Customer Profiling

A profile is a record with a fixed schema: a set of named attributes, each with an allowed range of values, attached to a person or an account. Building one means deciding the schema first, then filling each field from whichever source is cheapest for it, such as a form for stated attributes, a data provider for firmographics, and product telemetry for usage. Identity resolution stitches those sources onto a single record by matching email domain, company name or a known identifier.

A profile is only as useful as its least populated required field, so completeness governs everything downstream. You raise completeness either by asking, which costs form conversion, or by enriching, which costs money and introduces accuracy risk. Data also decays: people change roles, companies change headcount and tooling, so a record untouched for a year describes someone who may no longer exist in that shape. Wider schemas decay faster, because every extra field is another thing that must be kept current.

Progressive profiling is the usual answer. Ask two or three fields at first contact, then request different fields at each later interaction rather than repeating the same form. A scorecard quiz suits the first pass well, because respondents answer structured questions willingly in exchange for a result, and each answer maps directly onto a schema field instead of arriving as free text that someone must clean. The completed profile then drives routing, list membership and the copy shown afterwards.

Profiles describe attributes, not readiness, so a fully populated record still says nothing about whether the account will buy this quarter. They also flatten committees: one contact profile cannot represent a group in which the user, the budget holder and the security reviewer want different things, which is why account-level and contact-level records are normally kept apart. Self-reported fields carry aspiration too, especially around budget and timeline, so treat those values as claims rather than as facts.

Track fill rate per attribute rather than a single completeness score, because one blank required field can block a routing rule while the average still looks healthy. Alongside it, count how many records fail the rules that consume the profile and note which field caused each failure. That breakdown tells you exactly which attribute to prioritize collecting next.

Then test predictive value. Group closed-won and closed-lost accounts by each attribute and look at which ones separate the two populations. An attribute that appears in similar proportions on both sides is decoration, however complete it is. Recheck periodically, because attributes that separated well when the product was narrower often stop working once the market broadens.

The usual failure is a schema that grows without a consumer. Someone adds a field because it might be useful one day, sales never fills it, and reporting quietly drops every record where it is blank. Before adding an attribute, name the campaign, routing rule or report that will read it, and delete any field nothing has read in a year. A narrow schema kept complete outperforms a wide one that is mostly empty.

The second is asking the same questions repeatedly. A returning visitor who already supplied headcount and industry meets the identical form on the next download, concludes the answers went nowhere, and starts entering junk. Hide fields you already hold, request the next missing one instead, and demonstrate that the profile is being used by changing what the person sees after submitting rather than showing everyone the same page.