Absent or in an ivory tower: the two pitfalls of IT architecture
IT architecture suffers from two opposite pitfalls. In some organisations, it is absent: each project chooses its technologies, creates its own interfaces and duplicates the data it needs. The information system grows through successive additions, point-to-point flows multiply, and nobody can say any longer which application is authoritative for a customer, a contract or a piece of equipment.
In others, it exists but remains at a distance. A team of architects produces comprehensive maps, out of date before they are finished, and a target that projects discover when they go before the committee. Architecture is then experienced as a toll: a late approval, which can only delay or rubber-stamp.
In both cases, the flaw is the same. Decisions that commit the information system for a long time are taken without an overall view, or with an overall view that arrives after them. Our involvement aims to bring architecture back to the moment when it is useful: before spending is committed, in contact with projects.
Enterprise architecture, urbanisation, solution architecture: the concepts
Three terms are in circulation, and they do not sit at the same scale.
Enterprise architecture considers the whole organisation. It links strategy and processes to applications, data and infrastructure, and describes the roadmap for moving from the existing situation to a target. Frameworks such as TOGAF, published by The Open Group, offer a method for carrying out this approach, and languages such as ArchiMate a notation for representing it.
IS urbanisation applies the reasoning of a town plan to the information system. It divides it into zones, districts and blocks with distinct responsibilities, and lays down traffic rules between them: a piece of data has a single responsible block, and one block does not directly access another block’s data.
Solution architecture sits at the scale of a project. It describes how an application or a set of applications meets a need while respecting the common framework.
These approaches read the information system through superimposed views. Urbanisation distinguishes four: business (the processes), functional (the functions to be provided, independently of the tools), application (the applications and their exchanges) and technical (the infrastructure). The value of this breakdown is practical. It makes it possible to observe, for example, that three applications cover the same function, or that a critical process relies on an unsupported component.
Integration: choosing between APIs, events and files
Integration is the area where the absence of rules becomes costly the fastest. Three exchange modes coexist in many information systems, and none replaces the other two.
| Mode | Principle | Suitable when | Point of caution |
|---|---|---|---|
| Synchronous API | One application queries the other and waits for the response | An immediate response is needed to continue | The caller depends on the availability of the called system, versions must be managed |
| Events | One application publishes a business fact, subscribers react to it | Several consumers, deferred processing acceptable | Consistency achieved with a delay, message ordering and duplicates, schema governance |
| Files and batches | Data is transmitted in bulk at a fixed time | Large volumes with no freshness requirement | Error recovery, processing window, dated data |
Beyond the mode, a few rules prevent many drifts. Each interface has a written, versioned contract and an owner. Each piece of reference data has a master application. No application reads another’s database directly. The integration platform routes and secures exchanges, but carries no business rules: these remain in the applications responsible for them.
Flows intended for analysis and performance management obey other constraints of volume and history. They fall within the data platform, with which the integration architecture must be designed consistently.
Our approach: architecture in contact with projects
Our consultants work within your teams; decisions remain yours.
Starting from the decisions to be taken. We do not start by mapping the entire information system. We identify the decisions awaiting an architecture answer (rationalising an application domain, opening data to partners, preparing an application modernisation) and we map what they require, at the useful level of detail.
Defining a target that can be reached in stages. A target without a roadmap guides nobody. We describe intermediate stages that each leave the information system in a consistent and operable state. If the programme is interrupted, what has been done retains its value.
Writing few principles, but enforceable ones. A list of forty principles is rarely applied. We recommend retaining those that settle real disagreements, each with its justification, its consequences for a project and the cases where an exemption can be discussed.
Writing architecture documents that set out a decision. A project document presents the requirements, including non-functional ones (availability, security, volumes), the options actually studied, the one chosen and what is accepted in choosing it. Structuring decisions are recorded in a log, in the form of short notes, so that their rationale remains accessible after their authors have left.
Setting up governance that serves projects. The architecture body intervenes early, on criteria known in advance, and responds within a time frame compatible with project schedules. Exemptions are allowed, tracked and given a deadline for resolution. The most effective route remains making the right choice easy: ready-to-use templates, shared services, preconfigured foundations, in the spirit of platform engineering.
The drifts to avoid
- Mapping for the sake of mapping. A map that nobody consults to make decisions is not kept up to date, and becomes wrong.
- Applying a framework in full. Enterprise architecture frameworks are toolboxes. Rolling them out in full keeps architects busy producing documents rather than working on projects.
- Reducing governance to an end-of-design committee. At that stage, the choices have been made and the budget committed. The body can only approve or block.
- Concentrating business logic in the integration layer. Business rules lodged in flows become invisible and make every change dependent on a single team.
- Splitting up on principle. Multiplying independent services when neither the organisation of the teams nor the needs for change justify it adds operational complexity without benefit.
How to measure the contribution of architecture
Architecture is judged by its effects on projects and on the information system, not by the volume of its deliverables. The indicators we suggest tracking fall into three categories.
On the functioning of governance: the share of structuring projects reviewed before the main spending is committed, the response time of the body, the number of open exemptions and how long they have been open.
On the state of the information system: the number of interfaces without a contract or owner, the share of reference data with a master application, redundant applications actually retired, the date of the last update of the maps.
On usage: the number of projects that reuse an existing API or service instead of creating a new one. These measurements also feed the decisions of project portfolio management, where a project’s impact on the architecture becomes one prioritisation criterion among others.