What a Workflow Map Looks Like in an AI Engagement

The workflow map is produced before any AI is configured.

It is the document that tells the AI builder exactly what the workflow is; the inputs it needs, the decisions it contains, the human checkpoints that cannot be automated, and the output it should produce.

Without it, the AI is built from assumption. With it, the AI is built from the ground truth of how the work actually runs.

This article shows what a completed workflow map looks like for one of the most common workflows in a $10M–$25M professional services company: the client proposal process.

The format shown here is the Phos AI Labs workflow map standard.

It produces a plain-text document that any AI system owner can read and any future builder can use to reconstruct the workflow.

Any new team member can follow it to run the task manually if the automation ever needs to be paused.

It is not a flowchart. It is a structured plain-text specification; the format that produces the most useful AI configuration and the most maintainable AI system.

The workflow map standard: the five components in context

The workflow map uses a consistent five-component structure for every workflow. Each component serves a distinct purpose in the AI configuration:

The complete map for any workflow is typically one to two pages. It is written in plain English, not in process notation or technical diagram form.

Anyone who can read the map can understand what the workflow is supposed to produce and where humans are required.

A complete workflow map: the client proposal process

WORKFLOW MAP
=============

WORKFLOW NAME: Client proposal — first draft
COMPANY TYPE: Professional services firm ($10M–$25M)
WORKFLOW OWNER: Account manager / senior consultant
VERSION DATE: [Current date]
MAPPED BY: [AI Foundations engagement]

---

TRIGGER
-------
Type: Manual initiation
When: After a discovery call with a qualified prospect has produced a clear brief
Initiating role: Account manager
What the initiator has: Discovery call notes; prospect company information; stated
  requirements and constraints; specific questions or concerns raised in the call
What makes the trigger ready: The account manager has enough information to make a
  recommendation; the discovery call produced a clear picture of the problem; the
  budget range; and the timeline

---

INPUTS
------

STABLE INPUTS (loaded from context pack; do not change between runs):
- Company voice guide and tone standards
- Service descriptions and pricing framework
- Client archetypes (the relevant one for this prospect type)
- Proposal structure standard (section order; format requirements)
- Competitive positioning (how the company describes its differentiation)
- Standard commercial terms (payment terms; IP; confidentiality)

VARIABLE INPUTS (loaded for this specific run; change with each proposal):
- Prospect company name; industry; and company size
- Prospect contact name and role
- Stated problem or challenge (in the prospect's own words from the call)
- Stated constraints (budget range; timeline; internal capacity limitations)
- Specific questions or concerns raised in the discovery call
- Any prior relationship history (referral source; previous conversations)
- The service or engagement type being proposed

CONTEXT INPUTS (the account manager's knowledge; loaded as a brief before running):
- Which client archetype best describes this prospect
- Whether this is a competitive situation or a sole-source conversation
- The account manager's sense of the prospect's primary concern from the call
  (stated concern vs. the concern behind the concern)
- Any specific sensitivities identified in the call

---

DECISION POINTS
---------------

DECISION POINT 1: Which service or engagement scope to propose
QUESTION: Is this a Phase 1 Foundations engagement; a Phase 1 + 2 bundle;
          or a full four-phase proposal?
DETERMINING FACTORS:
  Company size and estimated AI maturity (companies using AI individually;
  no shared systems: Phase 1 only to start)
  Budget signal from the call (explicit budget range; or signals of
  constraint or flexibility)
  Urgency signal from the call (need for quick results favours a phased
  approach starting with Phase 1)
  Competitive situation (sole-source conversations have more room for
  full-scope proposals; competitive situations favour a more conservative
  starting scope)
AI HANDLING: AI uses the context inputs to recommend the scope; account manager
  makes the final call
DEFAULT IF UNCLEAR: Phase 1 proposal with a noted path to Phase 2 and beyond

DECISION POINT 2: How much emphasis to place on differentiation vs. approach
QUESTION: Does this proposal lead with what makes the company different; or with
          the engagement approach and what it produces?
DETERMINING FACTORS:
  Whether the prospect raised the competitive question explicitly in the call
  (if yes: differentiation-forward)
  Whether the prospect seems primarily motivated by risk reduction
  (approach-forward) or by choosing the right partner (differentiation-forward)
  Whether this is a first conversation or a recurring relationship
  (first conversations are more approach-forward)
AI HANDLING: AI defaults to approach-forward for a first proposal; account
  manager adjusts based on call context
DEFAULT: Lead with the specific problem and approach; differentiation in a
  dedicated section

