Most South African businesses already have a business intelligence dashboard. Power BI is open on someone's second screen, the monthly pack lands in an inbox, and the numbers are broadly correct. Yet when a plant stops, a delivery slips, or a branch underperforms, nobody can say why within the hour. That gap between reporting and acting is what operational intelligence closes.
Key takeaways
- A KPI dashboard reports what happened; operational intelligence tells you what is happening now and what to do about it.
- Dashboards stall for four reasons: data trapped in separate systems, refresh lag, no agreed definitions, and no named owner for the action.
- Operational intelligence needs a unified data layer, a semantic layer that defines each metric once, near-real-time signals, and a decision path attached to every alert.
- POPIA applies the moment operational data identifies a person — customer, driver, patient, or employee — so access control and operator agreements are part of the build, not an afterthought.
- Start with one question, one dataset, one owner, and one measurable outcome inside 90 days rather than a full data platform programme.
The short version
Business intelligence answers "what happened last month". Operational intelligence answers "what is happening right now, is it normal, and who must act". The difference is not the chart library. It is the freshness of the data, the number of systems feeding it, and whether an alert has an owner and a next step.
Traditional reporting vs operational intelligence
| Dimension | Traditional BI reporting | Operational intelligence |
|---|---|---|
| Data freshness | Daily, weekly or monthly refresh | Streaming or near-real-time |
| Sources | One or two systems, usually finance and ERP | Every operational system: ERP, CRM, service desk, IoT and sensors, telemetry, spreadsheets |
| Primary audience | Executives and the board | Supervisors, operations managers, engineers, and executives |
| Trigger for action | A person opens the report | A threshold, anomaly, or forecast raises an alert |
| Typical failure | Beautiful, ignored | Alert fatigue when thresholds are untuned |
Why dashboards stall
The data lives in five places. Finance sits in the ERP, jobs in the service desk, movement in telematics, sensor readings in a historian, and the rest in spreadsheets on someone's laptop. Any question that crosses two of those becomes a manual exercise, so it gets asked once a month instead of once an hour.
The refresh is slower than the decision. A dashboard that refreshes overnight cannot support a decision that has to be made at 10:00. Operational decisions need operational latency.
Nobody agreed what the metric means. When operations, finance, and the plant each define "uptime" differently, the dashboard becomes a debate rather than a decision. A semantic layer that defines each metric once, in one place, is what ends the argument.
No owner, no action loop. An indicator turning red is not an outcome. Every signal needs a named owner, a threshold, an expected response, and a record of what was done.
What operational intelligence actually requires
A unified data layer
A data lake, or a lakehouse pattern on Microsoft Azure or an equivalent, consolidates structured and unstructured operational data without forcing every source into one rigid schema first. The point is not storage. It is that a single query can cross ERP, sensor, and service-desk data.
A semantic layer
One definition per metric, versioned and owned. This is what lets a Power BI dashboard, an alert, and an AI assistant all return the same number for "availability" or "cost per unit".
Near-real-time signals
Not everything needs streaming. Decide per metric: safety and uptime signals typically need minutes, commercial metrics often need hours, and strategic metrics are fine daily. Paying for streaming on metrics nobody reads before Friday is waste.
A decision path
Each alert carries an owner, a threshold, a runbook, and a place to log the response. Without this, operational intelligence degrades into a noisier dashboard.
Governed access
Row-level and role-level security, audit logging, and clarity on where data is processed and stored. In South Africa this is a POPIA obligation, not a nice-to-have.
What good looks like on the ground
Plant and mining operations. Sensor and historian data joined to maintenance history turns predictive maintenance from a slide into a work order: a bearing trending outside its normal signature raises a job before the shift stops.
Fleet and logistics. Telematics joined to job and fuel data answers whether a route is late because of traffic, loading delays, or scheduling — three problems with three different owners.
Retail and branch networks. Point-of-sale, staffing, and stock data in one view shows whether a weak branch is a demand problem or an availability problem.
IT and infrastructure. Service-desk tickets joined to monitoring and asset age turn a recurring incident pattern into a replacement decision, which is exactly the loop we run inside Managed IT Services.
Where AI genuinely helps, and where it does not
AI earns its place in three specific jobs. Anomaly detection: learning what normal looks like per asset, per site, per season, instead of static thresholds someone set in 2019. Forecasting: demand, consumption, and failure probability. Natural-language questions over governed data: letting a manager ask "which sites are running above their usual energy cost this week" without waiting for an analyst.
