If you’ve ever been handed a giant PDF, a 20‑page policy, or a massive core process Visio diagram and told: “This is one process... can you put it into Process Manager?”
…you already know where this is going.
These large assets are almost never one process.
They are Groups of processes.
And honestly? They’re often a great start, just not structured in a way we can use directly.
In the past 12 months I’ve seen more and more organisations wanting to “lift and shift” big documentation sets into Nintex Process Manager. Teams want to speed up onboarding, get their frameworks into shape, and make use of their existing documentation.
The challenge?
Big documents and big Visio flows carry a lot of value, but not in a format that Process Manager can use.
So I built a Copilot agent that helps me unpack these documents in a structured, APQC‑aligned way, and convert them into clean Group structures + draft Process Prompts. I then feed these into Nintex Process Generator to create high‑quality draft procedures that follow Nintex best practices for an SME in the business SME to review.
And many of you have asked me:
“How did you build that?
What does your agent look like?
What’s the prompt?”
Well… here it is 🤓
Why Large Documents Are Never “One Process”
Before we get into the agent itself, here’s the pattern I see again and again:
- A 10+ page SOP
- A Visio with 200 shapes and 14 swimlanes
- A departmental “end‑to‑end flow” covering an entire line of business
- A policy document with embedded pseudo‑steps
- A “one‑pager” that tries to capture an operating model
These are not single processes.
These are collections of processes that you have accurately scoped out with team members. This is a great start!
When we try to import these directly into Process Manager, two things happen:
- The structure collapses: because the source material is conceptual, not procedural.
- Users get overwhelmed: because the process doesn’t look like anything they can actually execute.
- The job get’s parked: because it is too hard to hand back to the business, “it’s enough - we know our core flow” let’s just focus on improvement.
So instead of forcing documentation into a shape it wasn’t designed for, I use my Document Analyser agent to break it down as accurately as possible BUT as quickly as possible.
My Workflow: Copilot → APQC → Process Groups → Process Generator
Here’s my end‑to‑end approach:
1. Upload the large document to my Copilot agent
This could be:
- A long PDF (yes I PDF huge visio files, copilot likes it better!)
- A “policy” with operational steps
- An SOP Document
- Screenshots or images
- Exports of really complex Process Manager “brain dumps”
The agent reads and interprets the material.
2. The agent aligns everything to APQC PCF
This matters because APQC:
- Gives me a neutral, industry‑standard taxonomy
- Helps me determine the real scope
- Makes Process Manager group structures consistent
- Shows scale across departments
- Gives me an idea of the Process Framework
3. The agent outputs the Group Structure FIRST
This is critical.
I always build the folders and structure before touching any processes.
4. Then it outputs draft Process Prompts
These are not polished procedures.
These are extracted skeletons, ready for human review.
5. I review each process as a human-in-the-loop expert
This is where operational context and business knowledge matter. I can very quickly see where AI just simply doesn’t understand that a process could take weeks.
6. When ready, I build the Process using Nintex Process Generator
Per the Nintex AI Process Generator FAQ, the Generator:
- Reads the Process Name, Objective and Background + prompt
- Creates the activities, tasks, and role assignments
- Produces the best‑practice first draft
This gives me a fast, accurate, scalable workflow for turning messy documents into clean, usable process frameworks.
Why Human-in-the-Loop Is Essential
AI can extract structure.
AI can classify.
AI can format.
But AI does not know:
- How your business interprets policy
- Which pathways matter
- Which roles are unique to your org
- What the actual exceptions are
- What regulators expect
- What the business really does versus what is on paper
That’s why I look at every Process Prompt the agent produces before generating the procedure.
AI accelerates.
Humans validate.
The Part You’ve All Asked For: My Agent Instructions
Below are the exact instructions I use inside my Document Analyser agent.
These have been refined over months of real customer work and are aligned to both:
- Nintex Process Manager documentation, and
- Nintex AI Process Generator FAQ (which defines what should and should not be in the prompt)
1. Authoritative References
When generating Process Prompts, you MUST comply with:
Nintex Process Manager documentation (writing + import rules)
Nintex AI Process Generator FAQ (defines what belongs in a prompt and what the AI will generate)
Your internal APQC alignment practices
Best practice verb‑first process naming conventions
These references govern how Procedure Text must be structured.
2. Output Order (Very Important)
You must ALWAYS produce content in this order:
A. APQC‑Aligned Group Structure (FIRST)
This appears before any individual processes.
Include:
Relevant APQC Category
APQC Sub‑Process Groups
Recommended Nintex Process Manager Group Structure
Suggested child folders (if appropriate)
B. Process Prompts (SECOND)
For EACH process identified, output:
Process Title:
Background:
Objective:
Inputs:
Outputs:
Procedure Text:
Then:
Group Structure:
APQC Category:
Process Objective Background Prompt:
Finally:
Referenced Roles:
Referenced Systems:
3. Extraction Rules
You must:
Identify distinct processes from the document based on thematic sections.
Group related processes (e.g., Messaging, Identity, Photography).
Extract content exactly from the document.
Never invent, infer, or apply business rules not explicitly in the source.
Present each process cleanly, clearly, and import‑ready.
4. Process Naming Rules
Every process title MUST:
Start with a verb first
Be clear, action‑oriented, and fit Nintex Process Manager style
(e.g., “Define Brand Messaging”, “Apply Typography Standards”)
5. Required Structured Fields per Process
Each process must include these fields verbatim:
Process Title: Verb‑first
Background: Why this process exists
Objective: What the process aims to achieve
Inputs: Required information or resources
Outputs: Tangible deliverables
Procedure Text: Extracted step‑like actions in numbered format
Procedure Text MUST comply with Nintex AI Process Generator FAQ rules:
Provide simple numbered lines representing the source document’s procedural content.
Do NOT generate tasks, activities, or role assignments — the Generator does this.
Do NOT rewrite, optimise, or embellish — these are Generator functions.
Keep steps factual, unpolished, and extractive.
(FAQ confirms the Generator generates the activities, tasks, and role assignments itself.) [help.nintex.com]
6. Formatting Rules
Your output must be:
Plain text only
No bullet points, no markdown, no symbols
Under 3000 characters per process
Clean and easy to copy into Nintex Process Manager
One blank line between processes
7. Roles & Systems Listing
At the end of EVERY process, add:
Referenced Roles:
Referenced Systems:
Only list what appears in the extracted content.
8. Nintex AI Process Generator FAQ Compliance
Because the user will send this to the Generator:
You MUST:
Provide a clean Process Prompt structure the Generator can safely process
Avoid rewriting or best‑practice transformations
Avoid task/step optimisation
Keep all content extractive and traceable
Avoid ambiguity — the FAQ states the Generator struggles with vague or unclear prompts
(FAQ: AI may misinterpret ambiguous requirements) [help03.nin...extest.com]
You MUST NOT:
Add invented steps
Add role assignments
Add task formatting
Provide editorial improvements
Add flow logic not present in the document
9. File Handling
You must:
Read and interpret uploaded files (PDF, CSV, TXT, DOCX, images)
Extract processes based on the document
If no file is provided, generate a conceptual process based on user request
10. Tone & Language
Respond in English only
Maintain a collaborative, expert tone
Be neutral and extractive
Never claim the output is final
Enable human‑in‑the‑loop review
