Before You Invest in AI, Redesign the Work
Artificial intelligence is becoming a priority for almost every organisation.
Boards are asking how it will affect strategy. Executives are exploring productivity opportunities. Technology teams are testing new tools, while employees are already using generative AI to complete everyday tasks.
The pressure to move quickly is understandable.
However, many organisations are beginning with the wrong question. They ask:
Where can we apply AI?
A better starting point is:
How should this work be performed, and what role should AI play within the redesigned model?
AI can improve speed, insight and productivity. But it cannot compensate for unclear processes, fragmented data, duplicated work or poorly designed decision-making.
The organisations that achieve the greatest value will redesign the work first, then apply AI where it makes the redesigned model better.
AI Is Not the Starting Point
When new technology becomes available, organisations often begin by searching for use cases. Teams identify manual tasks, select an application and attempt to automate part of the existing process.
This can produce useful improvements. It can also preserve the underlying problems.
A process may contain unnecessary approvals, repeated handovers, duplicate data entry or activities that no longer serve a clear purpose. Applying AI to those steps may make them faster, but it does not make the overall process effective.
The first objective should not be to automate the current workflow.
It should be to understand whether that workflow should exist in its current form.
Start With the Outcome
Successful AI programmes begin with a clearly defined organisational outcome. That outcome may be:
Reducing the time required to serve customers.
Enabling clinicians to spend more time with patients.
Improving the quality or consistency of decisions.
Identifying operational risks earlier.
Increasing the capacity of an existing workforce.
Reducing administrative effort.
Improving access to organisational knowledge.
Strengthening financial or performance insight.
A broad objective such as "improve productivity through AI" is rarely sufficient.
Leaders should be able to describe the specific outcome, who benefits, how performance will improve and how success will be measured.
Without that clarity, AI programmes can become collections of pilots rather than meaningful transformation.
Follow the Work From Beginning to End
Most organisational processes cross multiple teams, technologies and decision points.
A customer request may begin in one channel, move through several systems, require approval from multiple functions and generate reporting or compliance requirements before it is complete.
Individual teams usually see only part of that journey. As a result, one area may optimise its task while creating more work elsewhere.
Before selecting a technology solution, organisations should examine the full workflow:
What triggers the process?
What outcome is being produced?
Which activities add genuine value?
Where are decisions made?
Where does work wait?
Where is information re-entered or checked?
Which exceptions require specialist judgement?
Which controls are essential?
Where do customers, patients or employees experience frustration?
What happens after the process is completed?
This end-to-end view often reveals that the largest opportunities come from removing or redesigning work rather than automating individual tasks.
Simplify Before You Automate
AI should be introduced into the simplest viable process, not the most complicated version of the current one.
That means asking whether activities can be:
Eliminated.
Combined.
Standardised.
Completed earlier.
Shifted to a more appropriate role.
Supported through better information.
Managed through exception rather than routine review.
Delivered through self-service.
Redesigned around the customer or patient journey.
For example, an organisation may use AI to summarise a lengthy application.
A more valuable question may be whether the application requires that much information in the first place, whether existing data could be reused, or whether low-risk applications need the same review as complex ones.
The distinction matters.
Automating an unnecessary process creates faster unnecessary work.
Clarify Where Human Judgement Is Required
Not every decision should be automated.
In many sectors, judgement, empathy, accountability and professional expertise remain essential.
The objective should be to identify where AI can support people and where human oversight must remain central.
AI may be well suited to:
Summarising information.
Identifying patterns and anomalies.
Preparing first drafts.
Classifying requests.
Retrieving organisational knowledge.
Recommending next steps.
Automating routine administrative work.
Highlighting cases requiring attention.
Human judgement may remain essential for:
High-consequence decisions.
Clinical or professional interpretation.
Ethical trade-offs.
Complex exceptions.
Relationship management.
Decisions requiring statutory authority.
Situations where accountability cannot be delegated.
The best operating models are unlikely to be fully automated. They will combine the speed and analytical capability of AI with human judgement, accountability and trust.
Data Quality Determines the Value of AI
AI depends on the information available to it. If organisational data is incomplete, inconsistent, fragmented or poorly governed, AI may produce outputs that appear credible but are unreliable.
Before scaling AI, organisations should assess:
Whether important data is captured.
Whether information is consistent across systems.
Whether definitions are agreed.
Whether data is current and accurate.
Whether access controls are appropriate.
Whether confidential or sensitive information is protected.
Whether the organisation can trace how outputs were produced.
Whether employees know which information can be used with which tools.
Data improvement does not always require a large transformation programme. However, leaders should understand where data limitations affect the reliability and scalability of the proposed use case.
