How to Redesign Business Processes Around AI — Instead of Adding AI to Broken Workflows
One of the easiest ways to introduce AI into a business is to take an existing process and add AI to it.
Employees spend hours producing reports.
Add AI to write the reports faster.
Customer Service manually categorises enquiries.
Add AI to categorise them.
Sales teams spend time researching prospects.
Add AI to accelerate the research.
Finance teams manually reconcile information.
Add AI to assist the reconciliation.
These improvements can create genuine value.
But they can also hide a much bigger opportunity.
Because the most important question may not be:
How can AI make this task faster?
It may be:
If we designed this process today with AI available, would we still design it this way at all?
That is the difference between AI task optimisation and AI-enabled process redesign.
And organisations that only add AI to existing workflows risk making inefficient processes faster rather than creating fundamentally better ways of working.
What is AI-enabled process redesign?
AI-enabled process redesign is the deliberate restructuring of a business process around the capabilities of humans, AI, automation and AI agents.
Instead of beginning with the existing process and asking:
Where can we add AI?
it begins with the required business outcome and asks:
What is the best way to achieve this outcome now that AI exists?
That can lead to very different answers.
Some tasks remain human.
Some become AI-assisted.
Some are largely performed by AI.
Some become automated.
Some disappear completely.
And some previously separate steps can become one integrated workflow.
For more on why AI transformation requires more than simply getting employees to use AI tools, read AI Transformation vs AI Adoption: What's the Difference — and Why Does It Matter?
Why adding AI to existing processes can limit value
Most business processes were not designed for AI.
They were designed around:
human limitations;
legacy technology;
organisational structures;
historical controls;
manual hand-offs;
information scarcity.
A process might require five people to move information between systems simply because those systems could not previously communicate intelligently.
AI may change that constraint.
But if the organisation simply inserts AI into one of the five existing steps, the rest of the process remains untouched.
The result may be:
10% faster execution
when a complete redesign could potentially remove several steps entirely.
This creates an important Stratify principle:
Do not automate the process before challenging the process.
The automation trap
Imagine a customer complaint process.
Today it looks like:
Customer submits complaint
↓
Employee reads complaint
↓
Employee categorises it
↓
Employee searches customer history
↓
Employee retrieves relevant policy
↓
Employee drafts response
↓
Manager approves response
↓
Employee sends response
↓
Employee updates CRM
↓
Employee creates follow-up task
Leadership identifies AI.
The first instinct might be:
Use AI to draft the response.
That could save time.
But a redesigned process might look more like:
Customer submits complaint
↓
AI classifies + retrieves customer history + retrieves policy
↓
AI determines whether case falls within approved routine category
↓
Routine case:
AI prepares resolution → human approves → systems update automatically
Exception:
AI prepares case summary → specialist handles
Now the organisation has not merely accelerated one task.
It has redesigned the flow of work.
Start with the outcome, not the workflow
Existing workflows create cognitive anchoring.
People look at the process they already have and ask:
Where could AI fit?
Instead, start with:
What outcome does this process exist to create?
For example:
Not:
Produce a monthly management report.
But:
Give leadership timely insight into business performance and emerging issues.
Now the possibilities change.
Perhaps leadership does not need a 40-page report generated faster.
Perhaps AI should continuously monitor performance, identify exceptions and present management with the five issues requiring attention.
The outcome remained the same.
The process changed completely.
The Stratify AI Process Redesign Framework
I would redesign AI-enabled processes through seven stages:
OUTCOME → MAP → CHALLENGE → ALLOCATE → CONTROL → MEASURE → REDESIGN
This gives us a repeatable methodology.
1. OUTCOME — What are we actually trying to achieve?
Define the purpose of the process.
Examples:
Resolve customer problems.
Convert qualified prospects.
Approve invoices accurately.
Identify financial anomalies.
Onboard employees.
Produce management insight.
Do not begin with individual tasks.
Begin with the business outcome.
Then define how performance is measured today.
For example:
cycle time;
cost per transaction;
conversion;
quality;
error rate;
customer satisfaction;
employee time;
revenue.
That gives us the baseline against which redesign can later be measured.
2. MAP — How does the work happen today?
Map the current process.
Not the process described in a procedure document.
The process employees actually perform.
Capture:
tasks;
decisions;
systems;
data;
approvals;
hand-offs;
waiting;
exceptions;
rework.
This often exposes something important:
work and process are not the same thing.
A ten-minute task may sit inside a three-day process because information waits between teams.
AI might save five minutes from the task while leaving the three-day process almost unchanged.
That is why process-level analysis matters.
3. CHALLENGE — Which work should exist at all?
Now challenge every step.
Ask:
Why does this task exist?
What happens if we remove it?
Is it creating value or compensating for another problem?
Is this approval still necessary?
Is this hand-off required?
Could information flow automatically?
Could the decision happen earlier?
Could the requirement disappear entirely?
This is where transformation begins.
Do not ask only:
Can AI do this task?
Ask:
Should anyone be doing this task?
The Stratify REMOVE Test
For each step, ask whether it should be:
RETAINED
Still necessary.
