Service · Software development

Custom real-time analytics platform development.

Process streaming data from machines, web shops, IoT sensors and transaction systems as it happens, with dashboards, alerting and, where desired, automated action. Custom for situations where standard BI with scheduled refreshes falls structurally short.

Arrange an introductory call See our work

An analytics platform is more than a dashboard.

A KPI dashboard looks at metrics, often based on data from last night or an hour ago. A real-time analytics platform does much more: it continuously ingests streaming data, enriches and analyses it in motion, alerts on the outcome and, where sensible, triggers a follow-up action straight away. The dashboard is then just one of several outputs, alongside alerting via PagerDuty, Slack or Teams, an API that drives other systems, and machine learning models that detect anomalies before a person would notice them.

Standard SaaS tools such as Snowflake, Tableau Online, ThoughtSpot, Sisense, Looker, Power BI, Domo and Qlik are excellent for batch analysis and reporting at daily or hourly level. We don't replace them. We build custom when sub-second latency is required, when sector-specific calculations don't fit standard tools, when analytics needs to be embedded in your own customer-facing product, when the platform has to run multi-tenant or white-label for your customers, or when the scale and cost of SaaS pricing become unviable for the intended volume.

We have experience with streaming pipelines across a range of sectors: from OEE tracking and andon systems in manufacturing and real-time production data monitoring, to conversion and cart monitoring for e-commerce, fleet tracking for logistics, network monitoring for telecoms and transaction monitoring for financial services. The common thread: data arrives in milliseconds, needs enriching with context from other systems, and must lead to a decision or action within a window that a scheduled overnight refresh never meets.

A good analytics platform depends on choosing the right stream-processing layer, storage that can keep up with continuous writes and reads, a visualisation layer that doesn't choke at a few hundred events per second, and alerting that doesn't multiply false positives. Those ingredients are our focus, together with the data integration that connects your existing systems to the platform.

What sets a real-time platform apart.

Streaming-first

Events are processed the moment they arrive, not in the next batch.

Sub-second latency

From source to dashboard or alert within the time window the decision deserves.

Alerting & action

Not just showing, but warning and, where sensible, intervening automatically.

Industry logic

OEE, SPC, conversion attribution or churn models that don't fit standard BI.

Six applications we often build.

In practice they overlap. A real-time platform for a manufacturer often also includes IoT sensor data, and an e-commerce platform often relies on the same fraud detection as financial services. We determine together which combination fits, based on a conversation about your sources, latency requirements and desired actions.

Application 01

Manufacturing and production

OEE tracking, andon systems, SPC monitoring and predictive quality alerts for production floors. Sensor and MES data is read in continuously, enriched with order and quality data, and shown visually to production managers as soon as a line falls behind or a process moves outside its control limits. For organisations that see real-time production data monitoring as essential to meeting delivery dates and quality margins, often linked to an existing manufacturing software landscape.

OEE trackingAndonSPCPredictive qualityMES integrationSensor streams

Application 02

E-commerce real-time analytics

Live conversion tracking, cart monitoring, real-time pricing and fraud detection for webshops and marketplaces. The platform reads event streams from your storefront and back office, enriches them with customer context, and gives you insight within seconds into pages with poor conversion, falling cart values or suspicious transaction patterns. For e-commerce organisations that want to turn real-time data from their e-commerce systems and integrated flows into marketing, pricing and fraud decisions, without waiting for the next reporting cycle, often via our API integrations.

Conversion trackingCart monitoringDynamic pricingFraud detectionEvent streamsShopify / Magento

Application 03

Logistics and supply chain

Fleet tracking, real-time shipment status, route anomaly detection and SLA monitoring for logistics providers. GPS and telematics data is combined with order and route data, so planners and account managers can see within seconds where a shipment is deviating, where a driver has stopped, or where a delivery is at risk of breaching its SLA. Often combined with customer portals, where shippers and consignees track their own shipments live.

Fleet trackingTelematicsRoute anomalySLA monitoringCustomer portalGeo data

Application 04

Telecoms and network monitoring

Network health dashboards, customer experience monitoring and churn prediction for telecom, hosting and SaaS providers. The platform processes network events, customer interactions and service tickets, and flags incidents or customer groups with a sharply increased churn risk within seconds. Alerting flows reach the right teams at the right moment via PagerDuty, Opsgenie or your own incident flow.