What AI does not do is decide what a metric means, fix ownership, or compensate for data nobody trusts. Point a model at ungoverned data and you get confident, wrong answers faster. Governance first, models second.
POPIA and governance
The Protection of Personal Information Act applies as soon as operational data identifies a person — a driver, a customer, an employee, a patient. Three practical consequences. First, any partner processing that data on your behalf is an operator and needs a written operator agreement. Second, access must be scoped: an area manager sees their area, not the payroll. Third, you must be able to answer where data is stored and processed, and for how long. The Information Regulator expects the answer to be documented, not remembered. Aligning the platform to ISO/IEC 27001:2022 controls and King IV reporting duties makes that answer easier to give.
How to start in 90 days
- Pick one question that costs money. "Why do we lose production hours on line 3?" beats "we need a data strategy".
- Pick the smallest dataset that answers it. Usually two systems, not ten.
- Name one owner who is accountable for acting on the answer.
- Define the metric once and write the definition down.
- Set one measurable outcome — hours recovered, cost avoided, response time reduced — and measure it before and after.
Deliver that, then repeat. Platforms built one answered question at a time get used. Platforms built as a two-year programme get cancelled.
How BroadVision helps
BiVi is BroadVision's AI middleware and data intelligence platform. It connects disparate enterprise systems, unifies them into a single data lake, and turns operational data into contextualised, governed insight rather than another dashboard. Around it, Data Intelligence & Applied AI covers the modelling and analytics work, Strategic IT Services covers the governance, roadmap, and ownership questions that decide whether any of it gets used, and Connectivity & Infrastructure covers getting the data off the floor in the first place. If you want to test the 90-day approach on one question, contact us.
Related reading: Managed IT services in South Africa: what an MSP actually does and CIO-as-a-Service: when a business needs fractional IT leadership.
FAQ
What is operational intelligence?
Operational intelligence is the practice of analysing live operational data to support decisions as work happens, rather than reporting on it afterwards. It combines data from every operating system — ERP, CRM, service desk, sensors, telematics — into one governed layer, applies thresholds, anomaly detection and forecasting, and routes each signal to a named owner with an expected response. The distinguishing feature is latency and ownership: minutes and a person, versus a monthly pack and a discussion. BroadVision delivers this through BiVi, which unifies disparate systems into a single, contextualised data layer.
Is Power BI enough for operational intelligence?
Not on its own. Microsoft Power BI is an excellent presentation and self-service analysis layer, and most South African businesses should keep it. What it does not solve is the layer underneath: consolidating data from systems that do not talk to each other, defining each metric once so every report agrees, and refreshing fast enough for an operational decision. Power BI on top of a governed data layer is powerful; Power BI pointed at five disconnected exports is a monthly argument. BroadVision builds the layer underneath through Data Intelligence & Applied AI.
Do I need a data lake?
Only if your questions cross systems. A data lake earns its cost when you regularly need to join operational data — sensor readings, tickets, transactions, telemetry — that lives in different platforms with different structures. If every question you ask sits inside a single ERP, a well-modelled warehouse or even good reporting inside that system is cheaper and faster. Decide by listing your ten most valuable questions and counting how many cross two or more systems. BroadVision will do that assessment with you as part of Strategic IT Services.
How is operational intelligence different from business intelligence?
Business intelligence explains the past for planning; operational intelligence supports decisions in the present. BI typically refreshes daily to monthly, draws on one or two systems, and is read by executives. Operational intelligence refreshes in minutes or seconds, draws on every operating system, and is read by supervisors and engineers as well as executives. The two are complementary — the same governed data layer should feed both. BroadVision runs them off one foundation with BiVi rather than building two competing stacks.
Does POPIA apply to operational dashboards?
Yes, whenever the data identifies a person. Driver telematics, customer transactions, employee productivity metrics and patient records are all personal information under the Protection of Personal Information Act, and the Information Regulator expects you to justify the purpose, limit access, and document retention. In practice this means row-level and role-level security, audit logging, a documented data-residency position, and a signed operator agreement with any partner processing the data. BroadVision signs an operator agreement as standard and builds access control in from the start — see Managed IT Services.
How long before an operational intelligence project pays back?
A properly scoped first use case should show a measurable result inside 90 days, because it is deliberately narrow: one question, two data sources, one owner, one metric measured before and after. Multi-year platform programmes are where payback disappears — the business changes faster than the build. Ask any provider to commit to a first measurable outcome in one quarter and to state exactly which number will move. BroadVision scopes the first BiVi use case that way and reviews it against the baseline — start with contact us.