Uncia
FR / EN Demo

Glossary

Technical debt

Technical debt is the work an IT organization has postponed and will have to do eventually: an upgrade, a dependency to document, a workaround to replace. Until that work is done, every change costs more and takes longer.

Updated

What is technical debt?

Technical debt is the backlog of work a system owes itself: the upgrade that was postponed, the dependency nobody wrote down, the workaround that became permanent. None of it stops the system from running today. All of it makes the next change slower and riskier, and that extra cost is what people call the interest.

The metaphor comes from software development. In an experience report dated March 1992 and written for the OOPSLA conference, Ward Cunningham wrote that shipping first-time code is like going into debt: a little debt speeds development as long as it is paid back promptly with a rewrite, and the danger comes when it is not repaid. He was writing about code. The phrase "technical debt" took hold afterward, and usage stretched it.

Most articles on the subject stop at code. In an IT department, a good share of the debt sits in infrastructure, in the network and in documentation. Nobody wrote bad code there. An operating system reached end of support, a firewall rule outlived its migration, an architecture document stopped following the platform. Infrastructure technical debt has one awkward property: it lives in no code repository, so no code analysis tool will report it.

What does infrastructure technical debt look like?

Four kinds show up in almost every estate.

Unsupported systems. A server runs an operating system or database version the vendor no longer patches. It works, which is exactly why the migration slips from one budget to the next. This is the easiest debt to date, because end-of-support dates are public (see software obsolescence: definition and method).

Undocumented dependencies. An application calls a database on a server everyone believes is dedicated to something else. A nightly job drops a file on a share nobody listed. You find out the day the machine is switched off, or halfway through a hypervisor migration, when a cutover wave breaks an application that was not in it.

Firewall rules nobody can justify. The rule was opened for a project, the project ended, the requester left. It still allows a whole subnet into a sensitive zone, and nobody dares close it without knowing what uses it.

Stale documentation. The architecture document describes the platform as it was at go-live. Since then, machines have changed hosts and flows have been added. This debt makes the other three worse, because it hides them.

How do you inventory technical debt across an IT estate?

Asking teams what bothers them returns this month's irritants, not the risks. An inventory starts from the sources that describe what runs, and puts one precise question to each.

To the hypervisor export or the cloud inventory: which operating system versions are running, and which are no longer maintained? To the CMDB: which servers have neither an application nor an owner attached? To the firewall rules, set against the flow matrix: which rules match no written requirement? To the architecture documents: which have not been reviewed since go-live, and which contradict the exports?

Each answer becomes a line in a debt register. A useful line carries the item concerned, the kind of debt, the source that establishes it and its date, the applications that depend on it, and an owner. Without an owner the line will never be handled. Without a source it will be challenged in committee.

The first inventory is incomplete, and it should say so: undocumented dependencies are, by definition, in no file.

How do you measure technical debt and prioritize it?

Putting a dollar or euro figure on technical debt is tempting and rarely holds up. The cost of a migration can be estimated. The cost of not doing it depends on an incident that has not happened yet, and a total presented without a method loses its credibility at the first question.

More modest measures hold better: the number of unsupported machines and their share of the estate, the number of critical applications resting on at least one unsupported system, the number of firewall rules with no justification, the age of the last review of each architecture document. These counters do not add up to one number. They can be tracked from quarter to quarter, which is enough to tell whether the debt is growing or shrinking.

To prioritize, two questions usually do the job. What depends on the item? An unsupported server that carries payroll does not weigh the same as a test server, and an impact analysis is how you establish it. What deadline applies? An end-of-support date six months out, a hardware maintenance contract that ends, an announced audit.

The rank of a line comes from crossing the two answers. At the top, whatever carries a critical activity with a deadline that has passed or is close. At the bottom, whatever only affects an isolated scope. In between, group by project: clearing the debt of a platform during the migration that already touches it costs less than coming back to it later.

How do you present technical debt to a steering committee?

A steering committee does not decide on a list of versions. It decides on a risk, a deadline and a cost. So the presentation starts from the business: this activity relies on that system, which has not been patched since a given date.

Each topic needs very little. What is exposed, in business terms ("invoicing", not "srv-app-12"). The evidence, with its source and date, so that nobody argues about the finding. The request, framed as options: migrate this year, isolate the system in the meantime, or accept the risk in writing until a stated date.

Written acceptance matters as much as budget. Show as well what earlier decisions have closed since the last session.

Which tools help track technical debt?

A spreadsheet is the usual starting point. It is enough for a register of a few dozen lines kept by one person, and it can be shown as is. Every review means collecting the data again.

Code analysis tools measure the debt of the software you develop: duplication, complexity, outdated libraries. They do it well and see nothing of the infrastructure.

A CMDB holds owners and the relationships between configuration items. When it is fed by discovery, it gives installed versions. It does not usually say whether those versions are still maintained, or whether the documentation is accurate.

An enterprise architecture tool links a technology component to an application and a business capability, and can carry a lifecycle per component. It is the right place to reason about a roadmap. Its content is entered by architects and trails the estate.

The last approach is mapping derived from the sources: read the exports, the rules and the documents, reconcile them, and build the register from the result. It is only as good as the files it reads.

What Uncia brings, and what it does not do

Uncia reads the sources of your IT estate in read-only mode (connectors to clouds, hypervisors, CMDB, firewalls and scanners; dropped-in files such as an RVTools export, a flow matrix or an architecture document) and reconciles them in a knowledge base. Several lines of a debt register come out of it with their source attached. End-of-support dates are taken from endoflife.date, per product and per machine, and every sentence of the obsolescence report points to its fact. The matrix of required flows is compared with the firewall rules, with a verdict per flow, including "unjustified". Architecture documents you drop in are compared with the sources, and the gaps are shown. The impact of a shutdown tells you which applications and flows go down with a machine.

Uncia does not put a price on the debt and does not rank it: it supplies sourced findings, and a person decides. It changes nothing in your systems, does not observe traffic and does not analyze your application code. It only knows what your sources describe, so a dependency that no file mentions escapes it too. What it knows is updated at each collection.

Sources

Frequently asked questions

Where does the term technical debt come from?

The debt metaphor comes from Ward Cunningham. In an experience report written for OOPSLA in 1992, he compared shipping first-time code to going into debt, acceptable as long as it is paid back promptly with a rewrite. He was writing about code; usage later extended the image to the rest of IT.

Is technical debt only about code?

No. The term comes from software development, but an IT department also carries infrastructure debt: unsupported systems, undocumented dependencies, firewall rules with no justification, stale documentation. None of it shows up in a code analysis tool.

What is the difference between technical debt and obsolescence?

Obsolescence is one form of technical debt: software its vendor no longer maintains. Technical debt is broader and includes things with no end-of-support date, such as an undocumented dependency or an architecture document that fell behind.

Can technical debt be measured in money?

The cost of a remediation project can be estimated. The cost of waiting depends on incidents that have not happened, so any total remains an estimate. Simple counters tracked over time are easier to defend in front of a committee: unsupported machines, critical applications affected, rules with no justification.

Should all technical debt be paid back?

No. Some debt can be kept knowingly, for instance on an isolated system already scheduled for decommissioning. What matters is that the decision is made by someone with the authority to make it, written down, and given a review date.

Every week, an article to know your IT estate better

Architecture documents, end of support, NIS2, CMDB, migrations: methods you can apply with or without Uncia.

Get a call from our team

Thirty minutes with a member of our team, remote or on site: Uncia on a demonstration instance, then on part of your own estate if you wish. We reply within 48 hours.

Request received

Thank you. A member of our team will reply within 48 hours to set a date.