DECISION POINT 3: How to address cost
QUESTION: Does the proposal include a specific cost range; a "from" figure;
          or no cost reference at all?
DETERMINING FACTORS:
  Whether a budget range was discussed in the call (if yes: acknowledge it;
  if the scope fits: confirm; if it doesn't: address the gap directly)
  Whether this is a competitive situation (cost should be referenced at
  least directionally in a competitive context)
  Company pricing guidance: engagement pricing starts at $10,000/month;
  proposals should include "from" language unless a specific scope has
  been agreed
DECISION RULE: Reference investment level in the proposal if a budget was
  discussed or if this is a competitive evaluation; omit or add to a
  follow-up call if the discovery is genuinely exploratory
AI HANDLING: AI uses the pricing guidance and context inputs to determine
  which approach; account manager confirms

DECISION POINT 4: Tone calibration for the specific prospect
QUESTION: How direct; formal; or detailed should this proposal be?
DETERMINING FACTORS:
  Communication preferences from the client archetype (loaded as stable input)
  Any specific signals from the call (e.g. "the COO speaks in bullet points
  and interrupts long explanations")
  Whether the prospect has a technical background or is a pure business operator
AI HANDLING: AI applies the archetype communication preferences as the default;
  account manager adds a note if the specific prospect differs significantly
  from the archetype

---

HUMAN CHECKPOINTS
-----------------

CHECKPOINT 1: Scope and recommendation review (before the AI draft is produced)
WHO: Account manager
WHAT THEY ARE CHECKING: Is the proposed scope and recommendation correct before
  the AI drafts around it?
WHEN: Before the AI runs; the account manager confirms the scope decision and
  provides the context inputs
WHAT HAPPENS IF SKIPPED: The AI drafts based on incomplete context; the output
  will require fundamental revision; not just editing

CHECKPOINT 2: First-draft review and editing (after AI produces the draft)
WHO: Account manager (mandatory) + engagement lead review if proposal value
  above $100K (optional)
WHAT THEY ARE CHECKING: Accuracy of the problem description (does it reflect
  what was actually said?); appropriateness of the scope recommendation;
  tone calibration; cost language; and any specific details the AI could not
  have known
REVIEW STANDARD: The output should be 80%+ usable without significant structural
  changes; editing is expected but rewriting should not be necessary
WHAT HAPPENS IF BELOW STANDARD: The account manager provides specific correction
  notes and re-runs the workflow with additional context inputs

CHECKPOINT 3: Pre-send review (before the proposal leaves the building)
WHO: Account manager (self-review)
WHAT THEY ARE CHECKING: Is this the right proposal to send to this prospect;
  right now? Does it feel specific to their situation?
THE TEST: "If I were this prospect and I received this proposal; would I feel
  that the person who sent it actually understood my situation?"
  If yes: send. If no: identify what is missing and add it before sending.

---

EXPECTED OUTPUTS
----------------

OUTPUT FORMAT: PDF or formatted document (via the company's proposal template)

STRUCTURE (in this order):
1. Opening section (150–200 words): The problem we heard; a reflection of the
   prospect's specific challenge in their words; demonstrating that the discovery
   call was heard; not just conducted
2. What we'd suggest (200–300 words): The specific engagement recommendation;
   framed around what it produces for this company; not a service description;
   a recommendation
3. How it works (300–400 words): A plain-language description of the engagement
   phases relevant to the proposed scope; with specific milestones and what the
   company will have at each stage
4. What makes us the right fit (150–200 words): The two to three specific reasons
   this company is the right choice for this prospect; written honestly; not as
   a sales claim
5. Investment (50–100 words): The investment level and terms; framed appropriately
   based on the cost decision point above
6. Next step (50 words): A single; specific next step; not "let us know if you
   have questions" but "I'll follow up Thursday to confirm this scope fits your
   situation or to discuss adjustments"

TOTAL LENGTH: 900–1,200 words

QUALITY BAR:
A proposal meeting the standard:
  Uses the prospect's specific language from the discovery call at least twice
  Makes one clear recommendation rather than presenting options
  Addresses the primary concern identified in the call explicitly
  Reads as if written for this company; not adapted from a template

A proposal that does not meet the standard:
  Could have been sent to any company in the industry
  Opens with the company's credentials rather than the prospect's situation
  Lists services rather than recommending an engagement
  Uses vague investment language ("competitive pricing") when a specific
  range was discussed

---

KNOWN EDGE CASES
----------------

Edge case 1: The prospect explicitly asked for a cost breakdown before the proposal
Handling: Include a simplified cost table as an appendix; keep the main proposal
  narrative-forward

Edge case 2: The prospect is currently working with a competitor firm and is
  evaluating a switch
Handling: Do not acknowledge the competitor explicitly; address the reasons a
  company would switch (lack of implementation; generic outputs; no shared context)
  without naming the competitor

Edge case 3: The prospect wants a "quick overview" rather than a full proposal
Handling: Produce a one-page summary using the same structure at 250–350 words;
  flag for account manager review that this is a summary version

LAST UPDATED: [Date]
MAINTAINED BY: [AI system owner name / role]

How this map becomes the AI configuration

The workflow map is not produced and then filed. It is read by the AI builder and converted into three specific configuration elements.