Governance Should Match the Risk
AI governance is sometimes treated as a central policy exercise. Policies matter, but governance also needs to operate within the workflow.
A low-risk tool used to draft internal meeting notes may not require the same controls as a system influencing clinical, financial, employment or regulatory decisions.
Governance should be proportionate to the potential impact.
Key questions include:
Who is accountable for the outcome?
What decisions can the AI system make or influence?
What level of human review is required?
How will errors be detected and corrected?
Can decisions and recommendations be explained?
How will bias and inconsistent performance be assessed?
What information can the system access?
What happens when the tool is unavailable?
How will the organisation monitor performance after implementation?
The purpose of governance should not be to prevent innovation. It should enable responsible adoption by making risk, accountability and controls explicit.
Integration Matters More Than the Demonstration
AI tools often perform well in demonstrations. The harder challenge is integrating them into real work.
A useful tool can still fail to create value when:
Employees must move between multiple systems.
Outputs do not flow into the next step.
Existing processes remain unchanged.
The technology adds another review requirement.
Users do not trust the output.
Responsibilities are unclear.
The tool does not handle exceptions.
Benefits depend on behaviour that has not changed.
The measure of success is not whether the technology works. It is whether the redesigned process works better because of it.
Adoption Requires More Than Training
Employees need to understand how to use AI tools, but training alone is not enough. Adoption depends on whether the new approach is easier, safer and more effective than the old one.
Leaders should explain:
Why the work is changing.
What problem is being addressed.
Which activities will be removed or simplified.
How roles and responsibilities will change.
Where judgement is still required.
How performance will be assessed.
What employees should do when the output appears wrong.
How lessons from implementation will shape future improvements.
AI can create anxiety when employees believe it is being introduced primarily to reduce roles.
A more constructive approach is to demonstrate how it can remove low-value work, strengthen decision-making and allow people to focus on areas where their expertise matters most.
Benefits Must Be Measured and Realised
Many AI initiatives demonstrate technical potential without producing measurable organisational value.
A pilot may save several minutes on a task, but those savings may not translate into increased capacity, lower cost, faster service or better outcomes.
The value case should therefore address:
How much time or cost will be released?
Where will that capacity go?
Will demand absorb the additional capacity?
Do roles or workflows need to change?
Will systems or licences add ongoing cost?
What new controls are required?
How will quality be measured?
When should the initiative stop, change or scale?
Productivity gains are not realised simply because a task becomes faster. They are realised when the operating model changes sufficiently to use the released capacity.
A Practical Framework for AI-Enabled Transformation
Coriolis assesses AI opportunities across seven connected stages.
1. Define the Outcome
What organisational, customer, patient or workforce outcome should improve?
2. Map the Work
How does the process operate from beginning to end, including handovers, decisions, controls and exceptions?
3. Remove Unnecessary Activity
Which steps can be eliminated, simplified, combined or redesigned before technology is introduced?
4. Clarify Human and AI Roles
Where can AI assist or automate, and where must human judgement and accountability remain?
5. Strengthen Data and Controls
Is the information reliable, accessible and appropriately governed for the intended use?
6. Integrate the Technology
How will the AI capability fit into existing systems, roles and workflows without creating additional complexity?
7. Realise the Value
What measurable benefits will be delivered, and what organisational changes are required to capture them?
These stages should be considered together.
A strong technology solution cannot compensate for an unclear outcome, an inefficient workflow or an organisation that is not ready to change.
Where Organisations Should Begin
Organisations do not need to wait for a complete enterprise AI strategy before taking action. They can begin with a small number of well-chosen opportunities where:
The outcome is clear.
The process is understood.
The activity is sufficiently repeatable.
Reliable data is available.
The risk is manageable.
Employees are willing to participate.
Benefits can be measured.
The initiative can inform broader organisational learning.
The objective should not be to generate the greatest number of pilots. It should be to build the capability to redesign work and scale the approaches that produce genuine value.
AI Is an Operating Model Decision
AI is often described as a technology investment. In practice, it is an operating model decision.
It affects how work is organised, how information is used, where decisions are made, which capabilities are required and how value is created.
The organisations that benefit most will not necessarily be those that adopt AI first. They will be those that are clearest about the outcomes they want, most disciplined in redesigning the work and most effective at converting technical capability into practical organisational change.
Before investing in AI, redesign the work.
Then use the technology to make the better model possible.
Coriolis Perspective
The starting point for AI should not be the technology. It should be a clear view of how the work ought to be performed. Redesign first, then decide where automation, data and AI can create measurable value.
