AI operations layer for hospitality
Are you managing — or just watching?
Your dashboard says conversion dropped. Pragma says why — and what to do about it now. An AI operations layer for hotels, resorts and restaurant groups: it turns events into signals, signals into human-confirmed actions, and actions into a result measured in money.
E · S · A · I — Events → Signals → Actions → Intelligence: from raw events to governed action and learning.
Does not replace your PMS, POS, CRM, or ERP — sits above them as an AI action-control layer. Already running in production on a working resort's live data.
We accept only 3 properties for pilot through end of 2026. Pilot is 30–45 days, partially free.
What Pragma shows the manager every morning
One screen instead of scattered reports from PMS, POS, and messengers. Concrete signals, money at stake, and action buttons — not five tabs and a fifth cup of coffee.
- ● Weekend booking: guest left without an alternative dateupside 42,000 ₽
- ● Housekeeping on room 304 overdue by 18 mincheck-in at 14:00
- ● Breakfast complaint from the same guest2nd repeat
Illustrative signals from the first pilot. Real figures are published only with client consent.
E / S / A / I four pillars of the loop
The letters are not decoration — they describe what enters the loop, what matters, what happens, and what we learn.
Events from PMS, POS, CRM, ERP, tasks, messengers, reviews, guest calls and dialogues, appeals, and operational processes.
Variance, risk, overdue work, complaints, revenue dips, blockers.
Tasks, notifications, escalations, recommendations, promos, checks.
Outcome measurement, learning loop, better decisions next time.
Watching is numbers without an answer. Managing is an action with a proven effect. Pragma turns an event into a signal, a signal into a decision, a decision into a human-confirmed action, and an action into a measurable outcome.
How Pragma runs operational reality
The chain we anchor on a pilot: Event → Signal → Decision → Action → Outcome → Learning.
A guest checks in, a task slips, revenue dips, a complaint lands, a policy step is missed.
The system decides the event matters — risk, loss, blocker, or growth opportunity.
Decisioning and agents pick a safe action using rules, context, and past outcomes.
Pragma creates tasks, sends notifications, escalates, proposes upsell, or prepares a management decision.
We measure what changed: SLA, revenue, complaints, occupancy, discipline, repeat issues.
Strong plays are reinforced; weak playbooks are revised. The loop gets sharper.
Reservations call
A guest called asking for a room on the weekend.
AI detected booking intent, but no alternative dates were offered.
Pragma scored lost-booking risk and potential revenue.
The system created a manager task and drafted follow-up.
The manager re-contacted the guest; the outcome was recorded.
Pragma updated refusal reasons and recommendations for reservations.
Decision Card how AI reaches the manager's screen
When a signal calls for a decision on guests, money, or staff, AI does nothing on its own. It surfaces a card with source, loss estimate, recommendation, and a clear set of buttons. One interface — for calls, complaints, SLA breaches, penalties, HR signals.
The same card type — for a call, a complaint, a penalty, a task escalation, an HR signal. Managers learn one interface, not five.
Pragma brings decisions to you. And you can ask it directly.
The morning screen and the Decision Card work one way: the system spots a risk and brings it to your screen. But a manager also needs the reverse — to ask. So there is a second mode: a line where you ask a question in plain words across all connected data at once — PMS, POS, CRM, Yandex.Direct, finance, tasks, calls.
The answer arrives not as a chat message, but as a decision card: a number or chart, the data sources, a confidence score, and an action button — the same grade as the signals Pragma surfaces on its own.
Decisions, not a conversation
An answer is a stream of decisions with sources and a confidence score, not a conversation. If the data isn't there, Pragma says “no data” honestly rather than inventing a figure.
Across every integration at once
One question runs across PMS, POS, CRM, Yandex.Direct, finance, tasks, and calls — where before you had to open five systems and reconcile them by hand.
AI proposes — humans decide
Any action from an answer — a task, a follow-up, a change — goes through your confirmation and gets an Impact ID. The same safeguards as when the system surfaces a signal itself.
Data in your perimeter
Questions run over your data inside your own infrastructure, with guest personal data protected. Nothing leaves for third parties.
Free-form questions across all your data are a capability of the Pragma operations layer that we are rolling out on the assistant now. On the pilot we show it on your data and an agreed set of sources.
Operational signal sources
Pragma does not require one system of record. Every channel feeds events that become signals and actions.
Guest calls become another signal source: Pragma shows where a booking was lost, why a guest declined, and what action to take next.
Kitchen, storage and sanitary standards become another source of operational signals: Pragma detects temperature deviations, overdue checklists, write-off risks, missing evidence and recurring violations — and turns them into tasks, escalations and Impact IDs.
