Skip to main content

9 min readPublished November 8, 2026Talavishiraaj Groups
Technology & Engineering

Visibility is usually bought as a screen and fails as a data problem. If three systems disagree about where something is, a dashboard renders the disagreement faster. What makes visibility real is deciding which system is authoritative for each fact, capturing events rather than snapshots, and putting the result where somebody already works.

In short
  1. 01A dashboard built on disagreeing sources renders the disagreement faster. It does not resolve it.
  2. 02Decide which system is authoritative for each fact before building anything. That argument is the project.
  3. 03Capture events rather than nightly snapshots, or you can report what happened and never what is happening.
  4. 04Validate detection against incidents you already know about before trusting it on ones you do not.
  5. 05Visibility that arrives somewhere nobody already works will be checked for a fortnight and then ignored.

What is supply chain visibility?

Knowing where things are and what is about to go wrong, early enough to act. That second half is what separates it from reporting: a system that tells you accurately on Thursday what went wrong on Tuesday is a record, not visibility.

The distinction matters commercially because the two cost very different amounts. A record can be assembled from nightly extracts. Acting in time needs events as they happen, which means integration work against systems that were not built to emit them.

Most disappointment in this category comes from buying the first and expecting the second.

Why do visibility projects fail so often?

Because the hard part is not the screen. A transport system, a warehouse system and a spreadsheet will each hold a different version of where a consignment is, and a dashboard laid over the top renders all three faster without deciding between them.

Once that happens, trust goes quickly. People notice the number is wrong, check the underlying system instead, and the visibility layer becomes a thing nobody looks at that still costs money to run.

The failure is almost always upstream of the tool, which is why changing the tool rarely fixes it.

How do you tell a reporting problem from an integration problem?

Ask which system is authoritative for the fact you care about most. If somebody can answer immediately, you have a reporting problem and it is tractable: the data exists and is trustworthy, it just is not in front of the right person.

If the answer takes a meeting, you have an integration problem wearing a reporting label, and it is a larger piece of work. That is not bad news if it is discovered now. It is expensive news if it is discovered after a platform has been bought.

The same test applies per field rather than per system. It is common for a business to be authoritative about location and hopeless about expected arrival, and only the second one matters for acting early.

What does the work actually involve?

Deciding ownership of each fact, which is an organisational conversation rather than a technical one and is usually the longest part.

Capturing events rather than states. A nightly snapshot tells you what was true at midnight; an event stream tells you something moved at eleven. Only the second supports acting before a customer notices.

Establishing a baseline before adding detection. Anomaly logic validated against incidents nobody has confirmed produces alerts nobody believes, and the fastest way to kill an alerting system is to be wrong in its first fortnight.

And putting the output where people already are. An answer in a tool somebody has open anyway gets used. An answer requiring a separate login at a moment nobody chose does not.

When is buying a platform the right answer?

When your flows are standard, your carriers are already integrated with the platform, and the gap is genuinely a user interface rather than a data model. That is a real situation and the build case against it is weak.

When it is not the right answer: where the useful signal lives in systems the platform does not speak to, where your definition of an exception is specific to how you operate, or where the integration work is the project and the platform is the easy part at the end of it.

Worth costing the exit before signing either way. Switching cost on operational platforms accumulates quietly through integrations and workarounds, and it is far easier to estimate before the contract than after.

When should you not build this at all?

When nobody can name the decision that changes. If the honest answer to "and then what happens" is that someone looks at a screen, the value is not there yet.

When the volume of exceptions is low enough that a person checking twice a day is genuinely sufficient. Engineering and monitoring do not get cheaper because the problem is rare.

And when the underlying process is not written down. Visibility over an undocumented process shows you the symptoms of decisions nobody has recorded, which generates argument rather than action.

Where this work sits
Technology & Engineering →

We build and integrate the production systems a business runs on: internal platforms, customer facing applications, integrations, and the data layer underneath them.

Weighing a decision like this one?

Describe the situation and we will tell you how we would approach it.