What constitutes a signal?
Ask for a concrete event definition, source category, and what the signal can and cannot indicate. Distinguish raw observations from a modeled category label.
How recent is it?
Ask when activity was observed, processed, and delivered. A new file can contain old observations; require the useful window for your buying cycle.
How frequently is the data refreshed?
Distinguish new observations, eligibility review, removals, and re-delivery. Ask what happens when the source stops producing useful updates.
What is the unit: individual, household, business, or geography?
Confirm the audience unit before assuming an account or neighborhood signal belongs to a specific person. Ask how granularity limits activation.
How is identity resolved?
Request a high-level matching explanation, confidence limits, duplicate treatment, and unresolved-record handling. Do not assume perfect matching or channel eligibility.
What is observed versus inferred?
Ask which fields are direct observations, customer-provided facts, derived attributes, or probabilistic interpretations. Each should retain its meaning and provenance.
How is geography handled?
Confirm source precision, boundaries, coverage gaps, normalization, and service-area exclusions. More observations in an area can reflect better data coverage.
Can the audience be activated?
Verify identifiers, permissions, suppression, minimum requirements, and acceptance for the intended destination before buying the deliverable.
Which channels are supported?
Separate a tested native connection from a file handoff, custom implementation, and roadmap idea. Ask who owns delivery failures and destination rejection.
Which data is first-party versus external?
Ask what you supply, what comes from other sources, and what rights attach to each. First-party status does not automatically permit every subsequent use.
How is suppression handled?
Review opt-outs, existing-customer exclusions, stale signals, duplicates, and how removals reach already activated audiences. Agree refresh responsibilities.
What privacy restrictions apply?
Ask about permitted purposes, sensitive-category limits, source permissions, destination rules, consumer choice, and the customer’s responsibilities. Obtain appropriate review for the actual program.
Can customer outcomes be connected?
Identify the CRM and revenue fields, join keys, reporting period, and unmatched outcomes. A connection supports analysis, not automatic proof of causation.
How is quality measured?
Require definitions, denominators, sample coverage, period, and evidence behind any quality claim. Ask how missing or contradictory information is represented.
What happens when the data is wrong?
Agree how to report issues, correct records, remove unsuitable segments, evaluate a material delivery problem, and resolve it under the program’s terms. Do not assume a refund or remedy without an agreement.