Conversation Intelligence: the cause is where it happens — inside the guest conversation
Pragma analyses guest calls and dialogues to surface lost bookings, refusal reasons, unhandled complaints, and cases that need follow-up. AI does not replace the manager: it shows the signal, proposes an action, and routes it for human confirmation.
We reviewed 627 recorded reservation calls. Here is what the analysis found — and what becomes your baseline metric on the pilot:
of callers never got through — and no one reached back within 24 hours: 1,057 enquiries lost before a conversation even started
This is one property's baseline, not an industry benchmark. Your numbers will be your own — Pragma measures them on your recordings, and the “after” becomes a pilot metric.
Lost bookings
AI flags calls where the guest was ready to book but left without a next step.
Refusal reasons
Price, no dates, no alternative offered, weak terms explanation, no follow-up.
Follow-up tasks
Pragma creates a manager task and suggests message copy or a re-contact playbook.
Missed revenue
The system estimates potential loss and highlights guests worth re-engaging.
From a call debrief — to a prompt in the moment
The conversation is where the booking is won or lost. Pragma works right inside it: today through a post-call debrief, next through a live prompt to the operator during the call. We don't promise what isn't there — we've already validated the path to a live prompter technically.
A line-by-line prompter debrief: which move worked, where the booking slipped, and which phrase to use next time instead of the one that didn't land. Not “conversion dropped,” but a specific move — AI proposes, the manager confirms.
A live in-call prompt: the guest card, open dates and rates on the operator's screen in the moment. Feasibility is already validated technically; publicly we promise only what works in the pilot.
Roadmap The same speech analytics and prompter — not only for the sales desk but for internal communications too (employee ↔ support): dialogue debriefs, procedure prompts, and escalation of anything left unanswered. A direction of development — the internal-comms phase; publicly we promise only what already works in the pilot.
Call recording and analysis happen on your telephony stack, on the basis of guest consent and your data-processing policy. Pragma does not record calls on its own and does not store audio outside your property's infrastructure.
Get the same breakdown of your reservation desk
Send a period's call recordings and we'll return a breakdown of lost bookings, “let me think” reasons, and missed revenue in money. No rollout, no new system for the team.
The analysis runs on your recordings, under your consent and data-processing policy. Pragma shows the signal and proposes a follow-up — your manager confirms.
Reputation Ops: the negative review that never gets written — and never comes back
Pragma already collects reviews from the map and review platforms your property lives on into a single event feed. The next step is a closed loop: unhappy guests are noticed while still in-house, public negatives are answered on time, and a complaint can only be closed by fixing its root cause — with evidence. AI prepares the analysis and a draft reply; your manager decides and publishes.
We audited 2,166 public reviews of an operating resort hotel and checked every negative one for a reply. The team works fast — and still:
of negative reviews had a public reply, median response time — 1 day. Solid manual work — and also its ceiling
negative reviews were left unanswered — including two fresh public refund complaints that every prospective guest reads in high season
noise complaints within a single month went unaddressed: individual replies exist, but no one saw the systemic pattern
Most telling: not one of the 325 complaints can be shown to have its root cause fixed and not recurring — the industry simply has no such mechanism. That is exactly what Pragma adds. We can measure your property the same way — from public data, before any integration.
Interception before publication
A short survey on day two of the stay and on departure day. An unhappy guest gets the problem solved before check-out — instead of posting a 1-star review when it is too late.
No negative review gets lost
Every negative review carries a response deadline and an owner. Overdue items escalate to the GM and the owner automatically. A complaint cannot be closed for show: only a fix with photo evidence, or an explicit reasoned decision.
A reply with facts, not boilerplate
AI drafts a reply in your brand voice and inserts the fact from the closed task: “resolved on 12 July, work accepted by the GM”. A human always publishes — auto-publishing does not exist.
Verified resolution
After 4–8 weeks the system re-measures whether the same complaint returned and issues an honest verdict: resolved / recurred / insufficient data. This metric cannot be gamed by deleting reviews.
Review collection from the major map and review platforms into a single feed is already live. Plus a reputation audit snapshot from public data: reply coverage, response speed, systemic gaps — like the example above.
Pre-publication interception, the SLA and escalation loop, quality-guarded draft replies and resolution verdicts are rolled out on a pilot property. We only promise publicly what already works on a pilot.
Pragma does not inflate ratings, does not remove or commission reviews, and never auto-publishes replies. We do not promise rating growth: pilot metrics are response speed and coverage, the share of negativity intercepted before publication, and verified root-cause fixes. Brand replies are published only by your staff.
