Task-Level AI Automation Audit
Decompose a role into tasks and assess AI suitability one task at a time.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 95%
This audit evaluates AI at the task level rather than asking whether an entire job will disappear. Start with the role's outcomes, then inventory recurring responsibilities and break ambiguous activities into observable units such as classification, segmentation, interviewing, strategy, approval, or copy production. For each unit, assess whether AI can perform it, assist it, or adds little value. Consider task frequency, data availability, error cost, required empathy, accountability, tool maturity, review burden, and exception handling. Map suitable tools only after understanding the task. Pilot promising candidates and measure total cycle time and quality, including human verification. Finally, aggregate validated task findings to describe how the role may change. This reveals a redesigned allocation of work rather than producing an unsupported percentage based on the job title alone.
Origin
Kieran Flanagan demonstrated a prototype that decomposed marketing roles into tasks, assessed automation potential, and recommended AI tools.
Core principles
- 01Jobs are bundles of tasks rather than indivisible units.
- 02AI suitability varies by task and problem space.
- 03Automation estimates should follow task analysis, not precede it.
- 04Human responsibilities remain wherever judgment, empathy, or accountability dominate.
How to run it
- 1
Define role outcomes
State what the role must achieve independently of its current tools or routines.
Pro tip Separate business outcomes from inherited process steps.
Watch out Automating activity that no longer matters creates no value.
- 2
Inventory the tasks
List recurring responsibilities and split broad labels into discrete, observable work units.
Pro tip Include coordination, review, and exception handling.
Watch out Labels such as strategy or content are too broad to score reliably.
- 3
Score AI suitability
Assess each task's structure, data, frequency, error cost, empathy requirement, and need for accountability.
Pro tip Use separate ratings for automation and assistance.
Watch out Technical possibility does not establish operational desirability.
- 4
Map candidate tools
Identify tools whose demonstrated capabilities match the task and required inputs.
Pro tip Demand evidence from the actual workflow, not generic demos.
Watch out Tool recommendations can become obsolete quickly.
- 5
Pilot and measure
Compare the AI-supported workflow with the current one on quality, time, cost, and risk.
Pro tip Include prompting, checking, corrections, and failed runs in the measurement.
Watch out Ignoring review time exaggerates the automation benefit.
- 6
Redesign the role
Reallocate validated automatable tasks while preserving human ownership of judgment and outcomes.
Pro tip Use freed capacity for research, craft, empathy, and decisions.
Watch out Do not infer job elimination directly from isolated task automation.
In the wild
A team breaks product marketing into interview scheduling, transcript classification, segmentation, positioning decisions, copy variants, approvals, and launch coordination. It pilots AI on classification and first-draft variants while retaining human interviews, positioning judgment, and approval.
→ The role changes around verified task savings rather than an arbitrary automation percentage.
Common mistakes
Scoring the job title
A role combines tasks with radically different data, judgment, and risk profiles.
Ignoring verification labor
Automation can take longer when humans must investigate every output.
Starting with a tool
Tool-first analysis encourages teams to automate whatever a vendor demonstrates rather than valuable work.
Is it for you?
Best for
Managers and practitioners deciding where AI can improve a marketing or knowledge-work role.
Not ideal for
Organizations seeking a precise headcount forecast before testing tools in their own workflows.
From the transcript
“a lot of what we do are a series of tasks.”
“And AI is really incredible for automatic some tasks, not others.”
“I think that the change will happen little by little, sector by sector, problem space by problem space.”
From the episode
How To Rank #1 With Google’s Secret Algorithm (Google Leak Explained)