#Before you buy attribution software, repair your CRM

If you are a technical founder or commercial leader struggling to explain where pipeline comes from, the problem is probably not that your attribution model is too simple. It is that your CRM does not reliably record what happened.

When I audit a company’s attribution, I usually start with the CRM records rather than the model. It’s one of the most common breaks in the chain.

A more sophisticated model cannot recover a campaign that was never attached to a person, a person who was never attached to an opportunity, or an opportunity with no clear definition of when it became real. That points to a broken data contract wearing a reporting hat, rather than an attribution problem.

Before you debate linear attribution versus time decay, make the underlying chain trustworthy.

Attribution is a chain, not a dashboard

A useful attribution system needs to connect several events:

Campaign → Campaign Member → Lead or Contact → Opportunity → Opportunity Contact Role → Revenue

Each link answers a different commercial question.

Which campaign or channel created the interaction? Which person interacted with it? Did that person become part of an opportunity? Which opportunity was it? What role did they have in the buying process? Did the opportunity produce revenue?

If one of those links is missing, the system starts guessing. Sometimes the guess appears in a colourful dashboard, which gives it an unfortunate air of authority.

A CRM is often treated as a reporting database. That is too late in the process. It should be an operational system that records ownership, progression and commercial decisions as they happen.

Marketing owns some inputs. Sales owns others. RevOps, if you have it, should govern the rules rather than quietly repair everything in spreadsheets. The important point is that every important field has an owner, a definition and a moment when it must be recorded.

The first failure: source data gets overwritten

B2B buying journeys often involve several touches across search, referrals, events, partner activity, content and direct outreach. A contact may first arrive through a search campaign, return through a webinar, then speak to sales after a referral.

If the CRM stores only one value called `Lead Source`, most of that history disappears. The latest input wins. Or the sales representative selects “Other” because the approved list does not describe what happened. Or a form submission updates the original source with the latest UTM values during a later session.

Now the report can still answer a question. It just might not be the question you meant to ask.

At minimum, preserve first-touch, last-touch and intermediate campaign information as separate, queryable data. Do not flatten the whole journey into one source field and call it attribution. A source taxonomy should use an approved vocabulary, with clear definitions for channels and campaign types. “Partner” needs to mean the same thing in marketing, sales and the board report.

UTM parameters need the same treatment. Store the original values. Preserve later touches separately. Carry the relevant identity and source data through conversion. Use stable identifiers to connect the records instead of relying on a person’s name or an email address that may change.

The work lacks glamour, much like naming database columns. Both become surprisingly important when somebody asks why the pipeline number changed.

The second failure: people are not connected to opportunities

A lead or contact can exist in the CRM without proving that they influenced a deal. A common missing relationship is the Opportunity Contact Role.

Without it, you may know that a person attended a campaign and that an opportunity later appeared in the account. You do not know whether that person was involved in the opportunity, what role they played, or whether the connection is anything more than account-level proximity.

That distinction matters. A campaign member is not automatically an opportunity contact. A contact at an account is not automatically part of every opportunity at that account. Treating those relationships as interchangeable produces neat-looking reports and weak conclusions.

Opportunity Contact Roles are therefore not just a CRM feature for administrators to configure. They are part of the sales process. Someone needs to record them. The organisation needs to define when they become required and what roles mean in practice, such as champion, decision-maker, evaluator or procurement contact.

You do not need a grand buying-committee strategy to fix this. You need a reliable answer to a basic question: who was connected to this opportunity, and how?

If the answer is missing for most open and closed opportunities, no multi-touch model will make your revenue attribution credible. It will distribute credit across incomplete relationships with impressive precision.

The third failure: the pipeline itself has no shared definition

Attribution depends on lifecycle stages, and lifecycle stages are often treated as labels rather than operating rules.

What qualifies as an opportunity? Is it a sales-qualified lead, a completed discovery call, a documented business problem, a forecastable deal or simply a record that someone created because the quarter was looking lonely?

