Automating analysis steps
The analyst's own workflow is a process too: reusable prompt templates, the recurring-report playbook, and guardrails that keep 'automated' from meaning 'unwatched'.
Before automating anyone else's process, automate your own. Everything you did in Modules 1–3 — synthesis, analysis planning, memo drafting, chart specs, Q&A simulation — you did with ad-hoc prompts. The step up is turning the ones you'll reuse into templates: a saved prompt with blanks, plus a checklist for verifying the output. A template is a tiny piece of automation with you as the runtime — and it's how one analyst starts operating like three.
The template pattern
INTERVIEW SYNTHESIS v3 (team template - edit the [blanks]) Role: You synthesize BA discovery interviews. Investigation context: [one line]. Input: transcript below. Output: (1) themes with VERBATIM quotes only; (2) claims-to-verify checklist with the data that would check each; (3) interviewee's stated beliefs, labeled as beliefs; (4) open questions. Rules: no theme without a quote; no paraphrase inside quote marks; British spelling stays as spoken. ---VERIFY BEFORE USING OUTPUT--- [ ] spot-check 2 quotes against transcript [ ] beliefs section contains no data claims [ ] checklist items are actually checkable
The attached verification checklist is what makes a template a tool instead of a shortcut. Version them (v3), share them, and improve them when they fail — your team's template library becomes its quality floor.
The recurring report: automation's gateway drug
Every analyst owns at least one recurring report that eats a half-day per cycle. The playbook: standardize the input (same extract, same columns, every cycle — chase the source once, not monthly), template the analysis (this cycle vs. last, changes beyond threshold, one paragraph per notable change, drafted by AI from the pasted extract), keep authorship of the 'so what' (the paragraph of judgment on top is yours; it's also the only part the reader remembers), and date-stamp a review (quarterly: is the report still asking the right question, or just running on inertia?). Analysts who do this reclaim roughly a day a week — which, not coincidentally, is the time Module 5's capstone assumes you have.
The moment a workflow runs on schedule instead of on your attention, it needs what production AI systems need (LLMOps teaches the full version): a quality gate (the verification checklist, actually executed), a fallback (what happens when the input extract is late or malformed — fail loudly, never fill gaps silently), and a named owner (recurring automations without owners become the fragile flags of your team — nobody's job until they're everybody's incident). If your team later graduates to workflow platforms wiring AI steps end-to-end, that's the AI Automation & Workflows course's territory; the guardrail principles come with you unchanged.