Get a reputation audit snapshot of your property
Send a link to your property's public listing — we will return an audit built from public data: how many negatives are unanswered, how fast your team responds, and which complaints recur systemically. No integration and no access to your systems required.
The audit uses only your property's public reviews. Pragma shows the gaps and suggests an action plan — the decisions remain yours.
Pragma Hygiene Intelligence: kitchen and storage as a source of operational risks
Not a standalone HACCP log — a sanitary risk layer inside Pragma Command Center.
Pragma Hygiene Intelligence connects POS, 1C/storage, checklists, photo evidence and IoT sensors into a unified kitchen and storage control loop. It doesn't just record temperatures or cleaning. It identifies the risk, suggests a corrective action, assigns responsibility, requires proof of completion and saves the result to Impact ID.
Cold chain
Temperature deviations, prolonged door openings, manual probe verification, write-off risk and a task for the responsible person.
Sanitary checklists
Shift opening and closing, zone cleaning, disinfection, labelling, product placement, photo confirmations.
Shelf life & storage
Batches, stock levels, expiry dates, write-off risk, link to sales and procurement.
Audit Pack
Temperature logs, checklists, incidents, corrective actions and evidence for internal or external audits.
Roadmap · pilot loop for restaurants, kitchens, sanatoria and SPA hotels.
Pragma Risk Intelligence: where the business loses money — before the report shows it
Not a security add-on — a money-control layer inside Pragma Command Center.
Pragma Risk Intelligence looks for financial, operational and managerial anomalies across PMS, POS, CRM and 1C data, turns them into signals with a rouble loss estimate, explains the cause and surfaces a Decision Card with evidence to the manager. The AI never hunts for fraud on its own and never acts by itself: it shows the anomaly, gathers the evidence and hands the decision to a human.
Plan-vs-actual anomalies
Sharp revenue deviations, rising food and labour cost, falling RevPAR, occupancy that doesn't match revenue — on your existing plan-vs-actual thresholds.
Suspicious procurement
Prices above the median, payments split under the approval limit, repeating amounts, procurement growing without revenue, purchases bypassing policy.
Counterparties & 1C
Supplier risk dossier, repeating tax IDs and bank details, hidden links, suspicious detail changes, new counterparties with fast-growing turnover.
Operational abuse
Unjustified discounts, refunds and order cancellations, manual cheque adjustments, write-offs, night-shift operations and out-of-norm actions.
AI document auditor
OCR of contracts, invoices and acts, detail extraction, contract-invoice-act reconciliation, mismatch detection and risk explained with quotes from the documents themselves.
Every risk signal passes through rules and human approval — never straight into a fine. The loop expands as your sources connect: 1C, POS, documents.
Pragma Academy: standards live in shift clearance, not in a binder
Not a standalone LMS — a training and clearance loop built into your live policies and tasks.
Academy extends the live policy-and-testing loop into a full cycle: a new hire gets a role-based onboarding plan, takes courses assembled from your policies, proves skills in practice and earns clearance to work shifts. Managers see each person's readiness — instead of discovering it at peak hour.
Role-based onboarding
A day-one / week-one / month-one adaptation plan starts automatically on hire. A mentor signs off on practice; HR tracks every newcomer's progress.
Courses built from policies
A course is assembled from live policies: update the document and the training follows. Reading and test results are recorded.
Exams that test doing
Beyond recall: situational cases from real hotel practice and tasks with in-app photo evidence. AI drafts questions from the policy — the author approves every one.
Clearance and knowledge checks
Periodic micro-checks against knowledge decay. Shift-clearance status is visible before the shift; certificates verify by QR inside and outside the company.
In development · the core loop — policies with mandatory knowledge testing — is already live in the product.
Your property is already digital. Management is still manual.
- 🏨Reservations — in the PMS
- 💳Revenue — in the POS
- 📋Leads — in the CRM
- 📊Finances — in the ERP
- 💬Tasks and assignments — in messengers
- ⚠️Guest complaints — in conversations
- 🧠Control — in the manager's head
- 🌡️Sanitary logs, kitchen and storage — in paper, Excel or messengers
The data exists. But a dashboard shows the temperature, not the diagnosis: “conversion is down” is a result, while the cause lives inside a specific conversation, shift, or process. “Let’s tighten up” is not an action. What’s missing is the layer that turns the number into a confirmed action and a measurable result. Managers learn about a problem no earlier than the guest does.
When temperatures, cleaning, labelling, expiry dates and write-offs exist separately from POS, 1C and tasks, managers see violations too late. Pragma turns these events into signals and actions within the unified operational loop.