ENHANCED
Human work improved by AI.
MOVED
Performed earlier/later or by another part of the workflow.
OPTIMISED
Simplified or automated.
VANISHED
No longer required.
That gives us:
RETAIN → ENHANCE → MOVE → OPTIMISE → VANISH
The objective is to stop organisations automating work that should simply disappear.
4. ALLOCATE — Who or what should perform the work?
Once unnecessary work has been removed, allocate the remaining tasks.
We've already developed a useful Stratify model:
HUMAN → ASSISTED → AUGMENTED → AUTOMATED → AGENTIC
HUMAN
Human performs the work.
Best suited where the task requires significant:
judgement;
relationships;
empathy;
accountability;
negotiation;
creative direction.
ASSISTED
Human performs the task with AI support.
Example:
AI retrieves information while the employee makes the decision.
AUGMENTED
AI performs substantial preparation or analysis.
Human concentrates on judgement.
Example:
AI analyses a customer account and prepares a recommended action for the account manager.
AUTOMATED
The task follows sufficiently predictable rules that technology can complete it within defined boundaries.
AGENTIC
AI can coordinate multiple steps, use tools, make permitted decisions and take actions towards an objective.
The objective is not to move everything towards Agentic.
It is to determine the appropriate allocation of work.
For more on how increasing AI authority changes the operating model, read AI Assistants vs AI Agents: What Business Leaders Need to Know Before Giving AI Autonomy.
Human work should become more intentional
AI process redesign should not begin with:
How do we remove people?
A stronger question is:
Where does human involvement create enough value to justify remaining in the process?
Humans may be particularly valuable for:
complex judgement;
relationship management;
novel situations;
ethical decisions;
high-consequence approvals;
negotiation;
empathy;
exceptions.
AI can then absorb more of the:
search;
preparation;
administration;
analysis;
routine execution.
This can create better work as well as faster work.
5. CONTROL — Where are the boundaries?
A redesigned workflow needs controls.
If AI:
recommends;
decides;
communicates;
updates;
transacts,
leadership needs to determine where human authority remains.
For example:
Routine transaction below £100
AI can act.
£100–£1,000
AI prepares; employee approves.
Above £1,000
Manager approval.
The exact thresholds vary.
The principle is:
authority should be designed into the process.
For more on determining who has authority to approve, override or stop AI, read AI Decision Rights: Who Should Be Allowed to Approve, Override or Stop AI?
And where agents receive greater operational authority, read How Much Autonomy Should You Give an AI Agent? A Practical Framework for Business Leaders.
Design the exception path before automating the routine path
This is particularly important.
AI-enabled workflows will encounter situations outside normal conditions.
Leadership should ask:
What counts as an exception?
How does AI recognise it?
Where does it go?
Who owns it?
What information does the human receive?
What happens after the decision?
A poorly designed exception process can destroy the efficiency gained elsewhere.
Employees spend more time fixing unusual AI outputs than the AI saves on routine work.
So:
Design the exception path at the same time as the automated path.
6. MEASURE — Did the process actually improve?
Once the redesigned workflow operates, measure it against the original baseline.
Potential measures include:
cycle time;
cost per outcome;
throughput;
error rate;
customer satisfaction;
employee capacity;
human intervention rate;
exception rate;
rework;
revenue.
Do not measure only:
AI usage.
The objective is process improvement.
For more on separating productivity activity from realised value, read AI Productivity vs AI Value: Why Time Saved Doesn't Automatically Become ROI.
The Stratify Process Value Chain
The economic logic should be visible:
AI CAPABILITY
↓
PROCESS CHANGE
↓
PERFORMANCE CHANGE
↓
CAPACITY / QUALITY / SPEED
↓
BUSINESS OUTCOME
↓
FINANCIAL VALUE
If the process changes but the business outcome does not, leadership needs to understand why.
7. REDESIGN — What have we learned?
The first redesigned workflow is unlikely to be perfect.
AI performance improves.
Employees discover better ways of working.
Exceptions become visible.
Customer behaviour changes.
New technology appears.
So process redesign should become iterative.
Ask:
Which human interventions are still unnecessary?
Where are exceptions occurring?
Where is AI underperforming?
Where could authority safely increase?
Which controls create unnecessary friction?
Where is value leaking?
Then redesign again.
This creates a continuous cycle:
OUTCOME → MAP → CHALLENGE → ALLOCATE → CONTROL → MEASURE → REDESIGN
AI process redesign and change management are inseparable
A redesigned workflow can be technically excellent and still fail.
Employees need to understand:
why the process is changing;
what AI will do;
what they will do;
what skills they need;
how performance expectations change;
where accountability sits.
Managers need to reinforce the new process.
Otherwise employees may continue using the old workflow with AI added on top.
For more on managing this transition, read AI Change Management: How to Prepare People for AI-Enabled Ways of Working.
Managers need to manage the new work, not the old work faster
Suppose AI reduces administration by 30%.
If managers continue evaluating employees using the same expectations as before, the organisation may fail to capture the opportunity.
