Checking whether your funnel data is true before you act on it

4 days ago
3 min read

Introduction

Decisions get made from numbers without anybody having asked whether the numbers are correct. Usually they are approximately correct. Occasionally a form has been silently failing for six weeks, a phone line diverts to a voicemail nobody checks, or an entire channel stopped being logged when somebody changed jobs.

These failures are not rare and they are not obvious, because a missing enquiry leaves no trace. The report looks normal, slightly lower than last year, and the explanation offered is market conditions. Meanwhile the cause is a broken notification.

Half an hour of deliberate checking, twice a year, catches almost all of it.

1. Checking whether your funnel data is true starts with a test enquiry

The most valuable check.

Submit your own form from a phone on mobile data, using an address nobody at the company recognises. Then see what arrives, where, and how quickly. This finds more problems than any other single action. Do it on a Sunday evening as well as a Tuesday morning.

2. Test every route in

The completeness check.

Form, phone, email, messaging, the listing that has its own contact button. Each is a separate path and each can fail independently. Most businesses have never tested the ones they use least. The rarely used routes are the ones most likely to be broken.

3. Compare the log against the raw records

The reconciliation.

Take one week and count enquiries in your inbox and phone log, then compare with what was recorded. A gap of ten per cent is normal; a gap of forty is a process failure. Do it for one week rather than a month. Choose a busy week rather than a quiet one.

4. Check the invoices match the wins

The other end.

Jobs recorded as won should reconcile against the accounts for the same period. Differences reveal double-counting, cancelled work or jobs that never got logged. This is a quick check with a high hit rate. Reconcile a quarter at a time.

5. Look for impossible values

The data quality scan.

Negative durations, dates in the wrong order, values of zero, sources spelled four different ways. Sorting each column surfaces these in seconds. They are usually entry errors rather than anything deeper. Fix them at source rather than in the report.

6. Check the categories are being used consistently

The classification check.

If half the source column says web, online and website, the report is dividing one channel into three. Look at the distinct values once a quarter. Tidy them and constrain the entry if you can.

7. Ask whether anything stopped being counted

The drift check.

New person, new system, a channel that moved. This is the commonest cause of a mysterious decline. Ask it explicitly rather than waiting for someone to volunteer it.

8. Check the automated things still run

The silent failure.

Notification emails, forwarding rules, integrations between systems. All of them break without announcement, often after an unrelated update. Test rather than assume.

9. Put the check in the calendar twice a year

The habit.

Unscheduled verification does not happen, and the value comes entirely from doing it before you need it. Half an hour, twice a year. Write down what you found.

Be careful about assuming a fall is a data problem and a rise is real. The instinct to investigate only bad numbers means data faults that flatter you survive indefinitely, so apply the same check in both directions.

Conclusion

Submit a test enquiry through every route into your business, twice a year.

Reconcile one week of logged enquiries against your inbox and phone records, check that recorded wins match the invoices, sort each column to find impossible values, review the distinct entries in your source field for the same channel spelled several ways, ask explicitly whether anything stopped being counted, confirm the automated notifications and integrations still run, and book the whole check into the calendar.