Make Revenue Attribution Visible

Task

Create a source-of-revenue and source-of-lead attribution log.

Summary

Create enough attribution discipline to know which sources and channels produce opportunities.

Build an Attribution Log That Connects Leads to Revenue

Task ID: S1-05

A useful attribution log records how a real sales opportunity entered the business, preserves the evidence behind that classification, and connects the opportunity to revenue. It gives a founder-led company enough reliable information to compare referrals, warm relationships, inbound interest, campaigns, and other channels without pretending that a simple source field proves what caused the sale.

A founder can often explain the company’s first sales from memory.

One customer was introduced by a former colleague. Another came through an existing client. A third had seen a presentation months earlier, later visited the website, and finally sent an email. The stories are known, but they are scattered across inboxes, calendars, call notes, and the founder’s recollection.

That works until someone asks a basic operating question: Where is the company’s good business actually coming from?

The customer relationship management system, or CRM, may show opportunities and contract values, while the marketing system shows website visits and form submissions. Neither one records the full story. “Referral” is applied to almost everything involving another person. “Web” describes records that were entered through a form, even when the buyer originally heard about the company through a customer. Campaign names are missing, inconsistent, or overwritten. Closed revenue cannot be traced back to a credible source.

The immediate job is not to install a sophisticated attribution model. It is to create a small, dependable log that connects each legitimate opportunity to its best-supported source, channel, campaign, and referrer.

Start with the decision, not the attribution model

Attribution means assigning credit for an outcome to one or more events that preceded it. The term is often associated with advertising, but the underlying question is broader: What brought this potential customer into a real buying process?

That question needs a defined purpose. Research on digital attribution has warned that attribution is frequently used without first specifying the measurement objective or the decision it is supposed to support. A model designed to allocate advertising budgets is not necessarily the right model for understanding which relationships produce qualified consulting opportunities.

At this stage, the attribution log should support three practical decisions:

1. Which sources produce genuine opportunities rather than names or inquiries?
2. Which sources produce customers with attractive revenue, margin, retention, or repeat-purchase patterns?
3. Where should the company spend more attention, and where is its current evidence too weak to justify investment?

Those questions are especially important in a founder-led business built through referrals and a warm network. Referral business may behave differently from business acquired through other routes, but the effect is not uniform. A study that followed about 10,000 customers of a German bank for almost three years found higher contribution margins and retention among referred customers and estimated that their average value was at least 16% higher than that of comparable non-referred customers. The advantage varied by customer segment, however, which is precisely why a company should measure its own results rather than assume that every referral is equally valuable.

The operating principle is therefore:

Record the best-supported origin of every real opportunity, retain the detail behind the classification, and connect it to the eventual commercial result.

“Best-supported” matters. An attribution log is an evidence record, not a story constructed after the sale closes. The source should be based on information captured from the buyer, the referrer, a tagged link, a campaign record, correspondence, or another traceable source.

“Every real opportunity” also matters. Counting every newsletter subscriber or conference attendee may produce large numbers without improving the decision. The log should begin at a clearly defined opportunity threshold: for example, a buyer with a recognized problem, a plausible fit, and an agreed next sales step. The company must define that threshold before calculating its attribution rate.

Finally, “commercial result” means more than lead volume. The record needs to connect to the opportunity, customer, contract, and revenue data. Otherwise, the company can identify which sources generate activity but not which sources generate business worth building around.

Define the fields so they answer different questions

Source, channel, campaign, and referrer are related, but they are not interchangeable.

Official analytics documentation illustrates the distinction. Google Analytics defines a source as where traffic originates, a medium as how a person arrives, and a campaign as the specific marketing effort involved. Its standard manual-tagging parameters similarly separate utm_source, utm_medium, and utm_campaign.

A founder-led sales process needs a comparable structure adapted to both online and offline acquisition.

The last distinction prevents a common error. HubSpot’s documentation separates traffic source properties from its “Record Source” property. Original Traffic Source describes the first known web source, Latest Traffic Source describes the most recent, and Record Source describes how the CRM record was created—for example, by an import, a mobile application, or an integration.

