Decision Quality and AI Governance

Minimal viable friction for better AI and business decisions

Minimal viable friction adds the smallest useful pause at the decisions where assumptions, consequences and accountability need to be visible before action becomes difficult to reverse.

Framework: Minimal viable friction Focus: Decision quality Use: AI, automation and transformation Principle: Proportionate control

What is minimal viable friction?

Minimal viable friction is the smallest amount of intentional resistance needed to improve a decision without creating unnecessary delay.

It is designed for decisions that are uncertain, consequential or difficult to reverse. Routine and low-risk work should remain fast. Higher-impact decisions should include a short checkpoint that makes assumptions, consequences and ownership explicit.

The framework is not another approval layer. It is a decision-design method that protects judgement where speed can otherwise hide weak reasoning.

The central principle: keep routine work smooth, but add a deliberate pause where a poor decision would be costly, harmful or difficult to unwind.

Why removing every form of friction creates risk

Organisations have spent years simplifying approvals, automating handoffs and reducing waiting. Much of that work has improved service and productivity.

The problem begins when all friction is treated as waste. Some checkpoints create the moment where a decision owner tests what they believe, considers wider impacts and accepts responsibility for the choice.

When those moments disappear, speed can become a substitute for confidence. A recommendation moves directly into action because the system allows it, not because the reasoning has been tested.

Bad friction: repeated approvals, duplicated evidence and waiting that do not improve the decision.

Useful friction: a proportionate check that exposes assumptions, impact or uncertainty before commitment.

Why minimal viable friction matters for AI

AI can produce summaries, classifications, forecasts and recommendations quickly. Fluency and speed can make an output feel more authoritative than the evidence supports.

Minimal viable friction gives teams a practical way to decide when an AI-supported action should remain smooth and when human judgement must be made visible.

The New Zealand Public Service AI Framework emphasises human oversight, transparency and explainability across the AI lifecycle. These principles are useful beyond government because they connect technology use to accountable decisions.

What purpose is the AI output serving?
What information and assumptions support it?
What could happen if the output is wrong?
Who is authorised to review or override it?
What evidence must be retained?
What condition should trigger escalation?

See the New Zealand Government’s Public Service AI Framework.

Where minimal viable friction belongs

The framework should be triggered by the nature of the decision, not by organisational hierarchy alone.

Difficult to reverse The decision creates a financial, contractual, technical or operating commitment that will be costly to unwind.
Material impact The decision can affect people, customers, service quality, privacy, trust, safety or organisational reputation.
Significant uncertainty The preferred action depends on assumptions, incomplete information or an untested model.
Expanded authority An AI agent or automated system will move from recommending an action to taking it.
New data use The proposal changes how personal, confidential or commercially sensitive information is handled.
Operational scale A pilot, prototype or local workaround is about to become part of normal organisational delivery.

Where friction should be avoided

Minimal viable friction loses value when it is applied to every task.

Routine, repeatable and well-understood work should be easy to complete. Standard reporting, low-risk administration, approved knowledge retrieval and reversible workflow actions should not require a new decision paper.

Over-designed control

  • Every action requires approval
  • The same evidence is entered repeatedly
  • Routine decisions wait for senior leaders
  • Forms document activity rather than reasoning
  • Teams work around the process to deliver

Proportionate control

  • Low-risk work follows approved rules
  • Decision owners act within clear authority
  • Only material exceptions are escalated
  • Evidence is captured once and reused
  • Higher-risk actions receive stronger review

The three minimal viable friction checkpoints

The framework uses three lightweight checkpoints. They can be completed in a short written note, a meeting, a workflow form or an approval screen.

01

Assumption check

State what must be true for the decision to work and identify the assumptions that have not yet been tested.

02

Impact scan

Consider the effects beyond the immediate workflow, including people, customers, systems, incentives, privacy and service quality.

03

Judgement statement

Record why this is the right decision now, who owns it and what evidence would cause the organisation to revisit it.

Checkpoint one: make assumptions visible

Weak decisions often appear strong until their assumptions are stated plainly.

The decision owner does not need to document every possibility. They should identify the assumptions that materially affect value, feasibility, adoption or risk.

What user or customer behaviour are we expecting?
What are we assuming about data quality?
What capacity or skills must be available?
What level of accuracy is required?
What integration or supplier dependency exists?
What would make the proposal fail?

The purpose of the assumption check is not to eliminate uncertainty. It is to stop uncertainty being mistaken for fact.

Checkpoint two: scan the wider impact

A decision can optimise one workflow while making the wider system worse.

An automated triage process may improve speed but create more rework for the receiving team. A self-service tool may reduce internal handling while making support harder for vulnerable customers. A new AI assistant may save drafting time while creating additional review and governance work.

The impact scan checks whether the local gain creates cost, pressure or risk elsewhere.

Which teams receive the downstream work?
Will customers experience a clearer or harder process?
What new monitoring or review work is created?
Could incentives encourage poor behaviour?
Does the decision change privacy or data handling?
What happens when the normal pattern does not apply?

Checkpoint three: record the judgement

The judgement statement creates a concise record of the reasoning at the point of commitment.

