · Done. |
Read in at the start. Write back at the end.
The platform does not ask an organisation to move anything. It reads what the operator needs when a contact begins and writes back what the rest of the organisation needs when it ends — summaries and contacts, not raw material. Both platforms stay usable at once.
| System | Read into the platform | Written back | When |
|---|---|---|---|
| Customer / asset system of record (CRM, booking) | Customer, assets and vehicles, agreements and contracts, previous contacts | New and updated contact identities, customer data, interaction summaries | Read on every contact and on an identifier spoken mid-call; written on operator confirmation |
| A second internal system used by a downstream department | Optional, at the scope the organisation defines | Contacts and interaction summaries from calls, WhatsApp and email | At the end of each interaction |
| Microsoft 365 mailboxes | Incoming mail from the support mailboxes | Outgoing replies, from the same mailbox | Continuously |
| Incoming messages to the organisation's numbers | Operator replies | Continuously, within WhatsApp's own reply rules | |
| Corporate directory | Who the user is, which groups they belong to | Nothing | At sign-in |
| Other systems | Whatever an interface is built for | Call, message, task and summary events published outward | On demand, or as events occur |
One button, one person
The operator reviews and edits the summary, then presses one button. Nothing crosses that a person has not looked at; full transcripts and email threads stay on the platform.
No drift between systems
A new telephone number, a corrected email address or a newly created contact is pushed back without anyone remembering to do it.
Existing corporate accounts
No separate password; removing a corporate account removes access here. Several identity providers at once, with account linking — for organisations running more than one directory.
Everything an administrator configures, set in the browser.
Roles are operator or administrator. Administrative functions are closed to operators and enforced on every request rather than hidden in the interface — as is listening to a recording, downloading audio or reading a transcript outside one's own groups.
| Configurable | Form |
|---|---|
| Operators | Identifier, display name, email, role, group membership by checkbox. |
| Number groups & numbers | Groups created and named; numbers assigned to a group with a label — what operators see when choosing a caller ID. |
| AI assistant | On or off per group, queue-depth and waiting-time triggers, out-of-hours answering, the organisation's own material, pacing. |
| Wrap-up | Seconds after each call, per account, from none up to ten minutes. |
| Service-level threshold | The seconds within which a call counts as answered on time, and the target below which figures are highlighted. |
| Comfort, queue and hold audio | Uploaded per group. |
| Retention | How long recordings and transcripts are kept, and how much longer the summary. |
| Consent | Which consents are captured, and how the platform behaves when one is refused. |
| Testing | Simulated inbound calls, exercising routing, queueing and assistant behaviour without a live telephone line. |
The same live picture
Management sees every group's queue, who is online, who owns which call, and what is outstanding in the inbox and the worklist — from the same screens operators use.
Hard account boundary
Operators, groups, numbers, customers, conversations, tasks, campaigns and reports belong to exactly one account, and nothing crosses between accounts.
The operation in figures — per country, per operator, per channel.
Every figure is calculated from the platform's own record of what happened, and every metric carries a plain-language definition beside it, so two people reading the same number mean the same thing. Reports are filtered by date preset or range, by number group and by operator.
Wallboard: a large-format view for the contact-centre floor. The organisation chooses which figures appear. Illustrative values.
| Group of metrics | Content |
|---|---|
| Headline | Contacts in the period, share handled by AI, average talk time, share of callers who chose an operator, satisfaction score. |
| Service & demand | Offered, answered, abandoned and missed; answer rate; average speed of answer; service level against the chosen threshold; average and maximum wait and call duration; current queue depth; calls involving hold; collaborative calls; ownership changes. |
| Patterns over time | Calls by day, by hour and by day of the week, answered against abandoned; operators online against volume for the same period. |
| By number group | Offered, answered and abandoned with a high-abandonment flag, wait, service level, talk time and the operators in the group — per group or combined. |
| By operator | Answered, missed, speed of answer, talk time, share arriving from AI intake, share transferred, full call list — showing where coaching helps rather than ranking people on speed. |
| Channels, AI, outbound | WhatsApp and email volumes with answer rate, speed and handling time computed over threads; AI share and escalation rate; outbound volumes and campaign outcomes alongside inbound. |
Campaigns, branches and people with a phone in hand.
Operator calls and AI-led campaigns are one solution from the organisation's point of view: the same numbers, recordings, customer records and reports. And staff at branch counters, rental points and in the field are brought into the same picture without a desk or a headset.
AI-led outbound campaigns
| Element | Behaviour |
|---|---|
| Contact lists & batches | Uploaded with the data each call needs — name, reference, asset — so every conversation is specific. Work is grouped into batches that are started, paused and stopped; calling windows decide when calls may be placed. |
| Do-not-call & halt | An absolute list honoured without exception, alongside the marketing-consent flag. Calling stops at once, for one batch or for everything. |
| Outcomes | Each call is classified against outcomes the organisation defines — agreed, refused, no answer, call back later, wrong contact — so a campaign result is a figure. A person who asks to be called later becomes a scheduled task automatically. |
Three ways to bring distributed staff in
The workspace on a phone
A branch employee opens the web application, signs in and works as a desk operator does — takes calls, sees the customer, creates tasks. Nothing installed.
Anyone reachable by number
The operator presses a button, the platform calls the mobile and joins the person in. Recording, transcript, timing and statistics continue — for people outside the organisation too.
Presence and notifications
A small, specific purpose: push notifications when a call or task arrives, and an available / unavailable toggle telling the platform whether to route work to that person.
Isolation, consent, retention — and a rollout that starts where the value is clearest.
Humans stay in control, every action is on record, and adoption goes capability by capability rather than in one high-risk switch-over. Runs in the EU — hosted by Proof IT or in the organisation's own infrastructure — by an ISO 27001 and 9001 certified organisation; provider events are authenticated, duplicates collapsed, credentials never reach the browser.
Follows the corporate directory
Joining and leaving the organisation is the only place access is granted and removed. Group visibility is enforced on every request — including recordings, audio downloads and transcripts.
Honoured when the platform acts
A customer who has refused AI handling reaches a person; one who has refused marketing contact is left out of a campaign. Callback consent captured by the assistant travels with the task.
Audio expires, the record outlives it
Recordings and transcripts are removed automatically after the period the organisation sets; the summary is kept for a longer period of its choosing.
Enterprise terms, no training on your data
AI capabilities run under terms in which conversation content is not used to train models. Which capabilities are on is decided per group.
Delivery approach
Groups, numbers, people
The organisation as number groups; which lines, which countries, which operators.
Numbers and channels
Forwarding onto carrier numbers, caller-ID verification, WhatsApp and mailboxes attached.
Procedures and material
The call types that matter, the checklist behind each, what the assistant may say.
Read in, write back
Field list and direction agreed; summaries and contacts flowing to the systems of record.
Thresholds, then more
AI share raised as confidence grows; campaigns, branches and new countries added as groups.
The conversation, the customer, the follow-up and the numbers — in one place.
The Proof IT Contact Centre Platform suits organisations that take customer contact by telephone and message, in one country or several, and need every conversation on record, every promise followed up, and every figure available — with AI absorbing the routine load.