A manually created record may still have originated through a customer referral. A record imported from a spreadsheet may originally have come from a conference. A website form may have been the mechanism used to request a meeting, even though a trusted adviser created the initial awareness.

Treating record creation as customer acquisition erases the information the company needs.

Use a controlled channel list

Channel should be a short picklist rather than unrestricted text. A practical starting list for a founder-led business could be:

• Customer referral
• Partner referral
• Founder or team network
• Organic inbound
• Paid promotion
• Direct outbound
• Event or community
• Product or service expansion
• Other
• Unknown

The exact categories should reflect how the business actually sells. “Founder network” may be important now but should not automatically remain a permanent reporting category. As the company becomes less dependent on the founder, it may be divided into more durable sources such as customer referral, employee referral, investor network, partner, or outbound.

Do not create dozens of categories at the start. A category that receives one opportunity every two years is unlikely to support a useful comparison. Preserve its detail in the Source field and group it into an appropriate Channel value.

At the same time, do not use “Referral” as a universal category. A customer referral, employee referral, informal founder introduction, implementation partner, and paid referral agreement can involve different costs, trust mechanisms, ownership rules, and customer outcomes. Combining them makes the report easy to complete but hard to use.

Preserve both broad and detailed values

The broad channel is needed for management reporting. The detailed source is needed for investigation.

GitLab’s public operating documentation shows this two-level approach at a much larger scale. It maintains granular initial-source values and maps them into broader “lead source buckets” for executive reporting. Its documented categories roll sources such as conferences, web inquiries, referrals, prospecting tools, product interactions, and partner activity into standardized groups.

The lesson is not to copy another company’s categories. It is to keep both layers:

• the original descriptive fact;
• the stable reporting category.

If only the broad category is retained, useful detail disappears. If only granular text is retained, small spelling and naming differences make aggregation unreliable.

Keep campaign optional but governed

A campaign is a specific organized effort, not a substitute for source or channel.

A buyer introduced informally by a customer may have no campaign. That is valid. Entering “Referral campaign” merely to fill the field creates false precision.

Where campaigns do exist, use an agreed naming convention. For example:

2026-Q3 | Webinar | Operations Reporting

The convention should identify the period, campaign type, and recognizable name. It should not rely on one employee remembering what an abbreviation means.

Online links should use consistent tagging. Google warns that when manual UTM tagging is used, setting only some relevant parameters can cause missing values to appear as “not set”; it recommends applying all relevant parameters consistently.

Record evidence without collecting unnecessary personal data

The referrer field can contain personal information, so it should not become an unrestricted notebook.

Where an individual’s identity is genuinely needed—for example, to thank a referrer or administer an agreed program—use an approved contact record or internal identifier. Where the analysis only needs the relationship, “Existing customer—manufacturing segment” may be sufficient.

Canadian private-sector privacy principles provide a useful general rule: collect only personal information needed for an identified purpose, use and retain it only for that purpose, and keep it appropriately accurate. The specific legal obligations will depend on the company’s location, customers, and systems.

Build the log into the sales process

The attribution log can live in a spreadsheet when opportunity volume is low, provided that each row has a stable opportunity identifier and someone controls changes. Once several employees update opportunities, automation or reporting depends on the data, or records must connect across systems, the CRM should become the main source of truth.

The minimum viable record is not complicated:

A company may also record the customer problem, expected result, offer, market segment, sales owner, and delivery model. Those fields support later decisions about which customer and problem to build around, but they should not be allowed to obscure the immediate requirement: reliable origin data on the opportunity.

The workflow should make attribution part of normal selling rather than a cleanup exercise performed before a management meeting.

flowchart LR
    A[Inquiry or introduction] --> B[Create or update customer record]
    B --> C[Create qualified opportunity]
    C --> D[Capture source, channel, campaign, and referrer]
    D --> E{Evidence sufficient?}
    E -->|Yes| F[Mark confirmed or probable]
    E -->|No| G[Mark unknown and assign follow-up]
    F --> H[Connect opportunity to outcome and revenue]
    G --> I[Review unknowns weekly]
    I --> D
    H --> J[Compare opportunity and customer results by source]

Text description: when an inquiry becomes a qualified opportunity, the owner records its acquisition fields and the supporting evidence. Unresolved records enter an exception review. Won opportunities are connected to revenue so the company can compare commercial outcomes by source.