Configuration element 1: The context to load (from Stable Inputs and Client Archetypes)

The stable inputs section tells the AI builder what must be in the context pack before this workflow can run at quality.

Before the proposal workflow is deployed, the AI builder verifies that every item in the stable inputs section is in the context pack and is current.

Gaps become Phase 1 work before Phase 2 training includes this workflow. If the client archetypes are not complete, this workflow will not produce archetype-calibrated proposals.

Configuration element 2: The operating rules (from Decision Points)

Each decision point in the map becomes one or more operating rules in the AI configuration.

• Decision Point 1 (which scope to propose) becomes the scope selection rule
• Decision Point 3 (how to address cost) becomes the cost language rule

The AI does not make these decisions independently. It applies the documented rules and flags where the account manager’s input is required before the AI runs; as specified in the “AI Handling” field of each decision point.

Configuration element 3: The output prompt structure (from Expected Outputs)

The output section of the map becomes the output specification in the AI prompt.

• The structure (six sections in order) is written into the prompt
• The length ranges (900–1,200 words total) become length constraints
• The quality bar (“uses the prospect’s specific language from the discovery call at least twice”) is written directly into the prompt as a requirement

The quality bar also becomes the human checkpoint test; the account manager reviews the first draft against the quality bar criteria to determine whether it meets the standard before editing begins.

Common questions on workflow maps

”Does every workflow need a map this detailed?”

No. The detail level in the example above is appropriate for a high-value, high-frequency workflow that will be trained into the team and potentially automated.

For simpler workflows; a support ticket triage, a weekly expense report coding; a shorter map covering the key inputs, one or two decision points, and the output format is sufficient. Match the map depth to the workflow’s value and complexity.

”Who produces the workflow map: the founder or the Phos AI Labs team?”

In a Phos AI Labs engagement: the map is produced through a structured interview with the person who runs the workflow; typically the relevant team lead or the founder for workflows they own.

The person interviewed reviews the draft map and confirms it accurately reflects how the workflow actually runs.

Independent build: the person who runs the workflow maps it using the 20-minute interview structure from the workflow mapping article; with a colleague conducting the interview.

”How do I update the workflow map when the process changes?”

Update the map before changing the AI configuration; not after.

The map is the source of truth. When the process changes; new pricing, new service, new approval structure; the map is updated first.

The AI configuration is then updated to reflect the new map. This order ensures the AI always reflects the current ground truth.

”Can I use this format to map all my workflows before an engagement?”

Yes; and doing so accelerates a Phos AI Labs engagement significantly. The five workflows you map before the engagement begins are the five workflows that move to AI configuration fastest in Phase 1.

The workflow inventory article in this series describes how to identify which workflows to map first.

”What is the difference between a workflow map and a standard operating procedure?”

An SOP documents the official process; what is supposed to happen. A workflow map documents the actual process; what does happen; including the invisible judgment calls, the edge cases, and the human checkpoints that often differ from the official version.

The workflow map is also specifically formatted for AI configuration; the decision points, human checkpoints, and expected output components are not typically in an SOP.

”How many workflow maps does a typical AI Foundations engagement produce?”

Five to eight workflow maps covering the highest-priority workflows across sales, account management, operations, finance, and support.

This is enough to configure the team’s most valuable AI workflows and to give the AI system owner a complete, maintainable workflow library to build from.

Want your workflow maps produced and your AI built from the ground truth of how your business actually operates?

The workflow map is the document that makes AI configuration a build from specification rather than a build from assumption.

The five components; trigger, inputs, decision points, human checkpoints, expected outputs; translate directly into the context to load, the operating rules, the prompt structure, and the review protocol.

Every company that builds this document before touching an AI configuration tool builds an AI that reflects how the work actually happens; not how someone thought it happened when they wrote the prompt.

Path one: map one workflow this week. Choose the highest-value workflow in your business; the one that produces the most important output. Use the interview structure from the workflow mapping article. Produce the five-component map. Load the decision points as operating rules into your Claude Project. The difference in output specificity tells you immediately what the map is doing.

Path two: bring in a partner. If you want the workflow mapping sprint run correctly; producing complete maps for your five to eight highest-priority processes in the first two weeks of a structured engagement; that is the Phase 1 work Phos AI Labs does. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Thirty minutes, no deck. Start here.