Pragma does not replace your stack. It makes it behave as one loop.
Shelter, R-Keeper, amoCRM, 1C, and other tools hold data in their lanes. Pragma ties them to tasks, regulations, people, and management actions — so owners see signals and actions, not only disconnected reports. We reviewed some forty solutions built for this industry: AI appears in each of them in one spot — pricing here, calls there, cameras elsewhere. Pragma assembles those signals into a single management loop.
| Capability | ERP / accounting | BI | CRM | Pragma |
|---|---|---|---|---|
| Cross-system operational loop | — | — | — | ✓ |
| Tasks and escalations from signals | — | — | Partial | ✓ |
| Accountability trace | Partial | — | — | ✓ |
| Works between systems | — | — | — | ✓ |
| No PMS/POS/CRM replacement | — | ✓ | ✓ | ✓ |
| Agents and action playbooks | — | — | Partial | ✓ |
| Audit and approvals | Partial | — | — | ✓ |
What Pragma is not
Pragma is not your PMS, POS, CRM, ERP, telephony stack, or contact centre. It sits above them and turns their events into management actions.
Conversation Intelligence does not replace telephony and is not a call centre product. It is a Pragma module that treats calls and dialogues as operational signals: lost bookings, refusal reasons, complaints, and follow-up tasks.
The reputation module does not “manage ratings” and is not an SERM agency: it does not inflate scores, remove reviews, or publish anything on the hotel's behalf without a human. It is the Pragma loop that turns guest reviews and surveys into signals, tasks, and verified root-cause fixes.
Where implementation starts
Operations Control
Tasks, blockers, delays, escalations, and execution discipline.
Guest Experience
Complaints, guest requests, response speed, service paths, repeats.
Revenue & Occupancy
Occupancy, upsell, lost leads, marketing signals, revenue per guest.
HRM & Regulations
Shifts, workload, policies, knowledge checks, execution compliance.
Food Safety & Loss Prevention
Temperature deviations, sanitary checklists, expiry dates, write-off risk, photo evidence, Audit Pack.
In our ad-audit work, a sizeable share of paid budget — often a third or more — leaks into irrelevant traffic and competitor-brand poaching with near-zero return. The Revenue & Occupancy loop surfaces it as a signal — instead of you finding it at quarter's end.
Five examples: event to action.
The same loop in F&B, rooms, HR, and reservations — without turning the page into a technical manual.
Three focus zones for owners and directors
Pragma agent layer First wave
The product core already holds tasks, signals, and execution control. Below, honestly: which of the specialised loops run in the first release — and which turn on during the pilot as integrations land.
Pragma must prove impact — not only run
We baseline sites before go-live, then measure change at 30, 60, and 90 days. No promised “auto-growth”; a transparent method instead.
Each action gets an Impact ID: recommendation → approval → execution → outcome.
Concrete deltas come from your pilot data; we do not publish borrowed case figures or guaranteed percentages.
How the loop is built in brief
Impact sizing on the pilot
The dashes in the table are deliberate: we take potential figures from your data, we don't invent them. Here is what it looks like in practice — the first reservation-desk pilot already showed:
This is one property's baseline, not an industry benchmark. Your numbers will be your own — Pragma measures them on your data, and the “after” becomes a pilot metric.
| Illustrative upside | Payback horizon | |
|---|---|---|
| Small footprint | Based on your baseline | Case by case |
| Mid-size group | Based on your baseline | Case by case |
| Holding | Based on your baseline | Case by case |
We do not promise guaranteed revenue uplift — we agree measurable metrics up front and compare honestly afterwards.
AI proposes. Humans confirm. Pragma records the outcome.
On the pilot, Pragma does not run critical actions autonomously. AI forms the signal and recommendation; the manager confirms, edits, or rejects it; the system keeps the decision history and result.
Human-in-the-loop
Decisions on guests, money, and people require human approval.
Audit trail
Every action is logged: who saw it, who approved it, who executed it.
Privacy-by-design
Recordings and transcripts follow your access, retention, and consent rules.
Pragma does not act without a human
At pilot start, Pragma operates in observation mode only. Active actions only after your manager confirms.
- ✓Pragma does not change prices or rates without manager approval
- ✓Pragma does not charge money or apply penalties automatically
- ✓Pragma does not make HR decisions without a human
- ✓Every recommendation appears on the manager's screen with an explanation
- ✓Any type of recommendation can be disabled at any time
- ✓Your data stays in your infrastructure (self-hosted)
See where revenue leaks — and why the number is trustworthy
Not another dashboard your staff will abandon. A live scenario: a signal from the floor → your decision → a documented result. Step through it yourself.