Different teams can use the same stage name for different events. Marketing may count a qualified handoff. Sales may count an active deal. Finance may count only something with a plausible close date. The resulting reports disagree, and everyone blames attribution because it is the most visible part of the mess.

Define each stage in observable terms. Assign an owner. Record when the stage changed. Define the handoff between teams and the evidence required to make the transition. Then enforce those rules in the CRM as far as the system allows.

This is where operational discipline pays off. If a stage requires a problem statement, buying contact and next step, make those fields required at the relevant transition. If an opportunity can be created without an amount, owner or close date, decide whether that is genuinely useful or merely convenient.

Required fields can be annoying. That is their job. A seatbelt is also an inconvenience if your goal is to move quickly, but the complaint becomes less persuasive after the collision.

The aim is to keep the CRM from becoming bureaucratic while stopping important commercial facts from living in private memory, email threads and the one spreadsheet that nobody else can open.

Build the data contract before choosing the model

A data contract is a shared agreement about what data exists, what it means, who owns it and when it must be captured.

For attribution, yours should answer questions such as:

• Which source fields are preserved at first touch and later touches?
• Which campaign fields are standardised, and who governs the vocabulary?
• What identifies a person across forms, marketing systems and the CRM?
• What qualifies as an opportunity?
• Who owns each lifecycle transition?
• When must an Opportunity Contact Role be added?
• How are open pipeline and closed-won revenue reported separately?
• Which fields are required before a record can move forward?

Write these rules down before buying software. Then test them against real records, not an idealised process diagram.

Take a sample of recent opportunities and trace each one backwards. Can you identify the source history? Can you connect the relevant contacts? Can you see when the opportunity became qualified? Can you distinguish an open deal from revenue that actually closed?

If the answer is no, document the exact break. “Attribution is inaccurate” is too vague to fix. “Campaign members are not carried through lead conversion” is an engineering task. “Opportunity Contact Roles are missing at creation and never enforced later” is a sales-process decision. “Stage 2 means three different things” is a governance problem.

Those are useful diagnoses. They tell you where to work.

Model choice comes later

Once the data chain is reliable, model choice becomes a commercial decision rather than a ritual argument.

You can then ask what you want the model to help you decide. Are you allocating budget across channels? Learning which campaigns assist pipeline? Understanding the path to revenue? Evaluating demand creation separately from sales conversion?

Different questions may justify different views. There is no universal model that turns a complicated buying process into objective truth. Attribution is a useful representation of evidence, not a surveillance camera pointed at reality.

The model should also be governed centrally. If every team creates its own campaign naming conventions, stage filters and credit rules, the result is several incompatible systems rather than several perspectives on one system, and the disagreement gets called insight.

A dashboard is the final layer. It should expose decisions and exceptions, not conceal missing inputs behind a percentage with two decimal places.

Takeaways

1. Repair the chain before refining the mathematics. Connect campaigns, people, opportunities, contact roles and revenue.
2. Preserve source history. Keep first-touch, last-touch and intermediate campaign data queryable instead of overwriting it with one lead-source value.
3. Make contact roles part of sales execution. A contact at an account is not automatically a participant in an opportunity.
4. Define stages as events with owners. Record the transition and enforce the evidence required to move forward.
5. Separate pipeline from revenue. An open opportunity and a closed-won deal are different objects with different reporting questions.
6. Choose the attribution model after the foundation works. Otherwise, you are applying sophisticated maths to incomplete records.

Your next step

Take 20 recent opportunities and trace each one through the full chain: campaign, person, opportunity, contact role and revenue status. Record the first point where the evidence disappears or the definition becomes ambiguous.

That list is more valuable than another attribution demo. It tells you whether the next investment belongs in CRM configuration, sales process, data governance or modelling. Only after that should you decide whether you need a more advanced attribution tool at all.