Hidden Work and AI Process Discovery

AI shadow processes: finding the work your systems do not show

AI shadow processes are the unofficial workflows, manual workarounds and hidden coordination that sit outside formal systems. Understanding them can reveal process risk, staff pressure and practical opportunities for AI and automation.

Topic: AI shadow processes Focus: Hidden work and process discovery Approach: Understand before automating Control: Privacy and human validation

What are AI shadow processes?

AI shadow processes are unofficial or poorly visible ways of completing work that sit outside the organisation’s documented process, system of record or formal reporting.

They may involve spreadsheets, shared inboxes, copied templates, personal notes, side conversations, manual checks, unofficial approvals or AI tools used to bridge gaps in the formal workflow.

The term also describes the use of AI to discover these hidden process patterns. AI can analyse approved operational material and help identify repeated corrections, clarification requests, handoff friction and workarounds that conventional process data may not show.

The central principle: do not automate the official process until you understand the hidden work that makes it function.

Every organisation has an official process and a lived process

The official process appears in policies, workflow diagrams, CRM stages, ticketing systems, project plans and dashboards.

The lived process includes everything people do to make that official version work. They chase missing information, reconcile inconsistent records, translate unclear policy, fix forms, reformat reports and contact the person who knows how to get the issue resolved.

Much of this work is not logged. Leaders may see a completed task without seeing the manual recovery, judgement and coordination required to complete it.

Official process: the steps the organisation expects people to follow.

Lived process: the real sequence of actions, decisions and workarounds people use to produce the outcome.

Why AI shadow processes develop

Most shadow processes are not created to avoid responsibility. They emerge because staff are trying to meet an operational need that the formal system does not support well.

The formal workflow is too slow People create a quicker path to keep customers, suppliers or internal teams moving.
Systems do not share information Staff copy data between applications or maintain a spreadsheet that reconciles conflicting records.
Decision rights are unclear Approvals happen through chat, email or personal relationships because ownership is not explicit.
Information is difficult to find Teams build unofficial templates, notes and knowledge collections to answer recurring questions.
The documented process is incomplete Experienced people add checks and judgement that were never captured in the procedure.
Performance pressure rewards workarounds People are measured on completing the outcome, not on whether the official process made completion possible.

AI shadow processes are not the same as shadow AI

The two concepts overlap, but they are not identical.

Shadow AI

Shadow AI is the use of unapproved or ungoverned AI tools for organisational work. The primary concern is tool use, information handling, quality and oversight.

AI shadow processes

AI shadow processes are hidden workflows and workarounds that may use AI, or may be discovered and analysed with AI. The primary concern is how work actually happens.

For example, an employee using a personal AI account to draft a report may be shadow AI. The wider process of copying data from three systems, asking for approval in a chat and manually correcting the final document is a shadow process.

Why AI shadow processes matter

Hidden workflows affect cost, quality, resilience and decision-making. They can also determine whether an AI or automation project succeeds.

01

They hide the true effort required

A process may appear efficient because the repeated checking, chasing and correction are not included in formal workload data.

02

They create dependency on key people

Operational knowledge sits with the individuals who understand the workaround rather than in a stable shared process.

03

They distort management information

Dashboards measure system events while missing the work completed before, between or after those events.

04

They make automation risky

Automating the documented path can remove the human checks and recovery actions that currently protect the outcome.

05

They transfer process weakness to staff

Employees absorb complexity through extra coordination, memory and emotional effort that the organisation does not formally recognise.

Hidden work is not automatically waste

Some hidden work compensates for poor process design. Some represents valuable expertise that should be recognised and retained.

A service employee may rewrite a response because they understand the customer’s situation. An administrator may identify a risk that the form does not capture. A project coordinator may know when a stakeholder requires a conversation instead of an automated message.

Work that protects quality and safety
Work that applies professional or local judgement
Work that compensates for a broken process
Work caused by repeated information failure
Work that should become a formal process step
Work suitable for safe automation or AI assistance

The goal is not to remove everything that is invisible. It is to distinguish valuable judgement from avoidable process friction.