Ask at the point where the answer is still available

For an offline introduction, the sales owner should confirm the origin during the first substantive conversation:

“How did you first hear about us, and what prompted you to speak with us now?”

This question separates initial awareness from the event that triggered action. The answer may be, “I have known the founder for years, but a customer recommended the reporting service last week.” Depending on the company’s decision rules, the first source might be founder network while the opportunity-creating source is customer referral.

Rather than forcing one field to answer both questions, preserve two when the distinction affects decisions:

• First known source: where the relationship or awareness began.
• Opportunity source: the interaction most directly associated with entering an active sales process.

Latest touch can also be useful, but it should not overwrite either of those fields. HubSpot’s separation of original and latest traffic sources demonstrates why both histories may matter. It also notes that tracking blockers can affect automatically recorded web-source values, so self-reported information and sales evidence remain important even when analytics tools are installed.

Make opportunity creation the control point

A required field is most effective when it is attached to a meaningful workflow event. For this task, that event is usually opportunity creation or advancement into a qualified stage.

The system should prevent or flag progression when:

• source is blank;
• channel is blank;
• a referral channel has no source detail or referrer;
• a campaign-dependent source has no campaign;
• “Other” has no explanation;
• the record uses a retired or invalid value.

Do not force an employee to invent an answer. “Unknown” is better than a fabricated category. The system should permit Unknown, but it should require an owner, a reason, and a review date.

The data also needs a steward. In a very small company, that may be the founder or sales lead. As volume grows, responsibility can move to sales operations or revenue operations. The steward owns field definitions, approved values, exception review, backfills, and changes to reporting logic. Sales owners remain responsible for entering or confirming the source.

Government of Canada guidance on data quality recommends distinguishing mandatory from optional fields, applying validation rules, recording issues, checking data against trusted sources, and maintaining metadata about how information was collected and processed. Those practices translate directly into an attribution log that can be audited rather than merely populated.

Backfill selectively

Historical records are tempting because they appear to offer an immediate trend line. They also contain the weakest evidence.

Start by backfilling:

• open qualified opportunities;
• recently won and lost opportunities;
• current customers whose source can be verified from correspondence, forms, contracts, or direct knowledge;
• high-value customers relevant to the company’s next strategic decision.

Label inferred historical values separately from confirmed values. Do not reconstruct years of source data from memory and then report it with the same confidence as newly captured records.

From the go-live date onward, the new process should be the primary measure of operational quality. Historical completeness can be shown separately.

Learn from established attribution systems without copying their complexity

Large commercial systems demonstrate both the value and the limits of simple attribution.

Salesforce allows campaigns to be associated with leads, contacts, opportunities, and revenue. Its standard campaign reports can show who was targeted, who responded, and how much revenue was associated with campaigns.

It also distinguishes a primary campaign from other influential campaigns. That arrangement illustrates a useful compromise: one primary value supports straightforward reporting, while additional relationships preserve a more complete journey. The primary assignment is still a reporting convention. It should not be interpreted as proof that the selected campaign independently caused the sale.

GitLab’s public lead-to-revenue documentation shows the next level of maturity. Its reporting model consolidates person, opportunity, account, and touchpoint information so analysts can examine funnel stages, conversion, cohort behaviour, and movement from lead creation through opportunity close.

A young founder-led company does not need that infrastructure on day one. It does need the underlying identifiers and distinctions that would make such analysis possible later.

The first approach is the right baseline for this task. The second becomes valuable when several touches genuinely influence opportunities and the company can capture them reliably. The third is appropriate when leadership needs to estimate incremental effect rather than merely trace observed journeys.

Academic research explains why the distinction matters. Li and Kannan’s multichannel study modelled individual customer touch histories, carryover, and spillover effects. It found channel contributions that differed significantly from those produced by commonly used attribution measures.

