# How It Works > Connect, understand, predict, decide and act — the pipeline that runs behind every Veraius answer. Source: https://veraius.com/en/how-it-works-en/ --- How It Works # From Data **to Action.** Five steps run behind every answer Veraius gives: connect the systems, understand what the data means, predict what happens next, decide what to do, and execute it. This page walks through each one. [Request a Demo](https://veraius.com/demo/) [Explore the Platform](https://veraius.com/platform/) Automation ## How a problem is seen **before anyone reports it.** Most problems are visible in the data days before they reach a meeting. Veraius keeps reading, checking and comparing on its own, so the problem arrives with its evidence while there is still time to act. 01 ### Read on a schedule Connectors sync your ERP, CRM, MES and external data sources on a timetable you set, so the latest orders, stock, prices and records are already in place before anyone asks. 02 ### Turn data into signals Signal rules and agent skills turn raw figures into signals: cheap energy hours from day-ahead prices, currency and inflation pressure, competitor price pressure, market growth, a stock or cash threshold crossed. Each signal carries its confidence and is not raised when the data is too thin. 03 ### A signal starts the workflow You decide which signal starts which workflow. When it fires, the workflow starts on its own: the right agent gathers the evidence, rules and models evaluate the options, and the case is prepared without anyone opening a report. 04 ### Check without being asked Scheduled workflows run on each organisation's own calendar. Open decisions are re-evaluated every night and operations are reviewed every hour, so a decision that has gone stale is caught before it causes damage. 05 ### Spot what is out of line Agent runs include anomaly checks on the data they read, and a run that misses its service level or fails is recorded as a finding for review. 06 ### Hand the work on When one workflow finishes, automation rules start the next agent in the chain without a click, and cross-department steps continue as one process. 07 ### Notify the right person, approve in Veraius The finding reaches the right person with a link to the case. The decision itself is reviewed and approved in the Veraius app, and anything written back to an external system is written exactly once. The Five Steps ## Connect. Understand. Predict. **Decide. Act.** Most analytics programmes stop after the second step. The value is in finishing the chain. 01 Connect ERP, POS, MES, WMS, CRM and IoT — into one place. 02 Understand Operational data related and put in business context. 03 Predict Demand, stock, production and performance, ahead of time. 04 Decide Decision models and AI weigh the options with you. 05 Act Decisions executed through workflow and automation. Behind the scenes ## One question, **a full decision chain.** This is not a conceptual diagram. It is the pipeline that runs behind every answer: planning, parallel specialists, review, grounding in your own data, and an honesty check before anything reaches you. User**Question Asks when unclear ↻ Intent & Router ### Planner Splits the goal into sub-tasks ### Parallel Specialists Domain agents run concurrently Review ### Grounding Enterprise data, sourced answers ### Synthesis One answer, streamed live Final check ### Decision Reasoned, sourced, actionable User Question** A question in plain language. **Asks when unclear** Ambiguous? It asks back before guessing. **Intent & Router** Routes the question to the right domains. **Planner** Splits the goal into sub-tasks. **Parallel Specialists** Domain agents run concurrently. **Review** Checks the specialists’ output. **Grounding** Every claim tied back to your own data. **Synthesis** One answer, streamed live. **Final check** Says “I don’t know” when the data does not support an answer. **Decision** Reasoned, sourced, actionable. **Input **Planning **Reasoning **Knowledge **Output Step 01 ## Connect. **Everything in one place.** ERP, POS, MES, WMS, CRM and IoT data is ingested from the systems you already run. Veraius reads from them; it does not ask you to replace them. ### Operational Systems ERP, MES, WMS and POS records read at the granularity decisions actually need. ### Sensor & Telemetry IoT streams from machines, vehicles and facilities alongside transactional data. ### APIs & Files Any system with an interface, plus the spreadsheets that hold the rest of the truth. Step 02 ## Understand. **Data in business context.** Data is related and put in business context. A production order, a customer order, a stock movement and a sensor reading stop being four separate records and become one situation. ### Relationships Records from different systems linked by product, site, customer, order and period. ### Business Meaning Metrics defined once, so revenue and margin mean the same thing in every conversation. ### Grounding Every later answer traces back to a source record, not to a model’s recollection. Step 03 ## Predict. **Forward signals, ahead of time.