Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

AI in Marketing › Intermediate

Checkpoint: AI for Automation and Analysis

Scenario questions on the second module of the level: where a model belongs in a workflow, what an agent may do on its own, how to treat predictions, how to check an AI-produced figure and how ownership, change control and alarms keep automations right. Answer first, then compare your reasoning with the explanations.
CheckpointIntermediate levelModule 2: AI for automation and analysis4 minUpdated

What this checkpoint covers

The questions apply the guide to AI for marketing automation and the automation governance guide, with two drawing on the guide to AI for analysis and reporting. Each describes something that works on the day it launches and fails quietly later.

Task

  • Write your answer to every question before reading any explanation.
  • Give the decision and the reason, as you would to the person who owns the workflow.
  • Mark each answer right, partly right or wrong, and note what you misjudged.
  • Where an answer changes your design, revise it in the accounting firm report commentary exercise.

Questions

  1. A classifier sorts inbound enquiries into five labels and must always choose one. Complaints keep landing in the nurture sequence. What is missing from the design?
  2. A retailer lets a model write each abandoned-basket email in full at send time, including the discount offered. What would you lock, and how would you keep watch on the rest?
  3. An agent that drafts replies to product questions has had its drafts approved almost unedited for weeks. The team wants it to issue refunds and change lifecycle stages too. What do you advise?
  4. A churn score from the customer data platform (CDP) now triggers an automatic discount for every high-risk subscriber. Why is that risky, and what would you do instead?
  5. A plain-language query tool reports more new customers by channel last month than finance does. What do you check before anyone uses the figure?
  6. An anomaly alert says conversions dropped yesterday. The manager wants to pause the paid campaigns. What comes first?
  7. A lead nurture enrolled nobody for two months and nobody noticed; its builder left in the spring. Which governance controls would have caught it?
  8. The AI vendor releases a new model and the team switches every workflow to it the same afternoon. What should have happened first?

Answers explained

  1. An "unknown" route and a confidence threshold. A classifier forced to pick will pick wrongly with confidence. Low-confidence results go to a person, and labels that escalate, such as complaints, skip nurture altogether, as the guide to AI for marketing automation sets out.
  2. Lock the offer, prices, claims, legal copy and footer. Those are rules, never generated. Let the model vary only survivable lines, check for banned terms, claims and broken links before send, and read a weekly sample, as the automation guide's guardrail table sets out.
  3. Grant permissions per action, not per agent. Refunds and stage changes are acting, not drafting. Treat the new tasks as a new deployment with drafts approved first, and add logging and rate limits before any action runs unattended.
  4. A prediction is not a fact. Paying out on a probability teaches the inputs that looking like a leaver earns money. Use the score to prioritise and time a service check-in, keep a holdout that receives nothing, and review calibration monthly, as the automation guide recommends.
  5. The query, not the table. Confirm the definition of a new customer, the period and time zone, the source table and the Attribution model. The guide to AI for analysis and reporting advises querying the metrics layer so the tool inherits the firm's definitions.
  6. Confirm the flag in the underlying report. A flag is a reason to look, not a finding. Check for a broken tag, a release or a filter change first; the analysis and reporting guide notes most flags are not real changes in the business.
  7. Inventory, ownership and alarms. An inventory row with an owner and deputy, ownership transferred as part of the leaver process, and a volume alarm on enrolments would each have caught it, as would a quarterly audit. The automation governance guide describes all four.
  8. A regression test and a sign-off. Re-run a fixed set of inputs with known good outputs on the new model, have the owner sign off, log the change and keep a way back, under the change control the governance guide describes.
Illustrative example — not a real result

Reading your score

A learner marks themselves out of 8. With 7 or 8 right, they move on to the module on evaluating and rolling out AI tools. With 5 or 6, they reread the sections behind the misses and revise their Tallybrook workflow. With 4 or fewer, they redo the exercise with the module guides open. Misses on questions 3 and 5 matter most, because they let an agent act unchecked or a wrong figure reach a decision.

Where to go next

Report on any workflow you run with one KPI per level, as the automation guide suggests, so that quality and safety are read together. If agents, predictions or verification caught you out, redo the accounting firm report commentary exercise. Otherwise return to the intermediate AI in marketing level and start the final module, whose exercise is evaluating an AI tool for a holiday park operator.

Pages this lesson applies

Terms used on this page

Also related

Related