The business implication is straightforward: a primary-source log answers where observed opportunities came from under the company’s stated rule. It does not answer the counterfactual question, how many opportunities would have disappeared if that source or campaign had not existed?

That limitation does not make the log unhelpful. It makes its purpose clearer.

Measure coverage without confusing completeness with truth

The primary operating measure is the percentage of eligible opportunities with valid attribution.

“Eligible” must be defined. A sensible definition might include all new-business opportunities created during the reporting period after reaching the company’s qualification threshold. Test records, duplicates, internal records, and administrative renewals should either be excluded or reported in separate groups.

“Valid” must also be defined. An opportunity should count as attributed only when:

• the required fields are populated;
• the values come from approved lists;
• source and channel agree with the mapping rules;
• referral or campaign details are supplied where required;
• the source is supported by evidence or marked with an honest confidence level;
• the original value has not been improperly overwritten.

A record containing “Other” in every field is technically populated but not meaningfully attributed.

The working target of at least 90% tagged is a useful internal starting point, not a universal industry benchmark. Data-quality research emphasizes that completeness thresholds should be set for the purpose and context of the data and may differ across organizations and over time.

The appropriate target may differ because of:

• opportunity volume;
• the proportion of offline or indirect introductions;
• the length and complexity of the buying process;
• legacy records;
• data migration quality;
• the number of systems involved;
• legal or contractual limits on retained details;
• whether sources are captured automatically, manually, or both.

For a business creating ten opportunities in a quarter, one missing record changes the rate by ten percentage points. The report should therefore show both the percentage and the underlying count:

18 of 20 eligible opportunities validly attributed: 90%.

Separate recent operating performance from historical backfill:

Completeness is only the first quality check. Data-quality research distinguishes completeness from accuracy, consistency, integrity, accessibility, and timeliness. A CRM can be 100% complete and still be wrong.

A credible review should therefore include:

• Valid-value rate: the percentage using current approved categories.
• Unknown rate: the percentage honestly classified as unknown.
• Evidence rate: the percentage with a recorded evidence type.
• Audit accuracy: the percentage of a sample that agrees with forms, emails, call notes, campaign records, or buyer confirmation.
• Time to attribution: the time between opportunity creation and completion of the source fields.
• Reclassification rate: how often sources are changed after initial entry, and why.

Once the data is sufficiently reliable, compare business outcomes by channel and source:

• number of qualified opportunities;
• opportunity value;
• win rate;
• average or median contract value;
• sales-cycle length;
• revenue won;
• acquisition or referral cost, where identifiable;
• delivery cost and gross margin;
• renewal, repeat purchase, or expansion;
• concentration in one referrer, partner, or founder relationship.

These comparisons should show counts as well as rates. A channel with one win from one opportunity has a 100% observed win rate, but it does not yet provide strong evidence of a repeatable sales channel.

The analysis should also distinguish proof from interpretation.

Proof includes the recorded introduction, tagged link, buyer statement, opportunity amount, signed contract, payment, delivery cost, and renewal.

Interpretation includes judgments such as “customer referrals appear to produce better-fit buyers.”

A working hypothesis might be “referrals from customers in one segment are the most promising route for a repeatable offer.”

The log strengthens the hypothesis. It does not turn a small sample into certainty.

Recognize the failure modes before depending on the result

An attribution log can look finished while remaining unusable.

The field exists, but no one owns the definition

Different employees interpret “source” differently. One selects the first interaction; another selects the last; a third selects the campaign that appears most important. The report aggregates incompatible judgments.

The remedy is a one-page data dictionary with examples, edge cases, ownership, and an effective date.

Original source is overwritten by recent activity

A referred buyer later clicks an email, and automation changes the source to email marketing. The company gradually loses evidence of where relationships began.

Preserve original source as an immutable or tightly controlled field. Record subsequent touches separately.

Referral is too broad

Every warm opportunity becomes “Referral,” concealing whether it came from a customer, founder, employee, partner, investor, or professional adviser.