Network healthCX monitoringChurn predictionPagerDutyOpsgenieIncident flow

Application 05

Financial services and transaction monitoring

Live transaction monitoring, fraud detection and market data platforms for financial services, fintech and trading firms. The platform processes large streams of transaction or market events with sub-millisecond latency, compares patterns against trained anomaly models, and blocks or escalates suspicious cases without a human in the loop. For compliance-heavy environments, we build the audit layer and data lineage as an integral part of the solution.

Transaction monitoringFraud detectionMarket dataAnomaly modelsAudit logCompliance

Application 06

IoT, marketing and customer support

Three adjacent patterns that share much of their architecture with those above. IoT platforms for temperature monitoring in cold chains, vibration analysis for predictive maintenance, and occupancy tracking in buildings. Marketing platforms for campaign performance, A/B test results and multi-touch attribution. Customer support dashboards with ticket volume, response times and sentiment analysis across incoming channels. Each is built on the same streaming foundations, with the visualisation and alerting layer tailored to the work of the target users and running on an underlying cloud-native platform.

IoT / cold chainPredictive maintenanceCampaign performanceAttributionTicket volumeSentiment

What you get at the end.

A production-ready real-time analytics platform, plus everything needed to manage, extend and operate it independently and with integrity.

The platform

Production and staging environments, hosted in your cloud or with us.

Streaming pipeline

Ingestion, processing and storage, with end-to-end monitoring and a backpressure strategy.

Dashboards and alerting

Visualisation, alerting flows and APIs to systems that act within your landscape.

Codebase and documentation

Full source code, runbook, architecture overview and a clear handover to your IT team.

Managed service (optional)

Monitoring, security patches, capacity planning and further development in follow-up sprints.

When custom is the right choice.

Six patterns we often see before an organisation chooses its own platform instead of a standard BI or SaaS analytics tool.

Latency requirements

Sub-second, not 5-minute batches

Operational decisions in production, e-commerce or trading cannot wait for the next scheduled refresh. Standard SaaS tools usually work in batches of 5, 15 or 60 minutes, which is structurally too slow for the patterns where real-time really matters.

Industry logic

The calculations don't fit standard tools

OEE, SPC, multi-touch attribution, churn models, fraud scores on transaction patterns: calculations that fit poorly, or not at all, in a drag-and-drop tool or a SaaS formula language. A custom calculation engine is faster to build here, easier to test and still readable years later.

Embedded analytics

Analytics within your own product

Customers expect real-time insights to appear in your application or customer portal, not in a separate Looker or Tableau environment with different branding. Embedded analytics in your own UX, with the same login and permissions as the rest of your product.

Multi-tenant and white-label

A platform for your customers

SaaS platforms that want to offer their customers real-time analytics need multi-tenant isolation and white-label branding, requirements generic SaaS tools were not built for. Data separation happens at database level, branding per tenant, and billing based on usage rather than user licences.

Scale and cost

SaaS pricing becomes unworkable

SaaS analytics often charges per user, per event or per gigabyte. At hundreds of millions of events a day or thousands of end users, that scales to amounts that undermine the business case. A custom platform on an open-source stack or cloud-native managed services scales more predictably.

Hybrid edge and cloud

Latency calls for local processing

Production lines, vehicles or remote installations cannot tolerate a network round trip to a central cloud for every decision. A hybrid architecture with edge processing for the hot path and cloud aggregation for the slow path is then the only workable approach, which is difficult to achieve with standard SaaS.

How an engagement works.

01Introduction→ 02Mapping data and latency→ 03Build in sprints→ 04Rollout and management

No-obligation conversation

Introduction

Which events flow in, which decisions the platform must support, and which latency that requires. We listen to your situation and give direction where that helps.

Workshop with your team

Data model and architecture

Stream sources, event schemas, storage choice and latency budget. Output: a workable data model and a prioritised scope for the first releases.

Fortnightly sprints

Build and validation

A working build with real data every sprint. Your team tests along the way, alerting rules are tuned in production, and the platform grows with what we learn along the way.

Ongoing

Rollout and management

Phased rollout per use case or business unit, handover to your management team and further development in subsequent sprints. Monitoring and capacity planning continue.

Tech stack we work with.

