Documentation-First AI Automation
Capture workflows and decisions before automating or querying them
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 98%
Create the information substrate for AI before attempting broad automation. First, document how important workflows actually operate so the organization can distinguish tasks ready for automation from those that require future capability or process redesign. Next, record relevant meetings and store transcripts, decisions, experiments, and updates under a consistent nomenclature. This converts transient conversations and scattered files into searchable unstructured data. AI can then generate follow-ups, identify dependencies, compare experiments, and retrieve prior learning without employees manually reconstructing history from Slack, email, and decks. The method treats documentation as infrastructure: better source material improves every downstream assistant or automation, while missing, disorganized, or stale context limits even strong models.
Origin
Extracted from Marketing Against The Grain when Kip Bodnar described documenting workflows, recording meetings, and organizing transcripts to unlock unstructured data.
Core principles
- 01AI quality depends on accessible organizational context
- 02Document a workflow before deciding how to automate it
- 03Record decisions and experiments as they happen
- 04Use consistent naming so information remains retrievable
- 05Preserve failures and dependencies, not only successful outcomes
How to run it
- 1
Select consequential workflows
Choose workflows where missing context, repetitive coordination, or delayed information regularly creates cost.
Pro tip Start where employees already spend time reconstructing prior decisions.
Watch out Do not attempt to document every activity simultaneously.
- 2
Document the current process
Record triggers, inputs, steps, owners, decisions, exceptions, systems, and outputs for each workflow.
Pro tip Use technical writers or structured interviews to expose tacit knowledge.
Watch out Document what actually happens, not only the official procedure.
- 3
Sequence automation
Classify steps that can be automated now, steps needing human judgment, and steps that may become automatable later.
Pro tip Revisit deferred steps as model capabilities improve.
Watch out Do not automate a broken process merely because it is documented.
- 4
Capture live work
Record appropriate meetings and preserve transcripts, plans, experiments, outcomes, and follow-up decisions.
Pro tip Attach artifacts to the workflow or project they concern.
Watch out Obtain required consent and apply retention and access controls.
- 5
Standardize storage
Create naming conventions and repositories that make related material consistently discoverable.
Pro tip Include dates, project identifiers, artifact types, and owners in metadata.
Watch out A data dump without structure can still produce retrieval failures.
- 6
Activate and maintain the corpus
Connect approved AI tools to the repository, test retrieval tasks, and periodically remove stale or contradictory material.
Pro tip Test with questions whose correct answers are known.
Watch out Do not assume that ingestion guarantees accurate retrieval.
In the wild
A leadership team stores raw planning decks, meeting transcripts, previous experiments, and decision notes in one approved AI project. The team asks which dependencies are missing and what could cause the plan to fail. The model surfaces overlaps and unaccounted-for constraints, after which leaders revise the plan and record the decisions back into the repository.
→ The organization identifies planning risks in minutes and preserves the reasoning for later retrieval.
Common mistakes
Automating undocumented work
Without an accurate process model, automation may omit exceptions, ownership, or critical decisions.
Capturing without nomenclature
Recordings and documents become another information pile if they cannot be associated with a workflow or project.
Saving only successes
Omitting failed experiments prevents future teams and assistants from applying prior learning.
Is it for you?
Best for
Organizations whose workflows, experiments, and decisions are scattered across meetings, decks, documents, email, and chat.
Not ideal for
Teams unwilling or unable to establish lawful recording, retention, access-control, and documentation practices.
From the transcript
“There's two things we've decided recently to do on our team Kieran right which is one we're in the process of having technical writers document…”
“And then the second thing is every meeting is recorded.”
“So you have a library of all previous experiments, which again speaks to the fact the most important thing to do for AI to be…”
From the episode
How a $1B+ Crypto Company Really Uses AI in Marketing