Management needs to decide:
What should employees do with the capacity?
Should output increase?
Should customer interaction increase?
Should caseload increase?
Should working time move towards higher-value tasks?
Should future recruitment reduce?
AI transformation therefore requires management redesign alongside process redesign.
Process redesign changes roles
When work changes, roles change.
Imagine an analyst role currently contains:
40% data collection;
30% analysis;
20% reporting;
10% stakeholder discussion.
AI may change that to:
10% data collection;
30% AI-supported analysis;
20% exception investigation;
40% stakeholder interpretation and decision support.
The job may still exist.
But the value contribution of the role has changed.
That affects:
skills;
recruitment;
training;
performance management;
career paths.
This is why AI workforce planning cannot be separated from process redesign.
The Process Redesign Canvas
For each target workflow, Stratify could eventually use a standard canvas:
OUTCOME
What should this process achieve?
BASELINE
How does it perform today?
CURRENT PROCESS
What steps exist?
REMOVE
What should disappear?
HUMAN
What remains human?
AI
What should AI assist or augment?
AUTOMATION
What can execute automatically?
AGENTS
What can operate with greater autonomy?
CONTROLS
Where are approvals and limits?
EXCEPTIONS
What needs escalation?
TARGET PERFORMANCE
What should improve?
VALUE
What is the expected economic outcome?
This can become a practical consultancy tool.
How does this connect to the AI operating model?
Process redesign should not happen independently across dozens of teams without coordination.
The organisation needs a mechanism for:
identifying opportunities;
prioritising processes;
assigning owners;
providing technology;
applying governance;
measuring outcomes;
scaling what works.
That is why process redesign sits inside the wider operating model.
For more on connecting these capabilities, read What Is an AI Operating Model? A Practical Guide for Leadership Teams.
Where should organisations start?
Do not necessarily start with the biggest process.
Start where there is:
meaningful value;
repeatable work;
accessible data;
manageable risk;
clear ownership;
enough volume to measure impact.
This gives the organisation an opportunity to learn the redesign methodology before tackling more complex processes.
Ten questions before adding AI to a business process
Before implementing AI, ask:
What business outcome does this process exist to create?
How does the process perform today?
Which steps create genuine value?
Which steps could disappear completely?
Which work should remain human?
Which work should be assisted, augmented, automated or agentic?
Where must human judgement or approval remain?
How will exceptions be handled?
What performance improvement should result?
How will that improvement translate into business value?
If leadership begins with these questions, the AI conversation changes.
From:
Where can we put AI?
to:
How should this work now happen?
Don't make broken processes faster
AI creates enormous opportunities for productivity.
But productivity is not simply about completing existing tasks faster.
The bigger opportunity may come from removing:
tasks;
hand-offs;
waiting;
approvals;
rework;
manual coordination.
That requires organisations to challenge the architecture of work itself.
Because if a process was designed around limitations that AI has now removed, preserving the old process may preserve the old limitations too.
So before asking:
How can AI automate this workflow?
ask:
If we were designing this workflow today, would we build it this way at all?
That is where AI moves from productivity tool to transformation capability.
Frequently Asked Questions About AI Process Redesign
What is AI process redesign?
AI process redesign is the restructuring of business workflows around the capabilities of humans, AI, automation and AI agents rather than simply adding AI to existing tasks.
What is the difference between AI automation and process redesign?
Automation typically performs an existing task with less human intervention. Process redesign questions whether the task, sequence, hand-offs and workflow should exist in their current form at all.
Which business processes should be redesigned for AI?
Good candidates often combine meaningful business value, repeatable work, accessible data, clear ownership, sufficient volume and manageable risk.
Should businesses automate every task AI can perform?
No. Technical capability does not automatically justify automation. Organisations should consider business value, consequence, human judgement, risk and the economics of automation.
How should humans and AI divide work?
Allocate work according to comparative strengths. Humans may remain central for judgement, relationships, empathy and high-consequence decisions, while AI can support analysis, preparation, routine execution and coordination.
How do AI agents change business processes?
AI agents can potentially coordinate multiple workflow steps, use systems, make permitted decisions and take actions, allowing processes to be redesigned beyond individual task automation.
How do you measure AI process transformation?
Compare the redesigned process against a baseline using measures such as cycle time, cost, throughput, quality, error rates, capacity, customer outcomes and financial value.
Who should own AI process redesign?
The relevant business leader should own the business outcome and workflow, supported by Technology, Transformation, HR, Risk and other specialist functions as required.
Is Your Organisation Ready to Redesign Work Around AI?
AI transformation requires more than technology adoption.
Leadership needs to understand which processes should change, how human and AI work should be allocated, where controls remain and how redesigned workflows create measurable value.
Stratify's Executive AI Readiness Assessment helps organisations understand whether the foundations required for scalable AI transformation are in place.
Assess areas including:
Strategy and leadership
Commercial opportunity
People and adoption
Technology and integration
Governance and risk
Test Your AI Readiness
Because the biggest AI opportunity may not be doing today's work faster.
It may be designing a better way to do the work altogether.
Comments