For each project we choose what fits your data volume, latency requirements and existing cloud landscape. For stream processing we rely on Apache Kafka (Confluent Cloud), AWS Kinesis, Google Pub/Sub, Apache Flink or Spark Streaming. For storing streaming data we work with ClickHouse, Apache Druid, InfluxDB, TimescaleDB or Apache Pinot, depending on query patterns and retention requirements. For the query and analytics layer, Apache Doris, StarRocks, Materialize and Tinybird are increasingly common. Visualisation relies on Grafana or on custom React and D3 dashboards where standard components fall short. Alerting via PagerDuty, Opsgenie, Slack and Teams. For anomaly detection and predictive analytics we work with isolation forest, Prophet, LSTM networks and whatever suits the data pattern. We have no allegiance to a single framework; the choice is always a function of the use case, not a preference.

Stream processing

Apache KafkaConfluent CloudAWS KinesisGoogle Pub/SubApache FlinkSpark Streaming

Storage & querying

ClickHouseApache DruidInfluxDBTimescaleDBApache PinotMaterializeTinybird

Visualisation, alerting & ML

GrafanaReact / D3PagerDutyOpsgenieSlack / TeamsAnomaly detection

Frequently asked questions.

Do you replace Snowflake, Tableau or Power BI?

No. Snowflake, Tableau, Power BI, Looker, Sisense, ThoughtSpot, Domo and Qlik are excellent standard tools for batch analysis and reporting. We don't replace them, and often run alongside them. We build custom when there is a genuine real-time requirement (sub-second latency), when calculations don't fit standard tools, when analytics needs to be embedded in your own product, when the platform has to run multi-tenant or white-label, or when SaaS costs become unviable at the scale you need.

What is the difference from a KPI dashboard?

A KPI dashboard looks at metrics, often on data refreshed on a schedule (nightly or hourly). A real-time analytics platform captures streaming events the moment they occur, enriches and analyses them in motion, alerts teams or triggers automated action. The dashboard is then just one output channel. For scheduled reporting, a custom KPI dashboard is usually the right, cheaper choice.

What latency requirements can you meet?

That depends on the pipeline. For sensor or event streams from source to dashboard, we typically achieve sub-second latency, and with edge processing and optimised stream pipelines we can reach tens of milliseconds on the hot path. We never give a hard latency guarantee without first aligning the architecture to your sources and requirements, as that would be an empty promise. During the architecture phase, we determine what is realistic and sensible for each use case.

Can the platform be embedded in our own product?

Yes, and it is one of the most common reasons to choose custom. We build analytics components as a web component, SDK or iframe embed, with the same authentication and authorisation as the rest of your application. Multi-tenant isolation sits at database level, and branding can be set per tenant. Your customers see their own data in your look and feel, not that of a SaaS vendor.

How do you handle IoT and sensor data?

Through MQTT, AMQP or HTTP ingestion into a streaming broker (Kafka, Kinesis or Pub/Sub), with optional edge processing on-site for latency-critical decisions. Events then pass through a Flink or custom processor and land in a time-series store (InfluxDB, TimescaleDB) or a columnar store (ClickHouse, Druid), depending on query patterns. We build IoT platforms for temperature monitoring in cold chains, vibration analysis for predictive maintenance, occupancy tracking and more.

How do you set up alerting?

We work with PagerDuty, Opsgenie and direct integrations into Slack or Teams. More important than the channel is the rule set: thresholds, hysteresis, deduplication and escalation, so teams are not buried under false positives. We apply anomaly detection on streaming data (isolation forest, Prophet, LSTM) where static thresholds are too crude. Alerting flows are tuned in production on real data, not in advance in a spreadsheet.

What determines the cost of a real-time platform?

The largest cost factors are the number and complexity of source streams, the required latency, the choice between managed services (Confluent, Tinybird, Materialize) and open source on your own cloud, the degree of custom visualisation, and whether anomaly models or automated actions are included. A first use case with one stream and standard alerting is a project of a few sprints. A multi-tenant platform with several streams and custom ML is a larger project spanning several sprints. After our introductory conversation, we provide direction, with full transparency about what each scope decision means.

When is batch processing actually sufficient?

A fair question, and the answer is more often "yes" than clients expect. Where decisions are made at day or hour level, where dashboards are mainly used for monthly or quarterly reporting, or where no alerting or action flow is needed, a well-designed batch architecture with data integration and a custom KPI dashboard is almost always cheaper and more stable. In that case we do not recommend a real-time platform: unnecessary complexity is costlier to maintain, year after year.

Talk to us about your real-time analytics platform.

A free, no-obligation half-hour introductory call. We will listen to which streams come in, which decisions they need to support and what latency that requires. Would you rather start with broader context on custom software development? That is possible too.

Schedule an introductory call Email fabian.vandijk@appfront.nl