Use a broad referral channel only if a more specific source or relationship type is also retained.

Campaign is mandatory when no campaign existed

Employees enter invented campaign names to satisfy a required field. The completion rate improves while accuracy falls.

Allow “No campaign” as a valid state. Missing information and non-existent information are not the same thing.

The log stops at leads

Management celebrates the source that creates the most inquiries, even though another source creates fewer but larger, faster-closing, more profitable customers.

The source fields must follow the record into the opportunity and revenue data. Lead counts alone cannot answer the source-of-revenue question.

Closed-won records receive cleaner attribution than lost records

People naturally spend more time documenting successful deals. If failed and disqualified opportunities remain untagged, every source comparison becomes biased toward success.

Apply the same attribution requirement before the outcome is known, and include lost and disqualified opportunities in quality reviews.

A primary source is treated as causal proof

Observed journeys are affected by prior customer intent, channel interactions, selection effects, and missing touchpoints. More advanced research has shown that estimates can change materially when carryover and spillover effects are considered.

Use the log for tracing and comparison. Use experiments or carefully designed causal methods when the decision requires an estimate of incremental effect.

The company optimizes the tagging target instead of the data

A team facing a 90% target may fill blanks with defaults, choose the easiest category, or discourage Unknown. That behaviour produces a higher score and a less trustworthy dataset.

Review Unknown records as exceptions, but do not punish honest uncertainty. A lower rate with transparent evidence is more useful than a higher rate built on guesses.

Personal information accumulates in free text

Employees paste introduction emails, private relationship histories, or sensitive customer details into source notes.

Use structured fields, approved identifiers, limited access, and retention rules. Collect only what is necessary to classify and manage the source.

The work is complete enough to depend on when the following conditions are true:

• Every new qualified opportunity receives a stable opportunity ID.
• Source, channel, campaign status, and referrer detail have written definitions.
• Original source is preserved rather than overwritten.
• Record-creation method is separate from customer-acquisition source.
• Approved picklists and mapping rules exist.
• “Unknown” and “No campaign” are valid but governed states.
• A named owner reviews exceptions and category changes.
• Attributed-opportunity coverage is reported with numerator, denominator, and population definition.
• Accuracy is checked against a sample of supporting evidence.
• Won, lost, and disqualified opportunities are all included.
• Opportunity records connect to customer, contract, and revenue outcomes.
• Personal information is limited, protected, and retained for a defined purpose.
• Leadership can compare not only how many opportunities each source produces, but what happens to those opportunities after they enter the pipeline.

At that point, the company should be able to replace “Most of our work seems to come through people we know” with a more useful statement:

“During this period, these sources produced these qualified opportunities, these wins, and these customer outcomes. This portion is confirmed, this portion is probable, and this portion remains unknown.”

That is enough evidence to make the next decision clearer: which relationships, customers, problems, and routes deserve further attention—and which apparent patterns still need more sales before the company builds around them.

Sources

Primary and official sources

• Google Analytics, “Traffic-source dimensions, manual tagging, and auto-tagging.”
• HubSpot, “Understand Original and Latest traffic source properties,” updated June 11, 2026.
• Salesforce Trailhead, “Report on Your Campaigns.”
• Salesforce Help, campaign-influence documentation.
• GitLab Handbook, “Marketing Metrics,” including its published lead-source bucket mappings.
• GitLab Handbook, “Marketing Analytics Data — Lead to Revenue Models.”
• Government of Canada, “Guidance on Data Quality.”
• Office of the Privacy Commissioner of Canada, “PIPEDA Fair Information Principles.”

Open research

• Kelly, Vaver, and Koehler, “A Causal Framework for Digital Attribution,” Google Research, 2018.
• Li and Kannan, “Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment,” Journal of Marketing Research, 2014.
• Albrecht and colleagues, “Designing a Data Quality Management Framework for CRM Platform Delivery and Consultancy,” SN Computer Science, 2023.
• Schmitt, Skiera, and Van den Bulte, “Referral Programs and Customer Value,” Journal of Marketing, 2011.