I85PREP
⌘ K
Dotdigital Certification study materials
Dotdigital Certification study resources

Dotdigital Certification: Customer data: contacts, fields, address books and segments

Dotdigital Certification · Work through your exam, one topic at a time.

All guide sections

Customer data: contacts, fields, address books and segments

The Customer data domain tests the model everything else depends on: how contacts, data fields, address books, segments and consent fit together, and how behavioural and transactional data arrive from integrations.

The data model

  • A contact record is keyed on email address; data fields hold standard and custom attributes with defined types such as text, number, date and boolean.
  • Address books are collections a contact can belong to many of - membership records where someone came from, not what they are.
  • Segments are rule-based queries over data fields, address books, engagement and behaviour, evaluated continuously, so a contact qualifies by satisfying the rules rather than by being added manually.
  • Suppression lists block sends regardless of consent, and preference and consent records should be treated as first-class data, not a checkbox.
  • Behavioural data from the tracking script (page views, sessions, abandoned baskets) and transactional data from the ecommerce connector (orders, products, refunds) enrich the contact record.

Worked example

Build "high-value lapsed": total_spend greater than 500 AND last_order_date older than 90 days AND consent_status = subscribed AND NOT in segment "active subscribers".

Reasoning: spend and recency come from order data synced by the connector, so the segment refreshes as new orders arrive. The consent rule keeps the send lawful and self-correcting when someone withdraws. Excluding the active segment prevents overlap with an existing journey, so the contact is not messaged twice in the same week.

Traps that cost marks

  • Confusing address book membership with segment membership; a contact can qualify for a segment without being in the address book you expected.
  • Renaming or retyping a data field, which silently breaks segments, personalization variables and imports.
  • Deleting contacts rather than suppressing them, losing history and reporting continuity.
  • Importing purchased or rented data because it was "opted in" on someone else's list.
  • Assuming behavioural data exists when the tracking script was never placed on the site.

Reporting and analytics: reading metrics honestly

The Reporting and analytics domain tests whether you can interpret campaign, deliverability and revenue reporting correctly, and whether you can tell a real lift from a measurement artefact.

Core metrics

  • Sends, delivered, hard and soft bounces, unique and total opens, unique and total clicks, click-to-open rate, unsubscribes and spam complaints.
  • Hard bounces are permanent and should be suppressed; soft bounces are temporary and should be retried, not deleted.
  • Revenue and conversion reporting attributes orders to a send within an attribution window, using tracking that must actually be enabled.

Reading them honestly

  • Opens are inflated by privacy features that pre-fetch images, so treat opens as a soft signal and use clicks, revenue and replies as the real engagement measure.
  • Deliverability is separate from engagement: inbox placement testing via seed addresses tells you where mail landed, while bounce and complaint rates tell you how the receiving systems reacted.
  • A/B tests need enough sample per variant and a full business cycle before a winner is declared.

Worked example

A campaign reports 62% opens, 0.8% clicks and 12% hard bounces. The wrong reading is "great open rate". The right reading: 12% hard bounces points at list quality and possibly authentication problems, and 0.8% clicks against 62% opens is a large gap that suggests inflated opens or misaligned content. The action is to suppress hard bounces, check authentication and sender reputation, then test subject and offer separately.

Traps that cost marks

  • Comparing click rates across sends with different denominators without saying which one you used.
  • Celebrating open rates as evidence of engagement.
  • Calling an A/B test after a few hours or a few hundred sends.
  • Forgetting that conversion tracking must be configured before the send, not after.
  • Treating a drop in unsubscribes as success when the audience was simply never engaged.
  • Ignoring list growth: shrinking address books quietly erode every future campaign.

AI agents: capability, oversight and guardrails

The AI agents domain tests whether you understand what AI adds to a marketing platform, where it differs from rule-based automation, and what governance it needs before it touches customers.

Automation versus agents

  • A program is deterministic: the same conditions always produce the same path, and every step can be audited.
  • An AI agent is probabilistic: it interprets input, generates or classifies content, and can choose an action. The same prompt can yield different output.
  • Typical platform applications include content generation in the editor, per-contact send time optimisation, predictive engagement or propensity segments, product recommendations, and conversational agents answering customers in chat channels such as web chat or WhatsApp.

Oversight and guardrails

  • Human review before anything publishes: brand voice, factual claims, pricing and regulated language.
  • Consent and privacy: do not paste personal data into tools outside your agreement, and keep the lawful basis for any data the agent reads or writes.
  • Ground agents in approved knowledge and limit the actions they can take, such as looking up an order but not issuing an arbitrary refund.
  • Test against a control group, because perceived improvement is not measured improvement.

Worked example

A team enables per-contact send time optimisation on a weekly newsletter. The correct evaluation compares the optimised group with a holdout sent at the previous fixed time, over several weeks, measuring clicks and revenue per contact rather than opens. Reasoning: send time optimisation shifts when mail arrives, so opens will move for mechanical reasons as well as behavioural ones; without a holdout you cannot separate the two, and if the list has deliverability problems no timing model will fix them.

Traps that cost marks

  • Sending generated copy unedited, including invented offers or claims.
  • Assuming an AI feature repairs poor list quality or authentication failures.
  • Treating a predictive segment as ground truth without validating it against actual outcomes.
  • Giving an agent unrestricted action scope with no escalation path to a human.
  • Using AI output as the only evidence of success, with no holdout or baseline.