Azure Integration Services

The integration layer built properly: Logic Apps, Service Bus, API Management and the operational discipline around them.

Engagement

Assessment + implementation. Assessment and architecture first where an estate already exists. Build and configuration from there.

Who it is for

IT leaders, integration and platform teams, and the architects who inherit whatever gets built.

When people call us

Point-to-point connections have multiplied past anyone's understanding, an upgrade broke six things nobody had documented, or a BizTalk estate is out of runway.

The practice

01

Integration assessment

What exists, what it connects, what it costs to run, and which interfaces have no owner. Usually the first complete inventory anyone has had.

02

Integration architecture

Message contracts, ownership of records, synchronous against asynchronous, batch against event — designed before any resource is deployed.

03

Logic Apps

Standard and Consumption, chosen deliberately: Standard where you need VNet integration, predictable cost and local development, Consumption where the workload is genuinely occasional.

04

Service Bus

Queues and topics where systems must be decoupled, with sessions for ordering, dead-letter handling that someone actually watches, and replay when a downstream system comes back.

05

API Management

A front door with policy on it — authentication, rate limits, transformation and versioning — rather than a proxy that forwards whatever it is given.

06

Event Grid and Functions

Event-driven flows, and custom code where a connector will not do the job honestly. Small, testable, and owned.

07

Data Factory

Bulk and scheduled data movement, kept separate from transactional integration so one cannot take the other down.

08

Configuration and hardening

Managed identity instead of stored credentials, secrets in Key Vault, private endpoints where the data requires it, and connections that do not belong to a person.

09

Deployment and ALM

Infrastructure as code, environments that match, and a release path that is not somebody clicking through the portal on a Friday afternoon.

10

Monitoring and support

Application Insights and Log Analytics wired to alerts an owner receives, with runbooks for the failures that will happen anyway.

How a message survives a bad night

Decoupled, retried, replayed, watched.

A synchronous chain fails as a unit: one system down and everything upstream fails with it. This is the shape that does not.

  1. 01
    Published

    The sending system completes its own work and hands the message off. It does not wait for anything downstream.

  2. 02
    Queued

    Service Bus holds it — with ordering where the process needs it, and no loss if the receiver is offline.

  3. 03
    Processed

    Logic Apps or Functions do the work, against a contract both ends agreed to.

  4. 04
    Confirmed

    Success logged, metrics emitted, and the message removed from the queue.

When it fails, which it will

Retry with backoff, then dead-letter rather than discard. A queue somebody monitors, a replay path that has been tested, and an alert that reaches an owner instead of a shared mailbox. The integration you never think about is the one quietly dropping messages.

We assess the integration estate, design the architecture, and build it: Logic Apps, Service Bus, API Management, Event Grid, Functions and Data Factory — with managed identity and Key Vault, infrastructure as code, and monitoring wired to an owner.

The estate is the problem before the platform is

An assessment usually produces the first complete list anyone has had: every interface, what it moves, how often, who owns it, and what it costs. That list is what makes the rest of the conversation possible, because the honest answer for a third of those connections is that they should be retired rather than migrated.

Standard or Consumption is a real decision

Consumption Logic Apps are cheap until they are not. Per-action pricing on a high-volume interface, no VNet integration, cold starts, and cost that scales with a mistake. Standard gives you predictable pricing, virtual network integration and local development, at the cost of paying for the plan whether or not it runs. We make that call per workload, with the volume numbers in front of us.

Queues are what make it survivable

A synchronous chain of calls fails as a unit: one system down, and everything upstream of it fails with it. Service Bus decouples that — the sending system completes, the message waits, the receiving system catches up. What matters then is the operational discipline around it: dead-letter queues somebody monitors, replay that has been tested, poison-message handling, and alerts that reach an owner rather than a shared mailbox.

API Management earns its licence with policy

A gateway that only forwards requests is an expensive DNS entry. The value is in the policy layer: authentication and authorization in one place, rate limits that protect the systems behind it, transformation so a consumer is not coupled to an internal schema, and versioning so changes do not break every caller at once.

Migration off what you have

BizTalk estates, scheduled file drops, database links and the point-to-point sprawl that grew around them. We map what each interface does, decide what survives, and move what survives onto a design rather than lifting it as it is.

Integration estate nobody can map? Bring it to a briefing.

What you keep

The code and the data·Your existing relationships·Approval and control·The ability to stop·The off switchWhat that means

What we build on

The services this is built from.

Chosen per workload against volume and cost, not adopted as a set.

Logic Apps
Orchestration. Standard for VNet integration and predictable cost, Consumption where the workload is genuinely occasional.
Service Bus
Queues and topics where systems have to be decoupled, with dead-letter handling somebody watches.
API Management
Authentication, rate limits, transformation and versioning in one place.
Event Grid and Functions
Event-driven flows, and custom code where a connector will not do the job honestly.
Data Factory
Bulk and scheduled movement, kept apart from transactional integration.
Key Vault and managed identity
Secrets and identities that do not belong to a person who might leave.

Start with a 45-minute briefing.

No pitch. We’ll map your situation against what actually works and tell you honestly where to start.