Back to all articlesCRM measurement

Handle Unknown Lead Sources Without Inflating Channel Performance

Alactic Editorial10 min read

Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.

A channel report can reconcile perfectly and still overstate what you know. If leads with missing sources disappear from the denominator, or disputed leads quietly move into a preferred channel, the resulting conversion rates conceal an attribution decision.

Keep unknown and disputed as explicit reporting categories. Preserve the evidence behind each assignment, separate verified assignments from assumptions, and publish a sensitivity worksheet alongside channel conversion rates. The worksheet should answer one practical question: would a reasonable alternative allocation change the decision?

This approach treats incomplete evidence as a measurement limitation. It does not depend on recovering tracking data, repairing integrations, or changing lead-capture implementation.

1. Define the conversion question before judging source evidence

A source cannot be right or wrong without a defined attribution question. The channel that first introduced a prospect may differ from the channel that generated the enquiry being measured. A later sales conversation may supply another source claim without resolving either question.

Write a short measurement contract before reviewing records:

  • Unit: what counts once in the denominator, such as one unique lead.
  • Cohort: which leads enter the report and during what period.
  • Outcome: what event counts as a conversion.
  • Observation window: how long each lead has to reach that outcome.
  • Source rule: which acquisition event the channel label describes.
  • Evidence standard: what supports a resolved assignment and how conflicts are handled.

Hypothetical worked example: the following register covers 100 fictional unique leads created during one reporting month. Each has completed a 60-day observation window. A conversion means becoming a won customer within that window. Source means the channel that generated the initial enquiry, not the latest communication channel. All figures and records below are illustrative, not customer results.

Keeping the observation window fixed matters. A newly created unknown-source lead should not be compared with an older, fully observed referral lead. Attribution uncertainty and incomplete outcome observation are separate problems; this example isolates the former.

2. Give unknown and disputed different meanings

Unknown means the available evidence does not support one source under the agreed rule. A blank field is one possible cause, but so is a populated field whose origin cannot be established.

Disputed means relevant evidence supports conflicting assignments and the review has not resolved the conflict. Two different source values are not necessarily a dispute: an initial enquiry and a later contact may legitimately have different channels. Check that both claims concern the event being measured.

Resolved means the available evidence meets the reporting standard. It does not mean the organisation has proved the customer's entire journey.

Keep evidence status separate from channel. A useful record has a reported channel, an evidence status, the underlying observations, an assumption field, and a reviewer decision. For unresolved records, the reported channel remains Unknown or Disputed. Possible channels belong in a separate field, not in the reported channel.

Also distinguish acquisition source from contact medium. A lead contacted through WhatsApp is not necessarily acquired through WhatsApp. Communication history alone should not settle an acquisition-source question.

3. Fill a source-evidence register before reallocating anything

The fictional register below is grouped for readability. In an operational register, retain one row per lead so a reviewer can inspect each assignment. Each ID range below represents distinct leads, not duplicate records.

The register deliberately includes outcomes. They are needed for sensitivity calculations, but the source reviewer should apply the same evidence standard regardless of whether a lead converted.

FICTIONAL SOURCE-EVIDENCE REGISTER
Measurement: initial-enquiry channel; won within 60 days
Status: review complete for this reporting snapshot

IDs: P01-P30 | Leads: 30 | Won: 9
Evidence: enquiry records identify Paid under the agreed rule.
Status: Resolved | Reported channel: Paid
Assumption: none needed for assignment.
Decision: retain Paid; reviewer: reporting owner.

IDs: O01-O30 | Leads: 30 | Won: 6
Evidence: enquiry records identify Organic under the agreed rule.
Status: Resolved | Reported channel: Organic
Assumption: none needed for assignment.
Decision: retain Organic; reviewer: reporting owner.

IDs: R01-R20 | Leads: 20 | Won: 8
Evidence: initial-enquiry records identify a referral introduction.
Status: Resolved | Reported channel: Referral
Assumption: none needed for assignment.
Decision: retain Referral; reviewer: reporting owner.

IDs: U01-U03 | Leads: 3 | Won: 3
Evidence: no supported source for the initial enquiry.
Status: Unknown | Reported channel: Unknown
Scenario candidates: Paid or Referral only by assumption.
Decision: leave unresolved; owner: source reviewer.

