Operations & quality

Operations: Inbox, Analytics & Data Protection

This section describes the day-to-day operation of your Prezio account: you read conversations in the inbox, keep context on every customer in the CRM area, track your business figures in the analytics dashboard, set goals together with a weekly digest, manage appointments and orders – and you learn how deleting customer data (GDPR) works. This is where you come once the agent is already set up and you want to operate it in everyday use.

20.1 Inbox ("Conversations") – reading conversations

The inbox (the left-hand menu item "Conversations") shows every conversation your agent has had across all channels (web chat, SMS, WhatsApp, email, phone). The view is laid out like an email program: the list on the left, the reading pane on the right.

Step 1: Select an agent. - At the top left you select the agent via the dropdown. By default the first visible agent is preselected. Deleted agents do not appear.

Step 2: Search the list. - The search field ("search") filters the list as you type (with a short delay). It searches the conversations of the selected agent. - Each list row shows: name or contact details, time of the last message, number of messages, channel (e.g. "Web", "WhatsApp"), the status and – if present – a lead badge with status and score (e.g. "qualified · 72"). The coloured initials circle is tinted by lead score (green ≥ 70, yellow ≥ 40). - If a row shows the yellow note "Missing", the AI has not yet captured all required lead fields – a click reveals which ones in the tooltip.

Step 3: Open a conversation. - On desktop a normal click opens the conversation in the right-hand reading pane; the first conversation is opened there automatically, so the pane is never empty. On mobile the list fills the screen, and a tap opens the detail view (with a "Back" arrow). - With Ctrl-/Cmd-click or the middle mouse button you open the conversation in a new tab as usual; the address is shareable and bookmarkable. - 20 entries are shown per page; if there are more, you page through them at the bottom with the arrow buttons.

Step 4: Read the conversation history. - In the reading pane you see all messages as speech bubbles (customer on the right, AI/agent on the left with a bot icon). Phone and voice calls are stored as a single transcript and automatically split into bubbles by speaker. - Beneath each AI reply you find rating and training buttons (thumbs up/down as well as adopting it as a routing/few-shot/eval example). These belong to the AI-debug area (see the AI-debug chapter) and have no effect on the customer.

Tips & pitfalls - The inbox is read-only plus training feedback – you do not type your own replies to the customer here. Replies are sent by the agent over the configured channels. - If the search finds nothing, first check whether the right agent is selected at the top – the search is always restricted to a single agent. - The list sorts by last activity; the most recently active conversation is at the top.

20.2 The CRM sidebar – context on every customer

On the right (on desktop) or below the history (mobile), Prezio shows automatically aggregated customer information. These cards only appear when data is available.

Contact ("Contact"): Email (clickable as mailto:), phone (clickable as tel:), company and position, where captured. At the very bottom, the time of first contact. If nothing has been captured, a corresponding note appears here.
Lead status ("Lead status"): The lead score from 0–100 in a coloured field, the status (e.g. "qualified", "nurturing") and, where applicable, the qualification date. If the "effective" value differs from the score, it is shown separately in yellow. Missing required fields appear as a yellow warning. Via the "View lead" link you jump to the full lead detail page.
BANT: Four criteria of purchase readiness – budget, decision-making authority ("authority"), need ("need") and timeframe ("timeline") – each with a green tick (met), a red X (not met) or grey (unknown), including the detail text the AI has detected.
AI insights ("Insights"): Freely summarised key points, pain points ("pain points", red), interests (blue), tags and a suggested next step. This card only appears when the AI has derived at least one of these.

Note on lead capture: In CRM mode (the "auto_create_on_contact" setting), every contact on all channels (SMS, WhatsApp, web, email) becomes a lead; phone calls always create a lead. If CRM mode is off, a lead is only created when the AI decides to (the "capture_lead" setting). This explains why some conversations carry a lead badge and others do not.

Via a lead or an appointment booking you reach the customer profile page (the "regular customer file"). It bundles name and contact channels, notes and the entire appointment history.

Step 1: Maintain marketing consent ("Marketing consent"). - The toggle records the customer's explicit consent to marketing (GDPR Art. 6(1)(a)). Default: off. - When toggled, the date is automatically stored and shown beneath the toggle ("consented on …" / "unsubscribed on …"). Recommendation: only switch it on when the customer has demonstrably agreed – the date is your evidence if a marketing email is later challenged.

Step 2: Add internal notes ("Private notes"). - This free-text field is visible only to you/your team and persists across all appointments (e.g. "always wants a tonic", "allergic to PPD"). Click the text to edit it, or click "Add note", and save. - This profile note is separate from the internal notes per appointment (see the appointment list).

