The fragmentation
index, published in full.
Plant data fragmentation is scored here out of 100 across four components: how many of your systems exchange data on their own, worth 35 points; how much machine data is entered by hand, worth 20; where machine and shift logs actually live, worth 20; and how many hours a week your team spends moving data between systems by hand, worth 25. A higher score means more fragmented. The whole model is below, including every option and what it is worth.
We publish it because a score you cannot inspect is a score you cannot argue with, and the parts worth arguing with are the parts that will improve it.
The four components
| Component | What it asks | Points |
|---|---|---|
| Systems not sharing data | Of the systems you run, how many exchange data without a person exporting, reformatting or retyping anything | 35 |
| Machine data entered by hand | How many machines report their own data without a person entering it | 20 |
| Where logs are kept | Where machine and shift logs actually live, not where policy says they live | 20 |
| Manual data movement | Hours per week spent moving data between systems by hand | 25 |
| Total | 100 | |
Systems not sharing data, out of 35
This one is proportional rather than banded. Count the systems in real use, count how many of them exchange data with at least one other automatically, and the score is the share that do not, scaled to 35. Eight systems with two connected scores 26 of 35. Connect everything and it is zero.
It is the largest single component deliberately, and it is the only one that gets worse as you add tools. That is the behaviour we wanted: another system nobody has connected makes the problem bigger, not smaller.
Machine data entered by hand, out of 20
| Answer | Points |
|---|---|
| None | 20 |
| A few critical machines | 14 |
| Most of the main lines | 6 |
| Effectively all | 0 |
The step from none to a few is the largest in the table, because the first automatic feed is the one that proves it can be done and settles what the numbers are worth.
Where logs are kept, out of 20
| Answer | Points |
|---|---|
| Paper registers on the floor | 20 |
| Excel or PDF files on individual machines | 15 |
| A shared drive or folder | 8 |
| A system anyone authorised can query | 0 |
Paper and local files score close together on purpose. A spreadsheet on one machine is not meaningfully more readable than a logbook beside it. Both need a person to go and fetch them, and neither can be correlated with anything else.
Manual data movement, out of 25
| Hours per week | Points |
|---|---|
| Under 5 | 4 |
| 5 to 15 | 12 |
| 15 to 40 | 19 |
| More than 40 | 25 |
Under 5 hours still scores 4 rather than 0. Almost nobody is genuinely at zero, and a model that pretends otherwise reads as flattery.
The bands
| Score | Reading |
|---|---|
| 65 to 100 | Heavily fragmented |
| 35 to 64 | Partly connected |
| 0 to 34 | Well connected |
The thresholds mark where the character of the problem changes rather than its severity. Below 35 the useful next step is usually about asking better questions of data you already hold. Above 65 the plumbing is the problem, and no amount of reporting laid on top will fix it.
What counts as a system
Eleven categories, and three of them are routinely left out of this kind of count:
- ERP or accounting
- MES or production
- SCADA or historian
- Maintenance system
- Quality or lab
- Inventory or dispatch
- HR or attendance
- Sales or CRM
- Spreadsheets doing real operational work
- Chat and email threads where decisions actually get made
- Paper registers, logbooks, shift diaries and checksheets
The last three are the point. Leave them out and a plant looks better connected than it is, because they are exactly where the reconciliation work has gone to hide. A spreadsheet four people edit is a system, whatever the software register says.
The hours figure is not part of the index
It sits beside the score and is calculated separately. The weekly answer band maps to 3, 10, 27 or 50 hours, and that is multiplied by 48 working weeks to give an annual figure.
It is the respondent's own estimate and the report says so rather than presenting it as a measurement. Its value is not precision. It is that the same question asked again after a change produces a number comparable with this one.
What the model refuses to do
When a respondent says a component is outside their view, the index is not computed at all. It is null rather than partial, and the report names which components are missing and how much of the 100 points is actually held. A score counts as a measurement only when every component is present and the person answering works with the systems in question.
That decision has its own page: what counts as a measurement, and what is a guess.
These weights are a considered judgement about what makes plant information hard to read across. They are not calibrated against outcomes. No claim is made that a score of 70 predicts anything in particular, and the band thresholds are reasoned rather than derived.
The index has been computed three times in total, and all three were our own runs while building and checking the tool. No outside respondent has completed it yet, so there is no distribution to compare a score against, and we publish no benchmark.
Publishing an uncalibrated model is deliberate rather than an oversight. Inspectable means it can be disagreed with, and disagreement is the only thing that will improve it.
Get your own score
Twelve questions about how your systems, machines and registers actually hold information. No signup. You get the score, the hours estimate, and the three things worth fixing first.
Your answers and your score are recorded when the report renders on your screen, which happens before any email is asked for. The page says so before the first question.
Take the diagnostic