Find the pattern behind a change.
Patterns in your data
Explore groups, unusual transactions and changes in your business records. See which data supports a finding and where further investigation is needed.
Which customers are ordering differently?
Compare purchasing patterns over the chosen period. Review the customers and transactions behind a change before deciding how to respond.
Which accounts have changed their buying pattern?
Compare the same measure across two periods before investigating the cause.
Data miningIllustrative records| Customer | Previous period | Current period | Change |
|---|---|---|---|
| Alder UK | 72 | 54 | −25% |
| Harbour Foods | 84 | 84 | 0% |
| Atlas Trade | 120 | 132 | +10% |
Alder UK has the largest decline in this group.
Open the underlying orders and review the context. The pattern does not, by itself, explain why the change occurred.
- Groups — customers, suppliers or products that behave alike in the ways that matter to your trade.
- Things that go together — what is bought together, what is returned together, which account codes keep coming up on a supplier’s bills.
- The usual order of events — the order things tend to happen in, before process mining checks whether that order was allowed.
- When something changed — the date sales or costs really shifted, rather than the date somebody noticed.
- Odd ones out — figures that don’t fit, found in a way that one extreme value can’t throw off.
- Duplicates — the same customer, supplier or product entered twice under two spellings.
- Gaps in your data — where details are most often missing, which usually points to a process problem rather than a data one.
- Where forecasts go wrong — the cases where Cobalt ERP’s forecasts are consistently off, so they can be improved.
- Leads worth checking — figures that move together, offered as things to look into and never as conclusions.
- Clusters of closely linked records, where the links between them make that reliable.
A finding that comes with its caveats.
Each finding records which records it covers, how it was found, how strong and how steady it is, and the period it was compared with. That makes it hard to quote a number out of context, because the caveats travel with it.
- The question it set out to answer, and which records it looked at.
- The snapshot it ran on, so the run can be repeated.
- The method and settings used, and the type of pattern found.
- How common the pattern is, how big the effect is, and how steady it is.
- The period it was compared with, and a check that it isn’t a fluke from testing many patterns at once, where that applies.
- Examples — the actual records behind the pattern.
- Data warnings, its limits, and a label that can’t be removed, saying it describes what happened rather than why.
It never says one thing caused another.
Two things happening together can easily read as one causing the other, and AI writing tools will happily phrase it that way. Cobalt ERP blocks this twice. Every finding carries a label saying it only describes what happened, and the part that writes answers in plain English isn’t allowed to reword a pattern as a cause. Working out a cause is a separate job, with a higher bar for evidence.
Wherever a question starts with “which” or “what changed”.
- “Why did this quote lose margin?” — the discounting pattern behind it, compared with an earlier period so it means something.
- “Which customers behave like this one?” — a group, what defines it and how steady it is.
- “What else is bought with this?” — suggestions on a sales order, with how often they are bought together.
- “Where are we missing data?” — gaps grouped by who entered the record and when, which usually points to a process problem.
- “Have we got this supplier twice?” — likely duplicates, sent to whoever keeps your customer, supplier and product records tidy.
Frequently asked questions.
How is this different from forecasting?
Data mining describes what is already in your records. Forecasting estimates something you don’t know yet, such as next month’s sales. The maths overlaps but the questions are different, so they are kept separate, each with its own limits.
Why not just say the pattern causes the outcome?
Because usually it doesn’t, and software that quietly treats a coincidence as a cause will cost you money sooner or later. Claiming a cause needs a proper test, not just two figures that move together.
Can a pattern change a rule automatically?
No. It can suggest one. Approving a rule is a person’s decision, and it is recorded as theirs.
What stops a finding being cherry-picked?
Each finding shows the period it was compared with, how steady the effect is, and whether it was checked for flukes. A pattern that only holds in one period says so.
Try it on your own records.
Book a demo and ask it a real question about your business. Every answer shows how it was worked out and which records it used.
Book a demo