Step 3: Read the appointment history. - "Upcoming" and "Past" are grouped separately. For each appointment you see the service, status, time, where applicable the resource/staff member as well as the customer's completed intake answers – handy for seeing what was requested last time.

Tips & pitfalls - Name/email/phone are not editable on this page; these fields are maintained automatically by the agent from the channel data. They can only be corrected by the customer themselves (see GDPR below) or via the Rectify function. - Private notes are personal data: in the event of a deletion request from the customer they are deleted along with everything else (see 20.7).

20.4 Managing appointments ("Appointments")

The global menu item "Appointments" shows bookings across all agents for which appointment booking is enabled. On the agent detail page there is the same list, restricted to a single agent.

Step 1: Filter. - At the top you select "All agents" or a single booking-capable agent. - Via the status dropdown you filter by confirmed ("confirmed"), pending ("pending"), cancelled ("cancelled"), completed ("completed") or no-show ("no_show"). Default: all statuses. The total count is shown on the right. Up to 100 appointments are loaded.

Step 2: Work through appointments. Each appointment card offers actions: - "Done" / "No-Show": These appear only for confirmed appointments whose time is in the past. This is how you mark retrospectively whether the customer showed up – it feeds directly into the no-show rate in the analytics dashboard. - Resend confirmation (send icon): Resends the appointment confirmation via email/SMS. The success note states which channel actually delivered it (or failed) – helpful for the classic "I never got an email". - Cancel (X icon): Cancels the appointment. - Internal note: For each appointment you can record a note visible only internally (the pencil/"📝" field). This is appointment-specific, unlike the persistent profile notes from 20.3. - If an appointment carries a source ("source", e.g. which channel/AI it originated from), a series or seat marker, these are shown as small badges. If a link with a speech-bubble icon appears next to the customer, it takes you straight into the associated conversation.

Tips & pitfalls - Mark "No-Show" and "Done" consistently – the metrics "utilisation", "no-show rate" and the forecasts in the analytics area are only as good as this upkeep. - If you see no appointments, check whether booking is even enabled for the agent (otherwise it does not appear in the selection).

20.5 Orders ("Orders") – the kitchen board

For an agent with order intake enabled, the "Orders" page shows a live, kitchen-style board. It refreshes automatically roughly every 10 seconds (pauses when the tab is in the background).

Step 1: Read the columns. Active orders move through the columns Received → Confirmed → Preparing → Ready ("received → confirmed → preparing → ready"). An additional "Out for delivery" column ("out_for_delivery") only appears when delivery orders are involved. Prepaid-pending as well as completed/cancelled orders deliberately get no column of their own.

Step 2: Process orders. - Each card shows a short reference (e.g. "BK-A1B2C3"), the payment status (paid/pending/failed/refunded), customer name, time, where applicable delivery address and notes, all line items with options as well as the total. - With the large button you push the order to the next status. Via the X you open the cancellation dialog (with an optional reason). - For a new active order a short signal tone sounds and a note appears – so a busy kitchen misses nothing. (Some browsers only allow sound after a first interaction with the page; the visible note always comes through.)

Tips & pitfalls - Via the manual refresh button ("refresh") you force an immediate reload without waiting for the 10-second interval. - Scheduled orders ("scheduled_for") carry a badge with the requested time – do not confuse them with the expected immediate preparation.

20.6 Analytics & goals – your business figures

The "Analytics" menu item shows the evaluations per agent. At the top you select the agent and the time window (7 / 30 / 90 days, default 30). The data is organised into tabs; each loads only when opened.

"Overview" tab (core metrics): - Net cash ("Net cash"), booked vs. collected (collection rate), margin (revenue minus AI/channel costs), utilisation ("Utilization") against your target, no-show rate, customer mix (return rate), optionally MRR (for subscriptions) and deliverability of reminders. - Many tiles show a coloured delta versus the previous period (▲/▼). Below that, breakdowns of the AI costs per channel and the reminders sent. All provider costs are captured and feed in here.

Further tabs (overview): - Staff: Performance per staff member/resource (bookings, no-shows, utilisation, revenue, tips). - Demand: A weekday-×-hour heatmap of demand plus the lead time of bookings (in your time zone). - Sources: Where bookings come from (channel/source) including revenue. - Retention: RFM customer segments and churn risk ("churn risk"). - ROI: Estimated hours saved, labour value, AI bookings/revenue, AI costs and the ratio ("ROI"). - Forecast: Demand forecast for the coming days, expected no-shows and a risk list of individual appointments (score 0–100). - Benchmark: Anonymous comparison with similar businesses – appears only when enough comparable businesses are available.

