Our lead source field is mostly garbage. How do we fix it?

Why the field degrades so predictably

Ask a CSR who is mid-call, with another line ringing, to classify where a caller came from. They will pick whatever is first in the list or whatever they picked last time. Ask a customer directly and they will say they found you on Google, which is true of nearly everyone and tells you nothing about which of your channels did the work.

The result is consistent across companies: an oversized Other or Unknown bucket, Referral inflated because it sounds safe, and search over-credited because search is where people finish rather than where they started.

Make the system write it

Every source that can be detected automatically should be. Tracked phone numbers identify calls. Hidden form fields carry click identifiers and campaign parameters. Booking widgets can pass the referring channel through. What remains genuinely unknowable is a much smaller set than most companies assume.

Set the field on record creation and make it read-only afterward. If staff need to add context, give them a separate free-text note field so it cannot contaminate the structured value. This is standard practice in a CRM integration and applies identically to field service platforms.

Shrink and structure the list

• Fewer options, sharper definitions. Twelve well-defined values beat forty overlapping ones. If two values could describe the same lead, merge them.
• Separate channel from campaign. Channel is paid search, organic, LSA, social, referral, repeat. Campaign is the specific effort. One field cannot carry both.
• Keep repeat customers out of acquisition. Existing customers are not new leads. Mixing them makes every cost-per-lead number wrong.
• Add an explicit Unknown. Real and labeled beats fake and confident.

Track your unknown rate on purpose

Put the share of records with an automatically determined source on your reporting as a standing number. It is a data-quality metric, and watching it go up is the clearest evidence that tracking work is paying off. It also tells everyone reading the report how much of the picture is actually measured.

When the unknown share is large, do not fill the gap with assumptions. Use modeled comparison instead — correlating spend changes against total booked revenue through marketing intelligence gives a defensible read on channels that resist direct attribution, without pretending to a precision you do not have.