** Machine learning and predictive analytics produce forward signals — demand, stock risk, capacity pressure and performance drift — with confidence attached, while there is still time to act. ### Forecasts Demand, sales, inventory, production and capacity projected forward by period and location. ### Risk Scores Stock-out, supplier, margin and delivery risks expressed as probabilities, not opinions. ### Anomalies Deviations detected as they emerge, including the ones nobody thought to set an alert on. Step 04 ## Decide. **Options weighed, with reasoning.** Decision models and AI recommendations evaluate the possible actions against your rules, constraints and objectives — and present the trade-offs rather than hiding them. ### Decision Models Candidate actions scored against business rules, constraints and objectives. ### AI Recommendations A ranked next action, with expected impact and the assumptions it depends on. ### Explainability The reasoning and the sources travel with the recommendation, so it can be challenged. Step 05 ## Act. **The decision actually happens.** Decisions are executed via workflow, integration and automation — back into the systems that own them — and the outcome is measured so the next recommendation is better informed. ### Workflow Approvals, requests and cases opened and tracked to completion, with an audit trail. ### Integration The action written back into ERP, MES, WMS or POS rather than emailed to someone. ### Measured Outcome What actually happened is compared to what was expected, and fed back into the models. Veraius agents ## What your teams gain **from Veraius agents.** [ Finance ### Budget against reality Planned and actual spend are compared by cost centre, and variances above your threshold are escalated with their cause. See the Finance agents → ](https://veraius.com/en/ai-agents/finance-agents/) [ Manufacturing ### Capacity against orders Open orders are compared with available capacity so delivery risk shows before a due date is missed. See the Manufacturing agents → ](https://veraius.com/en/ai-agents/production-management-agents/) [ IT Operations ### Changes approved by risk Low-risk changes move fast; high-risk ones go to the change board with the evidence already gathered. See the IT Operations agents → ](https://veraius.com/en/ai-agents/information-technology-agents/) ### In every department - Agents prepare the work on your own data; your people approve what happens next - Plain-language questions answered from live ERP data, with tables and interactive charts - Grounding check: numbers that cannot be traced to a source are not shown - Explainability, confidence and risk scores with decision rationale - Slack, Teams, WhatsApp and Signal channels with an approval inbox - Department-based access policy with personal data masking [All agent departments →](https://veraius.com/en/ai-agents/) Questions ## Frequently asked questions Straight answers about integrations, deployment, security and cost — including where the honest answer is "it depends". [Ask your own question →](https://veraius.com/en/company/contact/) What do we need in place for this to work? The data has to sit somewhere readable — an ERP with three years of order history is enough to start. Clean data is not a precondition; the pipeline is built to work with what the operation actually looks like. A missing field is not quietly counted as zero, it narrows the decision's confidence interval. How long before it makes real decisions? Weeks to the first recommendation, longer to running unattended. The interval between the two is where trust is built, and shortening it artificially is how these projects fail. A decision intelligence platform can take over only as far as its decisions can be audited. What are the steps from data to decision? Connect, understand, predict, decide, act. Connect reads the source systems; understand resolves what the fields mean in your operation; predict produces the forecast the decision needs; decide applies constraints and objectives to choose; act writes the outcome back or hands it to a person. The record is written at the decide step, in the same transaction. How long does each step take in practice? Connecting is days. Understanding is where the calendar actually goes — agreeing which of two date columns is the promised delivery, whether a cancelled line stays in history. That work belongs to you because only you know, and it is unavoidable rather than a sign something is wrong. Where does a person stay in the loop? Wherever the action is hard to reverse. Approval thresholds are set per decision type and per magnitude, so a stock transfer runs unattended while a price change above a stated percentage waits for someone. The boundary moves outward as the override record supports it, not on a schedule. What does the system learn when a planner rejects a recommendation? It is captured with its reason, which makes it training data of a kind history cannot provide: it encodes why an expert disagreed, and that usually rests on a fact the system does not hold. A platform that discards rejections is as good in year three as in year one.