What would leaving actually cost?
What it costs to move a TIBCO BusinessWorks 5 estate onto Apache Camel. The shape of the estate and the current maintenance spend in; effort estimate and payback period out. Assumptions are shown, not hidden. Argue with them.
The estimate
pd is person-days: one person, one working day. Divide by about 250 for person-years, or by your team size for a rough calendar.
The saving is the maintenance line less the 15% that does not go away when the engine does: hosting, support, and the people who run it.
Where the effort goes
| Class | Count | Coefficient | Person-days |
|---|---|---|---|
| Processes the generator left clean: nobody opens them | 89 | 0 pd / process | 0 |
| Processes with a marker in them: open, read, decide, fix | 31 | 0.14 pd / process | 4.2 |
| Adapters and EDI | 0 | 3 pd / activity | 0 |
| Per component: build, packaging, deploy, smoke test | 20 | 0.5 pd / component | 10 |
| Estate fixed: platform, CI, observability, a way to compare behaviour | 1 | 20 pd | 20 |
| Build | 34.2 | ||
| Tests and equivalence runs, 25% on top | 8.5 | ||
| Total | 42.7 |
The coefficients
A number you cannot audit is worth nothing to you, so here is every one the model uses. Where the coverage split comes from is stated beside it. The effort figures are estimates from real migrations. They are the part to argue with, and the part that moves most between estates.
| Coefficient | Value | Where it comes from |
|---|---|---|
| Translated share | 96.4% | From the toolchain coverage report, weighted by how often each type occurs |
| Caveat share | 2.3% | From the toolchain coverage report, weighted by how often each type occurs |
| Manual share | 1.4% | From the toolchain coverage report, weighted by how often each type occurs |
| A process with a marker in it | 0.136 pd | Field estimate: open it, read the warning, decide, fix. Per process, not per activity: an estate this shape is mostly repetition, so pricing every activity as its own problem overstates a hand migration by more than an order of magnitude. What the generator buys is fewer processes to open, not a cheaper hour. |
| An adapter or EDI activity | 3 pd | Field estimate: no generator covers these |
| Per component | 0.5 pd | Field estimate: build, packaging, deployment, smoke test. Per deployable unit, not per .process file: in a real estate most .process files are subprocesses another process calls, and they are never deployed on their own. |
| Estate fixed cost | 20 pd | Field estimate: platform, CI, observability, a way to compare behaviour |
| Test overhead | 25% | Field estimate: applied to everything above |
| Retained run cost | 15% | Field estimate: the share of the maintenance line that survives the migration |
| Uncertainty band | ±30% | Field estimate: the honest width of a figure produced from five inputs |
What the model deliberately leaves out: parallel running, data migration, re-certification of anything regulated, and the calendar cost of a change freeze. They are real, they are specific to you, and a calculator that guessed at them would be lying.
The business-case template
A spreadsheet with this model in it, plus the slide structure these numbers are usually presented in: current spend, effort, payback, and the risks a change board will ask about. Leave an email address and the files appear below.
Want help applying this to your own estate? Email me directly.