How AI can identify shadow processes

Traditional process analysis relies heavily on interviews, observation, workshops and structured system data. These methods remain important, but AI can help analyse larger volumes of approved unstructured material.

Email and request analysis Identify repeated clarification, chasing, missing information and informal routing patterns.
Ticket and case analysis Surface recurring handoffs, returns, exceptions and unresolved root causes.
Meeting and action analysis Find recurring coordination work, repeated decisions and actions that do not enter the system.
Document comparison Detect repeated reformatting, copied templates and inconsistencies across versions.
Spreadsheet discovery Understand what unofficial trackers record that the formal system does not.
Feedback theme analysis Group repeated staff or customer comments that point to a shared process problem.

AI output should be treated as discovery evidence rather than a final conclusion. Patterns need to be validated with the people who understand the workflow and its context.

What AI should look for in shadow processes

The analysis should be designed around operational signals rather than employee surveillance.

Repeated manual correction or re-entry
Common missing-information requests
Work returned to an earlier team or stage
Unofficial approval or escalation pathways
Duplicate reporting and reformatting
Long waiting periods between handoffs
Repeated use of unofficial templates
Tasks completed outside the system of record
Recurring customer or staff confusion
Exceptions that rely on individual knowledge

Analytical principle: look for patterns in the work system, not evidence to blame individuals for adapting to it.

Process mining does not capture every shadow process

Process mining uses event logs to reconstruct and analyse work that occurs inside structured systems. It can be valuable where timestamps, activities and case identifiers are available.

However, many AI shadow processes happen outside those logs. They occur in conversations, shared documents, personal trackers, email threads and human judgement before an official task begins or after it is marked complete.

Event data can show: what happened, where it moved and how long it took.

Unstructured evidence can help explain: why it was delayed, corrected, escalated or handled outside the system.

The strongest discovery approach combines process mapping, staff evidence, system data and carefully governed analysis of relevant unstructured material.

Privacy and trust when analysing AI shadow processes

Email, chat, case notes and internal documents may contain personal information and sensitive organisational context. Analysing them without a clear purpose and transparent controls can damage trust.

The Office of the Privacy Commissioner states that the Privacy Act applies when organisations use AI tools in New Zealand and recommends understanding the system and completing a Privacy Impact Assessment before use.

A controlled AI shadow processes review should define the purpose, approved sources, access, retention, aggregation, human review and how findings will be communicated to staff.

Use only information relevant to the defined process question
Explain what information will and will not be analysed
Prefer aggregated process findings over individual scoring
Limit access to source material and generated outputs
Validate findings with the people doing the work
Set retention, correction and escalation controls

See the Privacy Commissioner’s Artificial Intelligence and the Information Privacy Principles guidance.

Hidden workload and psychosocial risk

Shadow processes often place additional demands on the people who compensate for system weaknesses. The effort may include repeated interruption, emotional labour, conflicting priorities and responsibility without authority.

WorkSafe New Zealand’s psychosocial-risk guidance identifies work design, workload, support and organisational conditions as factors businesses should recognise and manage.

Process discovery should therefore assess not only time and cost, but where hidden coordination and recovery work are creating sustained pressure.

A process is not efficient when its apparent performance depends on invisible employee effort.

See WorkSafe’s guidance on managing psychosocial risks at work.

A practical AI shadow processes discovery method

A focused review should begin with one process where the operational pain is already visible.

01

Select the process and outcome

Choose a workflow with repeated delay, rework, customer friction, manual tracking or staff pressure.

02

Document the official process

Map the expected stages, roles, decisions, systems, inputs and outputs.

03

Collect approved evidence of lived work

Gather relevant interviews, trackers, tickets, messages, notes and process artefacts within defined privacy boundaries.

04

Use AI to identify patterns

Cluster repeated issues, extract recurring manual tasks and identify likely handoff or information failures.

05

Validate with the team

Ask the people doing the work to explain which patterns are accurate, valuable, risky or incomplete.

06

Decide what to remove, formalise or automate

Redesign the process before selecting technology and retain human judgement where it protects the outcome.

Turning AI shadow processes into practical use cases

Hidden work is a strong source of AI opportunities because the operational need already exists. People are currently spending time compensating for it.