"Goals" tab – setting up: Here you set monthly goals; progress is always measured for the current calendar month (independent of the 7/30/90-day filter at the top).

Step 1 – Revenue target ("Monthly revenue target"): An amount per month in your currency. Empty = no target. Stored as an integer ≥ 0. Recommendation: set it realistically based on your last good month.
Step 2 – Utilisation target ("Utilization target"): As a percentage (enter e.g. "75"). Valid range 0–100 %. Suggested in the field: 75.
Step 3 – Max no-show rate ("Max no-show rate"): As a percentage (e.g. "10"). Range 0–100 %. Suggested in the field: 10.
Step 4 – Weekly digest ("Weekly digest"): A checkbox. Default: off.
Step 5 – Save. The three tiles at the top (revenue / utilisation / no-show) then show the monthly progress against your goals.

Weekly digest ("Weekly digest") in detail: When the box is ticked, Prezio sends at most once every 7 days a short summary of the past week to the stored operator email: cash collected, utilisation, no-show rate, new/returning, delivery rate (and MRR, if present). Anomalies are added as a warning – for instance if the no-show rate has risen by ≥ 5 percentage points, the delivery rate fell below 90 %, or costs exceeded revenue. Sending is "best effort": on a failure it is retried the following day, and duplicate sends are ruled out.

Tips & pitfalls - The Benchmark tab stays empty until enough comparable businesses have contributed data – this is not an error. - The month boundary for goals is calculated in UTC; right at the start/end of the month, progress can be off by a few hours. - For you to receive a weekly digest at all, the agent must be active and have a send-capable sender address. - A knowledge-base tip that becomes visible here: when the agent "misses" answers, this is often due to dirty, crawled pages (navigation/footer text) that inflate the match distance. The knowledge search works "recall-first" (threshold kb_max_distance, default 0.5): if no result clears the threshold, the best near-miss is used anyway (marked yellow as "recall-fallback" in the decision flow). The fix: re-upload clean text. Details in the AI-debug chapter.

Prezio distinguishes two levels: the data of your customers and your own account data.

A) Your customers' data (GDPR Art. 15/16/17/18/20/21). These rights are exercised by the customer themselves – not by you via the dashboard. In every appointment confirmation and on the booking-widget footer there are links ("My data", "Export data") that use the customer's personal token. With these the customer can: - View (Art. 15) and export as JSON (Art. 20) – contact details, appointments, orders, payments, leads and conversation histories. - Rectify (Art. 16) – correct name, email, phone themselves (this is also carried over to the appointments). - Request erasure (Art. 17) – the data is anonymised, not deleted entirely: name/email/phone, free-text entries, conversation content, lead identifiers, AI-debug traces and also your internal profile notes on this customer are removed; stored payment methods are detached at Stripe. Deliberately retained, for tax/commercial-law reasons (HGB §257, AO §147 – 10 years), are the transaction metadata (date, service, amount) as well as the payment records. The erasure is not reversible – "coming back" is only possible via a new booking. - Restrict (Art. 18) and object to marketing (Art. 21) – stops all marketing immediately.

What you do about this as the operator in the dashboard: you manage the marketing consent per customer on the customer profile (20.3, with an automatic date as evidence).

B) Automatic retention (storage limitation, Art. 5(1)(e)). Prezio regularly tidies up in the background (usually once a day): - Inactive contacts: After 1095 days (~3 years) without a new booking, contacts are automatically anonymised – except customers with marketing consent, with a future appointment or under processing restriction (Art. 18). - IP addresses in admin logs: removed after 90 days (the log entry itself remains). - Voice transcripts: A separate retention period that is disabled by default (0 = off); it can be enabled to match your documented purpose (e.g. 365 days). In doing so only the spoken content is deleted; the conversation shell and any lead remain. - AI-debug traces: A short, separate period of 30 days (they contain message text + knowledge snippets and are deleted entirely).

C) Your own operator account is managed under "Account": - Step 1 – Export my data ("Export my data", Art. 15/20): One click downloads a JSON file with everything stored about your account. - Step 2 – Delete my account ("Delete my account", Art. 17): Starts a 30-day grace period. During this time the page shows a red banner with the end date and a "Cancel deletion" button. After it expires, the account is permanently deleted. You can also change name/email here (an email change requires your password) and the password.

Tips & pitfalls - A customer deletion is irreversible and also removes your internal notes on this customer – copy out important treatment/allergy notes beforehand if needed for business (and in a data-protection-compliant way). - Retention periods are platform-wide defaults; they cannot be changed per customer in the dashboard. Anyone who wants to be deleted earlier uses the deletion function (point A). - The grace period when deleting an account is your safety net: as long as the banner is visible, nothing is final.