The cost of running five systems is not five licence fees. Licences are the visible part, and usually the smallest.
The expensive part is what happens in the gaps. It does not appear in a budget, has no owner, and is absorbed by people as a normal part of their day. That combination makes it very durable.
Attention does not resume where it left off
Moving between applications is not free even when it is fast. Each switch requires reloading context: which record you were looking at, what you had already checked, what you were about to do next.
The mechanical time is trivial. The reconstruction is not, and it is repeated every time. Work that crosses four systems does not cost four times the effort of work in one, because each crossing also costs the re-establishment of where you were.
The visible symptom is that tasks take longer than the sum of their steps, and nobody can say quite where the time went.
Someone becomes the integration layer
When two systems do not talk, a person carries information between them. This is rarely anybody's stated job. It is absorbed into whoever's role sits nearest the gap.
That work is genuinely skilled — it requires knowing what both systems mean by a given field, which one is authoritative, and what to do when they disagree. It is also completely undocumented, because it was never designed. It exists in the habits of whoever has been doing it.
The exposure becomes obvious when that person is unavailable. Work does not slow down proportionally; it stops, because nobody else knows the translation.
Every duplicate entry is a place to be wrong
Entering the same information twice creates the possibility of two versions of it. Over time that possibility becomes a certainty.
The failure is rarely dramatic. It is a vendor name spelled differently in two places, so a report undercounts. A rate updated in one system and not the other, so an invoice does not match. A start date entered from memory, so payroll is a week out.
Each is individually small and individually fixable. Collectively they produce a quieter problem: nobody fully trusts any single number, so decisions get delayed while figures are checked against each other.
Reconciliation is permanent work
Once two systems can disagree, someone has to check whether they do. That work never finishes, because it is not a project. It recurs every period, forever, and it grows with volume.
Reconciliation is also unusually demoralising. In the good case, the outcome is that nothing was wrong, and the effort produced no visible change. In the bad case, you have found a discrepancy and now have to trace its cause across systems.
Decisions get made on whatever is nearest
This is the cost that matters most and gets measured least.
When the current figure requires opening a different system, asking someone for access, and waiting for an export, people frequently decide using the number they already have. Usually it is close enough. Sometimes it is not, and the decision is quietly worse than it needed to be.
Nobody records this. There is no log of choices made on stale data. The cost is real and permanently invisible.
The overheads nobody budgets for
Beyond the daily friction:
- Access management. Five systems mean five sets of permissions to grant, review, and revoke. Revocation is the step most often missed, which is a security exposure rather than an inconvenience.
- Onboarding. A new joiner learns five interfaces, five sets of conventions, and the informal knowledge of which system is authoritative for what.
- Audit. Answering a straightforward question about a past decision means assembling evidence from several places, none of which was designed to be read together.
- Vendor overhead. Five renewal cycles, five support relationships, five sets of terms.
How to estimate your own
You do not need a study. Take one routine process that crosses systems — onboarding a vendor, or hiring one person — and trace it end to end.
Count how many systems it touches, how many times the same information is entered, and how long it takes in elapsed rather than active time. Then ask who would be blocked if the person who normally does it were away.
That last answer is usually the most revealing. If the process depends on undocumented knowledge held by one person, you are not paying for five tools. You are paying for five tools plus a fragile, unrecorded integration layer made of somebody's memory.
Why this persists
Because no single instance is worth fixing. Each duplicate entry takes a minute. Each reconciliation is an afternoon. Each context switch is seconds.
The aggregate is significant, and it never presents itself as a single problem large enough to schedule. It presents as work, which is why it survives indefinitely.