How to Measure B2B Dark Social Without False Precision

TL;DR. To measure B2B dark social, combine self-reported discovery, customer conversations and tagged distribution. Record what each source actually tells you, reconcile differences at the account level, and keep unknowns visible. Use the evidence to decide what to publish and where to distribute it; do not assign exact revenue credit to interactions you did not observe.

What are you trying to measure?

For this framework, dark social means sharing whose path to a buyer you cannot directly observe: a forwarded document, a private group discussion or a recommendation between colleagues.

Define the measurement question before collecting data:

• Where do buyers remember first hearing about you?
• Which material do they describe sharing or receiving?
• What prompted them to start a buying conversation?

These are different questions. A recommendation might explain discovery, while a later article explains why someone booked a meeting.

Treat the goal as understanding those paths well enough to make a decision. Start with the buyers defined in your ideal customer profile, so unrelated enquiries do not drive your conclusions.

How should you collect self-reported discovery?

Add an optional, open-text question to a relevant enquiry or demo form:

How did you first hear about us?

Use neutral helper text: “If you remember, include the person, community or resource.”

Keep the original answer. Alongside it, add a category for reporting, such as colleague recommendation, community, content, search or unknown. Allow multiple categories when the answer describes several steps.

Do not force “a colleague sent an article” into either referral or content alone. Preserve both the recommendation and the resource.

Ask what prompted action separately, preferably during the conversation. Combining discovery and motivation in one question makes the answer harder to interpret.

Record these answers as buyer recollections. Do not present them as a complete history or let them overwrite recorded visits.

What should customer conversations add?

Use conversations to clarify the sequence and the role of the material. Ask:

• What was happening when you started looking for help?
• What did you see or receive before contacting us?
• Did anyone share something that helped you evaluate the problem?
• What prompted you to act?

Start with open questions. Ask about a specific resource only after the buyer mentions it or gives enough detail to identify it.

Separate the buyer’s account from your interpretation. “A colleague forwarded a checklist” supports a reported sharing event. It does not establish who originally found the checklist or whether it caused the purchase.

Record only the detail needed for the decision. There is usually no need to collect private messages or identify people in a group.

Use consistent notes across marketing and sales. The agreement described in MQL to SQL Handoff provides a useful starting point for deciding who records the answer and where it belongs.

How can tagged distribution help?

Give each planned placement a consistent campaign label and tagged link. Keep a distribution log with the resource, placement, intended audience and label used.

For example, distinguish a checklist shared through a partner newsletter from the same checklist shared in a community.

Read a recorded tagged visit narrowly: the visitor arrived using that link. Do not assume it identifies every person or conversation involved before the visit.

If someone forwards the link, its original label does not tell you who forwarded it. If they share the contents without the link, the label supplies no evidence of that exchange.

Use tagged distribution to compare observable responses to placements. Pair it with buyer accounts to investigate what happened beyond those observations.

Evidence	What to record	What to leave unresolved
Discovery answer	What the buyer remembers	The complete sequence
Customer conversation	Reported context and sharing	Unmentioned interactions
Tagged visit	The link used to arrive	Who passed it along
Account outcome	Opportunity or purchase status	Revenue caused by private sharing

How do you reconcile conflicting signals?

Keep different signals separate until you know they describe the same event.

Suppose the recorded enquiry visit came through search, the form names a colleague, and the conversation mentions a forwarded guide. Those observations can fit one sequence: the colleague introduced the business, the guide supported evaluation, and search helped the buyer return.

Use the following review process:

1. Identify the question each signal answers. Discovery, evaluation and enquiry are separate steps.
2. Preserve the original records. Add a synthesis without deleting the underlying answers.
3. Check the account context. Different people may describe different paths into the same purchase.
4. Label the conclusion. Use observed, buyer-reported, inferred or unresolved.

If two records genuinely disagree, ask a neutral follow-up when practical. Otherwise, leave the conflict unresolved.

Avoid a scoring formula that turns several uncertain signals into a precise attribution number. More records can add context without establishing causation.

What would a worked example look like?

Illustrative example: these numbers are invented to demonstrate the method.

A team reviews 20 qualified enquiries:

• 14 include a discovery answer.
• 6 mention a colleague or private community.
• 4 of those 6 include enough detail to identify a resource.
• 3 of those 6 accounts also have a recorded tagged visit.

The groups overlap. They are not separate pools of buyers.

A defensible report would say:

“Among 20 qualified enquiries, 14 provided a discovery answer. Six reported a colleague or private-community introduction. Four of those identified a resource, and three also had a recorded tagged visit. Six enquiries had no discovery answer. These records support further investigation of sharing, but do not establish its revenue contribution.”

If one account later buys, record the purchase beside the evidence. Do not count its full revenue once under recommendations and again under tagged content.

A reasonable next action is to clarify which resources buyers found useful and test another relevant placement. The evidence does not justify claiming an exact return on private sharing.

What should the reporting view include?

Keep the report small enough to audit:

Report item	Reporting rule
Collection coverage	Show enquiries reviewed, answers received and blanks
Reported discovery	Show counts; disclose overlapping categories
Resource mentions	Separate identifiable resources from vague recollections
Tagged responses	Use consistent labels and a stated review window
Account outcomes	Count each account once; retain multiple evidence labels
Uncertainty	List missing answers, conflicts and unsupported inferences

Keep the collection method and qualification criteria stable when comparing periods. If either changes, explain the break before interpreting a difference.

Avoid multiplying respondent answers across everyone who did not answer. Report the sample you have.

For a few important conclusions, add a short evidence note: what supports the conclusion, what remains unknown and what action follows. This is more useful than a confidence score with no stated basis.

How do you avoid the common mistakes?

• Do not rename all direct traffic “dark social”. An absent recorded source does not establish private sharing.
• Do not turn a discovery answer into revenue credit. It describes recollection, not the purchase’s counterfactual outcome.
• Do not double-count overlapping evidence. One account can have several labels and one recorded purchase.
• Do not lead the buyer. Ask what they encountered before suggesting your preferred channel.
• Do not change categories silently. Keep definitions and document revisions.
• Do not optimise for mentions alone. Check whether the evidence concerns the buyers and problems you intend to serve.

When is this approach insufficient?

This framework is insufficient when the decision requires a causal estimate: for example, whether increasing distribution spending creates additional demand.

For that decision, design a separate test with a defined intervention, comparison, outcome and decision rule. Treat discovery answers as supporting context.

It is also insufficient for tracing every private interaction. Leave that question open. Set a narrower objective, such as learning which resources to improve, using the distinction between objectives and measures in OKRs vs KPIs.

Frequently asked questions

Related reading

• How to Build an ICP — define whose discovery paths matter.
• MQL to SQL Handoff — agree how marketing and sales record useful context.
• OKRs vs KPIs — separate the decision you want to improve from the measures you track.