Which growth tasks should be automated, and which decisions need people?
Elena Verna proposes building repeatable growth foundations into product infrastructure. Hornpiper asks a further question: which work can we automate, with which controls?
Last reviewed:
Hornpiper recommends separating data preparation, interpretation and permission to act in external systems. Human review should reflect the impact of an error and how reversible the work is. This control approach is our application advice.
Verna’s thesis: stop rebuilding the foundations
Verna argues for repeatable growth foundations, such as analytics and experimentation infrastructure, to arrive built into the product. The aim is to free teams from repeatedly setting up the basics so they can focus on strategic and creative work.
Hornpiper’s interpretation: automation does not require unlimited authority
When applying this idea, we distinguish tasks by their risk. Drafting a report and messaging a customer require different permissions. Rather than postponing automation, we recommend separating the work into parts and identifying what can be deployed with appropriate controls.
Example: from campaign review to a budget decision
An agent could combine permitted campaign and CRM records into a draft for review, identifying missing data and unmatched records. An interpretation layer such as Luna could examine possible explanations. Changing a budget is a separate permission; access to produce a report does not automatically authorize that decision.
Illustrative use case, not a specific client installation or a performance claim.
Questions to answer before deployment
- What output is expected, and what does an acceptable example look like?
- What should the task do when data is missing or contradictory?
- Which actions require approval and which may run automatically?
- How will errors be detected, and who owns correction, reversal or stopping?
- How will review effort and error costs be assessed alongside time saved?
These questions do not guarantee results. Controls depend on the data used and the consequences of the work.