The screen a manager opens first: not a list of all the data, but the one thing that is costing money right now and waiting for a single decision.
Step 1 · A signal arrives. Pragma captured the complaint with its source, estimated the loss, and proposed what to do.
- Over one shift this catches losses that would otherwise slip away unnoticed
- A decision in 40 seconds — no meetings, no message threads
- Every action stays in history — who decided, and why
The on-shift staff screen: exactly what to do now, and a button to confirm with a photo — in two taps.
Step 1 · A task for the shift. No five different apps — just what to do now and by when.
- Proof is a fresh in-app photo, not an old one from the gallery
- Two taps for staff — no training on a new system
- The manager sees confirmed work, not a vague "think it's done"
Your screen: where revenue leaks across every property — and why the number is trustworthy. The data underneath is logged by the shift, not by you by hand.
Step 1 · Where revenue leaks. The number is trustworthy: the shift logged it, the manager confirmed it — Pragma only compiled it. Next, one action.
- Every property on one screen — no manual report-gathering
- An anomaly is visible right away, not at month-end
- The task goes out with a deadline and stays in the history — you see whether it was done
These are three screens from the product. The rest is easier to show on your data than to describe.
Personal demo of the real productPragma doesn't just recommend. It proves impact.
- 13:42 · 📞 Guest call (PMS event 9821)
- 13:43 · ⚠️ Signal “lost booking intent”, confidence 0.87
- 13:44 · 🤖 Recommendation: follow-up + 3 alternative dates
- 13:51 · 👤 Alexey M. approved, chose “follow-up”
- 14:30 · ✅ Guest contacted, booked 2 nights
- 15:00 · 💰 ΔRevenue +21,000 ₽ credited to the property
Every Pragma action gets a unique identifier. You always see: what the system proposed, who confirmed it, what was done, and what the measurable outcome is.
AI proposes. Humans decide. Results — in money, time, and service quality.
How Pragma proves impact
Pragma is not a pitch deck. It is a four-step framework that ends not with an interface, but with a report — a measurable delta on the metrics you agreed up front.
We agree the metric set and capture current values for your property: tasks-in-SLA, time-to-first-action, lost follow-ups, policy execution, RevPAR, etc. Without this snapshot, every later step is meaningless.
Pragma watches and emits signals — but takes no actions. We compare what AI saw against what staff saw. This is when we calibrate confidence and cut out noise.
AI proposes actions. The manager confirms, edits, or rejects each one. Every decision gets an Impact ID. Override rate becomes a metric of its own: how often a human disagrees with AI.
Roll-up of baseline metric changes. Not opinion, but a quantitative gap, broken down by signal, with the chain recommendation → approval → action → outcome.
| Metric | Baseline | Day 90 | Δ |
|---|---|---|---|
| Time-to-first-action on guest complaints | X min | Y min | −Z% |
| Share of tasks closed within SLA | X% | Y% | +Z pp |
| Lost booking follow-ups | X/wk | Y/wk | −Z% |
| Repeat complaints (by topic) | X/mo | Y/mo | −Z% |
| Staff who passed policy tests | X% | Y% | +Z pp |
| Override rate (AI ↔ manager) | — | X% | target corridor |
We do not promise guaranteed growth. We promise a measurable before/after on the metrics you agreed — and we commit to showing it within 90 days.
A spa hotel in Anapa — Pragma's first reference
Active pilot, 2026. We will name the client only after they agree — for now we share the format and context anonymously. Measurements are published after the pilot, and only what the client approves. We do not promise guaranteed percentages — every property is different.
- nowBaseline — we capture the starting point on agreed metrics
- nowShadow Mode — Pragma observes and collects data, no active actions
- Recommendation Mode — AI proposes, a human confirms
- Impact Report — a measurable delta vs baseline at day 30 / 60 / 90
The rollout has only just begun — data collection and observation are underway. That is why there are no “growth percentages” here: the first measurable cuts will come after Shadow and Recommendation Mode. We will show only what is confirmed on the property’s data and approved by the client.
Full client name, logo, and specific numbers — only with written consent. This page shows format and context without revealing sensitive data.
Pragma documents
A short overview for a first look, plus the full product presentation. Grab them before we talk — the conversation gets sharper.
Stop watching.
Start managing.
Audit 50 calls for lost bookings
Tell us about your property and your goal — we reply within one business day. No sales scripts.
Before the pilot
Missing a question? Email info@anapaerp.ru