AI controlling
When controlling keeps up with the speed of business.
Plai covers the full annual controlling cycle — data collection, planning, prediction, variance analysis and conversational analytics. AI is an extension for your controller, not a dependency.
Status quo
Managers ask — controlling can't keep up.
Traditional CPM tools focus on data administration. Reports are slow, anomalies arrive late, the forecast is obsolete the moment it's approved.
- 01
Expensive time stuck in spreadsheets
Controllers spend 30–40 hours a month mechanically consolidating data instead of analysing it.
- 02
Managers wait
An ad-hoc question like "why is margin falling in SK?" takes two days. By the time the manager is back, the decision is made without data.
- 03
Anomalies surface late
A cost spike is discovered at month-end close. The reaction comes too late, losses can't be reversed.
- 04
Forecast goes stale
The yearly forecast is updated quarterly, but the market moves in weeks. The plan is obsolete the moment it's signed off.
Three AI pillars
AI as a colleague to the controller, not a black box.
Each pillar handles a different phase of the cycle — together they close the loop between data and decision.
- Pillar 01
AI Analyst
Managers ask in plain language, the AI returns a data-backed answer. No SQL, no waiting on controlling.
- Safe queries through validated AI calls (no direct SQL)
- Suggested interpretation and visualisation
- Live answers in real time
- Instant anomaly detection
- Pillar 02
Planning Co-pilot
Helps build plans — suggests edits, calculates what-if impacts, flags inconsistencies. Full control stays with the controller.
- Trend-driven adjustment suggestions
- What-if impacts in plain language
- Mismatch detection (sales vs. capacity)
- Forecast suggestions and summaries
- Pillar 03
AI Automator
Works in the background — watches data, surfaces anomalies, drafts executive summaries before you ask.
- Variance detection 15 % warning / 30 % critical
- Notifications to accountable owners
- Monthly forecast draft
- Executive summary for meetings
In practice
A BI dashboard you can ask questions in plain language.
Managers see the numbers they care about, refreshed every few minutes. For the detail they ask the AI chat in natural language. No Excel export, no waiting on controlling.
Use case
Monday leadership meeting — 9 a.m.
From two days of report preparation to a five-minute dialogue. 25× faster.
Before
Before: Friday–Sunday report preparation, controller as a data courier. Decisions rest on two-day-old data.
With Plai
With Plai: anomalies already identified, AI answers in plain language, forecast refreshes in seconds.
- 9:00
Plai opens the Executive Summary Q3 — drafted by the AI over the weekend.
- 9:02
"Why is margin falling in SK?" — AI returns an answer and a top-3 root-cause chart.
- 9:04
Anomaly −24 % at the butter supplier — suggested action: renegotiate the contract.
- 9:07
"What if we raise sandwich prices by 4 %?" — what-if impact on EBIT, live.
- 9:12
Yearly forecast recalculated. Decision made, meeting ends 20 minutes early.
Measurable impact
Payback under six months.
Reference case for a mid-sized company with three controllers: 32 hours saved per month × 12 months × CZK 800 hourly rate = CZK 921,600 worth of freed capacity per year. Without hiring an extra team member.
- time on reports
- −80 %
- saved / controller / month
- 32 hrs
- faster variance detection
- ~70 %
- payback period
- < 6 mo.
AI that even the head of IT can sign off.
Plai is designed so that data never leaves your premises and the AI does nothing it shouldn't.
- 01
Plai runs without AI too
AI is an extension, not a dependency. You can disable it at any time and the system keeps working — every AI feature has a manual alternative.
- 02
The AI doesn't write
The AI only reads and suggests. Data changes are always made by the user after confirmation. No "AI surprises", no runaway agent.
- 03
Full audit log
Every query, every answer, every call logged with a trace ID. Compliance and GDPR-ready, fully traceable after the fact.
- 04
Data stays in-house
The Ollama AI provider runs inside your perimeter. No outbound calls to the public cloud — and no compromise on capabilities.
- 05
Validated tool API
The AI has no direct database access. It works through typed, validated calls — no SQL injection, no unbounded queries.
- 06
Role-based access
The AI respects user roles — a controller sees different answers than a sales rep. Tenant isolation enforced at the API layer.
Tech stack
Cloud-native architecture with full on-prem support.
- Frontend
- SvelteKit · Node 22
- Backend
- .NET 10 · ASP.NET · SignalR for AI streaming
- Databases
- PostgreSQL 17 (metadata) · ClickHouse 24.8 (OLAP fact tables)
- AI provider
- Ollama (Llama 3 / Mistral, local) · Azure OpenAI · OpenAI · Anthropic · NullProvider
- Distribution
- Docker · Docker Compose · Kubernetes (Linux x86_64); air-gapped operation supported
Pilot
Try Plai on your own data in 2–3 weeks.
Non-disruptive pilot — Plai runs in parallel and doesn't touch your existing systems. Fixed scope, fixed price.
- Week 101
Infrastructure prep
One server / VM with Linux and Docker per the pilot brief — 2 vCPU / 8 GB RAM / 50 GB SSD, up to 20 concurrent users.
- Week 202
Installation and onboarding
Docker Compose deployment, access configuration, sample data import, onboarding from the NK document.
- Week 303
UAT and training
Live demo of every AI feature on your data, acceptance testing, user training.
Let's spend 30 minutes on your numbers.
A Plai showcase on your own data, no strings attached. We'll calculate the savings on your capacity and walk through your use case.