IDs: U04-U12 | Leads: 9 | Won: 0
Evidence: no supported source for the initial enquiry.
Status: Unknown | Reported channel: Unknown
Scenario candidates: Paid or Referral only by assumption.
Decision: leave unresolved; owner: source reviewer.

IDs: D01-D04 | Leads: 4 | Won: 4
Evidence: enquiry source says Paid; later account note says
Referral and explicitly describes the initial enquiry.
Status: Disputed | Reported channel: Disputed
Scenario candidates: Paid or Referral, supported by conflicting claims.
Decision: neither claim overrides the other; owner: source reviewer.

IDs: D05-D08 | Leads: 4 | Won: 0
Evidence: same type of initial-enquiry conflict as D01-D04.
Status: Disputed | Reported channel: Disputed
Scenario candidates: Paid or Referral, supported by conflicting claims.
Decision: neither claim overrides the other; owner: source reviewer.

Review checkpoint for unresolved records:
Next reporting close; resolve only with additional relevant evidence.
If no new evidence exists, preserve the unresolved status.

The unknown records have no evidential basis for restricting them to Paid or Referral. That restriction is an explicit scenario assumption, not a finding. The disputed records, by contrast, have two evidence-backed candidates. Keeping this distinction visible prevents a worksheet assumption from becoming a CRM fact.

For a working register, add references to the underlying records, the review date, and a brief decision history. A reviewer should be able to reproduce the assignment without relying on someone's memory.

4. Publish a baseline that accounts for every lead

The baseline should retain all 100 leads and all 30 conversions. Channel rates are calculated only from the leads currently assigned to each channel, while unresolved categories remain visible.

HYPOTHETICAL BASELINE
Category        Leads   Won   Won / leads
Paid               30     9       30.00%
Organic            30     6       20.00%
Referral           20     8       40.00%
Unknown            12     3       25.00%
Disputed            8     4       50.00%
Total             100    30       30.00%

Resolved-source coverage: 80 / 100 = 80.00%
Unresolved-source share: 20 / 100 = 20.00%
Unresolved share of conversions: 7 / 30 = 23.33%

Unknown and Disputed are evidence categories, not acquisition channels. Their rates help describe which outcomes remain unattributed; they do not identify a channel to fund.

Among resolved records, the overall rate is 23 divided by 80, or 28.75%. That is not the cohort's overall conversion rate, which remains 30%. Label both denominators if you report both figures.

Dropping unresolved records does not always inflate every numerical rate. The more general failure is inflated certainty: a report can make Referral's apparent lead over Paid look settled when a substantial portion of the cohort has no settled source.

5. Build the sensitivity worksheet around actual outcomes

For each scenario, allocate whole leads with their observed outcomes intact. Moving a converted lead adds one lead and one conversion to the destination channel. Moving a non-converted lead adds one lead and no conversion. Never allocate conversions separately from the leads that produced them.

The calculation is:

Scenario channel rate =
(resolved channel wins + allocated unresolved wins)
/
(resolved channel leads + allocated unresolved leads)

The following fictional scenarios use the register above. They are alternative treatments, not revised evidence. Organic remains unchanged because these particular scenarios allocate unresolved records only between Paid and Referral.

HYPOTHETICAL ALLOCATION WORKSHEET
Notation: allocated leads / allocated wins

Scenario A: even lead split, mixed outcomes
Unknown to Paid:       6 / 2
Unknown to Referral:   6 / 1
Disputed to Paid:      4 / 2
Disputed to Referral:  4 / 2

Scenario B: Paid-favourable stress case
Unknown to Paid:       3 / 3
Unknown to Referral:   9 / 0
Disputed to Paid:      4 / 4
Disputed to Referral:  4 / 0

Scenario C: Referral-favourable stress case
Unknown to Paid:       9 / 0
Unknown to Referral:   3 / 3
Disputed to Paid:      4 / 0
Disputed to Referral:  4 / 4

RESULTING CHANNEL RATES
Treatment      Paid             Organic         Referral
Baseline       9/30 = 30.00%    6/30 = 20.00%   8/20 = 40.00%
Scenario A    13/40 = 32.50%    6/30 = 20.00%  11/30 = 36.67%
Scenario B    16/37 = 43.24%    6/30 = 20.00%   8/33 = 24.24%
Scenario C     9/43 = 20.93%    6/30 = 20.00%  15/27 = 55.56%

CHECKS FOR EVERY ALLOCATED SCENARIO
Channel leads sum to 100.
Channel wins sum to 30.
Each unresolved lead is allocated exactly once.
Overall conversion remains 30.00%.

