How do I check whether the lead source field in our field service software is actually accurate?
Pull a month of jobs, filter to the ones that started with a tracked phone call or a web form, and compare the recorded source against what the tracking system says. Three numbers tell you most of what you need: percent blank, percent set to the default or Other, and percent that disagree with mechanical tracking. Any of the three above a modest level makes source-based reporting unusable.
Run the audit on a sample, not the whole database
Take one recent month. Restrict to first-time customers, because repeat customers legitimately have an older source. Join each record to the call tracking log by phone number and time, or to the web form log by email. You now have two opinions about where each lead came from: what a person selected and what the system observed.
Three hundred rows is plenty. You are looking for pattern, not precision.
The three numbers, and what each one means
- Blank rate. Source is not required, or is required in one intake path and not another. This is a configuration fix, and it is the easiest win.
- Default-value rate. If the field defaults to a value and nobody changes it, that value inherits every lazy entry. A source that suspiciously matches whatever sits first alphabetically is a default, not a finding.
- Disagreement rate. The customer came in on a tracked number assigned to paid search but the office marked them as referral. Some of this is real — people are referred and then search your name — but a high rate means the field is guesswork.
The bias you will find in almost every account
Human-entered source systematically over-reports word of mouth and under-reports paid channels. The reason is conversational: a customer asked how they heard about you usually says a friend recommended you or that they found you online, and the intake person maps that to the friendliest bucket. Meanwhile the tracked number that rang says otherwise.
The practical response is not to scold the office staff. It is to stop asking the field to do a job that instrumentation does better. Let tracked numbers, form parameters and click identifiers establish channel, and reserve the human field for what only a human can capture, like the name of the referring customer. Call analysis can also pull a stated source directly out of the conversation, which gives you a third opinion without adding a step for the CSR.
What to do with a field you cannot trust yet
Do not delete it and do not report on it as if it were clean. Publish it with its own accuracy metric attached — this month source was recorded on a given share of new customers and agreed with tracking on a given share of those. A number with a stated confidence gets used carefully. A number without one gets used to make decisions it cannot support.
Over time the audit becomes a monitoring job rather than a project, which is where marketing intelligence earns its keep: the drift gets caught the month it starts, not the quarter after someone makes a budget decision on it.
Topics: lead source · data quality · audit · attribution
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.