The Source Field Is a Capture Problem, Not a Report Problem
Every service business has a source field. In the CRM it is a custom field somebody added in the first month. On the intake form it is the question at the bottom that asks how did you hear about us. In the dispatch software it is a free text note. The field exists everywhere and is enforced nowhere, so the business concludes from the resulting numbers that marketing does not work, when the field was simply never captured.
The reason is structural rather than moral: the person who fills in the field is almost never the person who knows the answer. A technician at a house hears that the customer found you online and that the neighbour recommended you, in the same breath, and writes down whichever was mentioned last. A CSR with a queue waiting has exactly one value that never requires a follow up question, and it is referral.
So the field fills with three kinds of values. Some are true and useless, because they name a channel and not the campaign. Some are the word referral, which is a relationship and not a channel. Some are blank. The fix is to stop treating the field as a question somebody answers and start treating it as a value the system records, because the browser already knows the referrer, the parameters, the landing page, and the time, and none of that requires the customer to know or admit anything.
The reframe
If the source field accepts free text, you do not have a source field, you have a comment box that happens to sit in a column.
How Much of Your Source Field Can Be Trusted
Set the sliders to match your intake. The model shows how many leads carry a clean first-touch source, what the unreadable remainder is worth in revenue you cannot attribute, and what share of a backfill would be a guess.
Leads with a clean source
56
47% of 120 leads
Unusable for a channel decision
64
24 blank, 14 wrong, 26 channel only
Annual revenue you cannot attribute to a channel
$241,920
$20,160 per month, at 35% close and $900 per job
Office hours recovered
1.9
per month, hand reconciliation against a captured field
Reconstructed backfill
53%
would have to be labelled as a guess, not as data
What the source field actually contains
record: channel + campaign + landing URL + arrival time
Written by the system when the lead arrived, not by a person when somebody answered the phone. The only bucket that supports a channel decision without a human cleaning it first.
record: channel = Google
True and nearly useless. It says search brought the lead and nothing about which ad or offer. You can hold money still on this value, not move it.
record: source = Dave, or source = referral
A relationship record, which is worth storing, but not a channel. Nobody can total these rows against a marketing line item, so the row cannot be reported on.
record: source = null
The largest bucket, and the one nobody defends. The lead arrived, the job was booked, the field was never touched, and the report quietly describes a smaller business.
record: source = Google referral
Produced on an afternoon when the office needed the screen to close. Worse than blank, because blank is visibly missing and wrong reads as a real number in a real column.
The five buckets sum to all records by construction, and 12% of them carry a source somebody could act on without a human cleaning it first. Only the first bar is gold, because only the first bar is a value a report can move money on.
Illustrative estimates from the inputs above. Your own numbers will differ; these are not measured results or client data.
From Arrival to Report, Stage by Stage
Six stages from the moment the lead lands to the moment a report states which touch it is describing. Each names the system it touches, the payload it leaves behind, and where a naive capture breaks.
Target System
The website, the landing page, and the form
The Lead Arrives, and the Session Carries the Answer
The moment a lead lands, the browser already knows things the person filling in the form never will: the referring page, the campaign parameters in the URL, the landing page they read, and the time. This is the only moment when the source can be recorded without asking a question.
The form posts the name and the phone and throws the rest away, so the source is reconstructed later from memory.
Stage payload
Capturing Source Without Asking Anybody
Two mechanisms do almost all of the work, and neither is visible to the customer. The first is the hidden field on the form: the page reads the campaign parameters from the URL, the referring page, and the landing path, and posts all of them alongside the name and the phone. The customer sees four fields and submits five values, and the fifth is the one a marketing report runs on.
The second is the tracking number. A number unique to a channel, a campaign, or even a specific ad means the source is known the instant the call connects, before the CSR says good morning. The greeting is unchanged and the only difference is that the business knows which piece of spend produced the phone ringing.
Neither covers everything. A customer who drives past a truck, asks a neighbour, and then searches your name three days later arrives with directed traffic and no parameters, and that lead is genuinely unlabelled. The honest move is to let those records sit in a known bucket, because a report that says we do not know on a share of leads beats one that guesses on all of them.
Two capture points, in one line
The standard campaign parameter set
Google's own documentation for URLs it builds lists five parameters and marks three required: utm_source, utm_medium, and utm_campaign, with utm_term and utm_content optional. That standard set is enough. The failure mode in a service business is not too few parameters, it is four parallel tagging conventions because three people built links in three different months.
What the Capture Does Alone, With You, and Never
Recording is safe, deciding is not. The capture can write a source onto every lead and change nothing else, which is what makes it reasonable to point at a live intake flow.
Autonomous
- Read the referrer, the campaign parameters, the landing path, and the arrival time at capture
- Write the source onto the form submission without prompting the visitor
- Normalise an arriving value against the controlled list, including known synonym mappings
- Reject a value that is not on the list and route it to a queue rather than into the report
- Carry the source through the lead-to-job handoff so it lands on the job record
- Report the share of records in each source bucket, including how many are blank
Supervised
- Propose a corrected value for a record whose source is wrong or missing
- Propose a synonym mapping when a new value clearly means an existing one
- Propose a backfill for a period where the field was not enforced, flagged as reconstructed
Never
- Invent a source for a record that has none
- Report a reconstructed value as though it had been captured
- Ask the customer how they heard about you and store the answer as data
- Explain why a lead is missing a source when the record does not say
- Merge two channels into one value to make a report reconcile
- Publish a channel total that exceeds the jobs the business actually ran
Three Things We Will Tell You Not to Buy
A report built on a source field nobody enforces
Every dashboard vendor will chart a field that does not hold real values, and the chart will look authoritative because charts always do. If the field is free text or filled by somebody guessing, the report measures typing behaviour. Fix the capture first. The report is the last thing you build on this, not the first.
A how did you hear about us field treated as data
It is a fine rapport question and a bad data source. Asking the customer produces a polite answer, arrives too late to verify, and cannot be audited because no other record exists to check it against. Keep the question if it helps the conversation, but do not let it write into the field the attribution report reads.
A source taxonomy too granular for anyone to use
A list of forty source values looks thorough and gets ignored, because the person choosing from it is under time pressure and picks the first plausible option every time. Twelve values covering every channel you actually buy will be used correctly. Granularity belongs in the campaign field, not the source field.
How to Sequence the Field Cleanup
The order matters, because a report built before the field is enforced will be used to make a decision the field cannot support.
Week 1
Write down what the field holds today
Export every distinct value and count them. The number of spellings of Google is the argument for everything that follows, and nobody believes it until they see it.
Weeks 2 to 3
Replace the box with a list
Turn the field into a controlled list of about twelve values. Map the existing values onto it and let everything unmappable fall into an unknown bucket rather than being forced.
Weeks 4 to 5
Capture at arrival
Add the hidden form fields and the tracking numbers so new leads carry a source the customer never had to supply. Leave the historical records flagged as reconstructed.
Ongoing
Report one touch, and say which
Build the channel report on first touch, state that on the report, and reconcile the channel totals against the jobs actually run every month.
Questions About Lead Source Reporting
Where This Fits in Marketing and Reputation
This page owns the source field itself: its capture, its normalisation, and the reporting that reads it across every channel. The platform-specific reads and the dashboard that displays them are separate pages, linked below.
Find Out How Much of Your Source Field Is Real
Give us read access to your CRM and the export of your source field. In thirty minutes we will show you how many distinct values it holds, how many leads carry one you could act on, and what the unreadable share is worth in revenue you cannot currently attribute. You keep the analysis whether or not you hire us.