Enquiry and request triage Classify incoming work, identify missing information and route it to the correct owner.
Thread and case summarisation Prepare the history and outstanding actions before a handoff or decision.
Information-quality checking Identify incomplete forms, conflicting records or missing attachments before submission.
Knowledge retrieval Help staff locate approved policy, procedure and service information without relying on personal memory.
Exception detection Surface repeated failures and unusual cases that need process or human attention.
Structured task creation Convert approved notes, messages or documents into assigned workflow actions.

These opportunities can be assessed through AI use-case development, supported by process improvement and implemented through workflow automation or AI agents.

Governance should enable safe process discovery

AI analysis should have a defined owner, purpose and decision pathway. Teams need to understand how the findings will be used and how they can correct incomplete interpretations.

New Zealand’s Responsible AI Guidance for the Public Service emphasises safe, transparent and responsible use, clear governance and meaningful human oversight. These principles are also useful for private-sector process discovery.

Name the accountable process and data owners
Define the question before collecting information
Restrict tools and access to approved environments
Separate AI-generated patterns from verified findings
Provide a correction and challenge process
Measure process improvement rather than employee activity

See the New Zealand Government’s Responsible AI Guidance for the Public Service.

Measuring whether the hidden work has reduced

The discovery project should lead to measurable process improvement rather than another report.

MeasurePossible evidenceWhat it demonstrates
Manual handlingRepeated entry, checking, chasing and reformattingWhether avoidable hidden effort is reducing
Handoff qualityWork returned because information or ownership is unclearWhether the redesigned workflow supports completion
Process visibilityImportant actions completed inside the system of recordWhether management information reflects real work
DependencyTasks requiring intervention from particular individualsWhether operational knowledge is becoming shared and resilient
Staff effortTime, interruption and recovery work associated with the processWhether process improvement is reducing hidden workload
Customer outcomeResponse, resolution, repeat contact and complaint patternsWhether internal changes improve the external experience

How Changeable works with AI shadow processes

Changeable helps organisations find the gap between documented process and operational reality, then convert the evidence into practical improvement and implementation decisions.

Current-state process discovery Map the formal workflow and the lived process around it.
Guided stakeholder interviews Capture frontline evidence, workarounds and judgement without treating staff as the problem.
AI-assisted pattern analysis Analyse approved operational material for repeated friction and exceptions.
Process improvement Remove unnecessary work, formalise valuable controls and clarify ownership.
AI and automation design Build purpose-specific tools around validated workflow needs.
Governance and implementation Set privacy, review, accountability, adoption and monitoring controls.

Explore Changeable’s process improvement, AI governance and AI readiness assessment services.

Frequently asked questions about AI shadow processes

What are AI shadow processes?

AI shadow processes are unofficial or poorly visible workflows that sit outside documented systems. They may use AI, or they may be discovered by analysing operational evidence with AI.

How are AI shadow processes different from shadow AI?

Shadow AI concerns unapproved AI tool use. AI shadow processes concern hidden workflows and workarounds, whether or not those workflows currently use AI.

Why do shadow processes develop?

They usually develop because official systems are slow, incomplete, disconnected or poorly aligned with the real work people need to complete.

Can AI find hidden work?

Yes. AI can help identify repeated corrections, clarification requests, handoff friction and unofficial process patterns in approved operational material.

Should every shadow process be removed?

No. Some hidden work represents valuable expertise or quality control. The organisation should distinguish useful judgement from avoidable process friction.

Is analysing shadow processes a privacy risk?

It can be if the purpose, sources, access and use of findings are unclear. A controlled review should use relevant data, transparent governance, privacy assessment and human validation.

Can Changeable help identify AI shadow processes?

Yes. Changeable can map the formal and lived process, capture frontline evidence, analyse approved material, identify use cases and design governed improvements or automation.

About Changeable: Changeable is a New Zealand AI and automation consultancy. We help organisations understand how work actually happens, improve the process and build governed AI systems around real operating needs.

Find the work your systems are not showing you.

Bring us the fragile workflow, repeated workaround or hidden administration problem. We will help identify the real process and determine where improvement, AI or automation can create practical value.