BAU vs Pilot Framework
Summary
Before deciding what to automate, classify each task as Pilot (strategically critical to the business model) or BAU (business-as-usual; routine, repeatable). The classification tells you how much human oversight a task needs — not simply whether to automate it.
The two categories
- Pilot — work that is core to the business model’s success. For a course business: the personal brand and the content about the course. Getting it wrong hurts the moat. High stakes, often higher creativity/judgment.
- BAU — work that keeps the lights on. Welcoming a new community member, syncing inventory, posting a scheduled clip. Routine, repeatable, low individual stakes.
The original examples (kept verbatim in spirit)
| Task | Classification | Why |
|---|---|---|
| Course business: personal brand + making content about the course | Pilot | It is the differentiator for that model |
| Updating/welcoming a user entering the community | BAU | Routine, repeatable, low stakes |
| Peptide brand: growing directly on TikTok | Hard / poor channel fit | → restructure into 10 niche knowledge accounts (see Part C1 - Case Study - Peptide TikTok Network) |
The important correction
Pilot does NOT mean "don't automate"
The intuitive read — Pilot = keep human, BAU = automate — is too blunt and will leave huge leverage on the table. The better rule:
- BAU → automate fully autonomously. Let agents run end to end.
- Pilot → automate the production, gate the publication. Agents draft, research, and prepare; a human approves before it ships.
Concretely: even “Pilot” course content can be 80% agent-produced (angles, drafts, edits, scheduling) with a human approval gate before publish. You keep the judgment where it matters and still get the speed. This converts the binary into a spectrum of autonomy levels. Detailed in Part B1 - Automation Decision Matrix.
Reframed as autonomy levels
| Level | Human involvement | Fits |
|---|---|---|
| L0 — Manual | Human does it | New/unproven, highest stakes |
| L1 — Assisted | Agent drafts, human edits & ships | Pilot, creative |
| L2 — Approval-gated | Agent does it all, human clicks approve | Pilot, repeatable |
| L3 — Autonomous | Agent runs end to end, logs for audit | BAU |
| L4 — Self-optimizing | Agent also proposes policy changes (human ratifies) | Mature BAU only |
Worked examples
- Community welcome message → BAU → L3 autonomous.
- Community relationship/answering → Pilot-ish → L1 assisted (keep humans; see Don’t over-automate community).
- TikTok niche content production → repeatable but brand-shaping → L2 approval-gated (the whole Part C1 - Case Study - Peptide TikTok Network pipeline ends in an approval/scheduling gate).
- Inventory→website sync → BAU → L3 autonomous.
- Ad budget change → money movement → L2 at most (never autonomous; see The approval-gate rule).
How to use it
- List the task.
- Ask: is this core to the moat (Pilot) or keeping-lights-on (BAU)?
- Ask: how repeatable / rule-bound is it?
- Assign an autonomy level (L0–L4) using Part B1 - Automation Decision Matrix.
- Record the decision in the decision log.
Related
- Part B1 - Automation Decision Matrix — the scoring tool
- Part A2 - Multi-Agent Orchestration — how the automatable tasks get built