It should be short enough to complete and useful enough to guide later review. The objective is not defensive documentation. It is clarity about why the organisation acted.

Why is this the preferred option?
Why is the organisation acting now?
What alternatives were considered?
Who owns the outcome?
What remains uncertain?
What would trigger pause, adjustment or reversal?

Applying minimal viable friction to AI use cases

The level of review should match the intended use, information involved and potential impact.

AI decisionUseful frictionEvidence to retain
Approve an AI use caseConfirm the business problem, users, data, benefit and riskUse-case definition and accountable owner
Move a pilot into productionReview performance, exceptions, adoption and operating ownershipTest results, controls and implementation decision
Use AI output without reviewDefine the permitted task, quality threshold and excluded situationsValidation results and monitoring rules
Give an agent authority to actSet permissions, limits, approval thresholds and recovery stepsAuthority model, logs and escalation path
Process personal informationAssess necessity, privacy, access, retention and disclosurePrivacy assessment and approved controls
Automate a material decisionExamine fairness, explainability, appeal and human accountabilityDecision design and review requirements

Privacy is a trigger for stronger friction

Where AI or automation changes the handling of personal information, the organisation needs a clear purpose and proportionate privacy assessment.

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

This does not mean every low-risk AI task requires the same review. It means personal information, sensitive decisions and new forms of disclosure should trigger stronger evidence and accountability.

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

Minimal viable friction and investment decisions

The same framework can improve technology, transformation and operating-model decisions.

New Zealand Treasury’s Better Business Cases approach gives decision-makers a structured way to consider strategic need, value, viability, affordability and achievability before committing to investment.

Minimal viable friction can sit inside an existing business-case or investment process by focusing attention on the few assumptions and impacts that require explicit judgement at the current stage.

Business case: develops the evidence required for an investment decision.

Minimal viable friction: makes the critical judgement visible at the point where the decision moves forward.

See New Zealand Treasury’s Better Business Cases guidance.

How to embed the framework without creating bureaucracy

Minimal viable friction should be built into existing work rather than added as a separate governance process.

01

Identify the inflection points

Find the moments where decisions become expensive, material or difficult to reverse.

02

Define the trigger

Use simple criteria such as data sensitivity, impact, uncertainty, authority or financial commitment.

03

Assign the decision owner

Name the person responsible for the reasoning, trade-off and outcome.

04

Capture only useful evidence

Keep the artefact short and reuse information that already exists in the workflow.

05

Monitor the decision

Track whether assumptions remain valid and whether the outcome requires adjustment.

06

Remove unnecessary friction

Reduce or eliminate checkpoints that do not change decisions, improve evidence or prevent material harm.

Measuring whether minimal viable friction works

The framework should improve decisions without creating a new source of delay.

Fewer high-impact decisions reversed because assumptions were missed
Clearer ownership of AI and transformation outcomes
Earlier identification of downstream impacts
Better evidence when pilots move into operation
Fewer routine decisions waiting for unnecessary approval
More consistent escalation of material exceptions
Improved ability to explain why an AI-supported decision was made
Shorter review artefacts with stronger decision relevance

How Changeable applies minimal viable friction

Changeable uses the framework to connect AI governance, process improvement and implementation decisions.

AI use-case development Test the business problem, value, assumptions and risks before investment.
AI governance Translate principles into practical triggers, decision rights and human review.
Process improvement Remove wasteful control while preserving the checks that protect quality.
Workflow automation Build approval and escalation points into the operating flow rather than beside it.
AI agent design Define authority, permissions, handoffs and conditions requiring human intervention.
Fractional AI leadership Provide senior guidance for prioritisation, risk and pilot-to-production decisions.

Explore Changeable’s AI use-case development, AI governance, process improvement, workflow automation and AI agents services.

Frequently asked questions about minimal viable friction

What is minimal viable friction?

Minimal viable friction is the smallest intentional pause or control required to improve a decision without adding unnecessary delay or bureaucracy.

Is minimal viable friction another approval process?

No. It should be embedded into existing decisions and used only when uncertainty, impact or irreversibility justify additional judgement.

When should minimal viable friction be used?

Use it for decisions that are difficult to reverse, materially affect people or the organisation, rely on uncertain assumptions or expand automated authority.

When should friction be avoided?

Avoid adding it to routine, reversible and well-understood work where an approved process already manages the risk adequately.

How does minimal viable friction apply to AI?

It helps teams decide when AI outputs, automated decisions or agent actions require assumption checking, human review, evidence and escalation.

What are the three checkpoints?

The three checkpoints are an assumption check, an impact scan and a short judgement statement recording ownership and the conditions for review.

Can Changeable embed minimal viable friction into our workflows?

Yes. Changeable can map decision points, define proportionate triggers and build the checkpoints into AI governance, operational workflows and automation.

About Changeable: Changeable is a New Zealand AI and automation consultancy. We help organisations improve decisions, design proportionate AI governance and build practical systems around real operational needs.

Protect judgement without rebuilding bureaucracy.

Bring us the AI use case, automation decision or governance process. We will help identify where work should remain smooth and where a small, deliberate pause will protect the outcome.