Scenario A is a balanced allocation by lead count, not a probability estimate. Scenarios B and C deliberately align outcomes with opposite channel assignments to expose sensitivity. They are feasible under the stated candidate restrictions, but there is no evidence here that either pattern is likely.

These results are not confidence intervals. They show the consequences of specified allocations. If Unknown could also include Organic or an unlisted channel, add scenarios for those possibilities. The worksheet above does not establish that Organic's rate is immune to source uncertainty.

6. Translate sensitivity into a decision statement

The baseline says Referral converts at 40% and Paid at 30% among their currently resolved records. Scenario A preserves that ordering but narrows the gap. Scenario B reverses it. Scenario C widens it in Referral's favour.

A defensible interpretation is therefore narrower than “Referral performs best.” It is:

Referral has the highest observed conversion rate among resolved-source records. However, 20% of leads and 23.33% of conversions remain unresolved. Under the tested allocations, Paid and Referral can exchange rank, so their conversion-rate ordering is not robust to attribution uncertainty.

This statement preserves the observed result without presenting it as settled. It also identifies the decision at risk: a channel preference based on this conversion-rate ranking alone.

Choose scenarios that match the decision. If the team is considering a threshold rather than a ranking, test whether alternative allocations cross that threshold. Avoid filling the report with scenarios that cannot affect the proposed action.

Even a stable ranking would not establish that a channel caused better outcomes. This worksheet addresses incomplete source evidence, not differences in lead mix or other explanations for conversion performance.

7. Prevent convenient assumptions from becoming permanent history

Several shortcuts make reports cleaner while weakening their meaning:

  • Assigning every blank to a default channel. An empty source is not evidence of Organic, Direct, or any other channel. Restore unsupported assignments to Unknown.
  • Letting the winning outcome settle a dispute. Commercial importance does not strengthen source evidence. Re-review converted and non-converted records using the same rule.
  • Applying the same allocation percentage to leads and wins. This can create fictional lead-outcome combinations. Allocate identifiable records or clearly labelled aggregate scenarios that preserve feasible counts.
  • Replacing the baseline with a modelled allocation. Keep the evidence-based snapshot and the scenario output separate. Otherwise an assumption can later appear to be an observed fact.
  • Rewriting prior reports without explanation. Preserve the original snapshot, publish the revised version, and show how reassignment changed denominators, conversions, and rates.

If a published report has already excluded unknowns, start recovery with reconciliation. Recover the full cohort count, separate source uncertainty from other exclusions, and reproduce the old result before presenting a corrected view. Do not silently overwrite the original channel totals.

There is a trade-off between simplicity and visibility. A compact dashboard can show the baseline plus unresolved share, while a linked worksheet holds the detailed scenarios. What should not disappear is the warning that the headline comparison depends on unsettled evidence.

8. Run a measurable review at the next reporting close

For the next completed cohort, assign a reporting owner to produce four items: the measurement contract, the source-evidence register, the reconciled baseline, and a sensitivity worksheet for the decision being considered.

  1. Confirm that every eligible lead appears exactly once.
  2. Review unknown and disputed records without using conversion success as an assignment rule.
  3. Separate evidence-backed candidate channels from assumed candidates.
  4. Calculate at least one balanced scenario and opposing decision-relevant stress cases.
  5. Record whether the proposed channel conclusion survives those alternatives.

Track unresolved leads as a share of the cohort, unresolved conversions as a share of all conversions, and the number of unresolved records without a documented review decision. These measures make uncertainty visible without rewarding unsupported reassignment.

The immediate completion test is concrete: another reviewer can account for every lead, reproduce the reported rates, and explain which channel conclusions change when unresolved attribution changes. A lower Unknown count is not success if it was achieved by replacing missing evidence with confidence.

CRM measurementLead attributionConversion reportingData quality

Put the next conversation to work.

Explore Alactic Communications for WhatsApp conversations, campaigns and AI-assisted workflows. Choose the product and plan that fit your team.

Explore Communications