🧭 Dojo Compass
Module: Finance, Risk Management and Long-Term Resilience
Focus Area: Technology, AI and Future Readiness
Key Article Point
Artificial intelligence is changing the nature of work.
The first phase of AI adoption was relatively simple.
An employee discovered an AI tool and began using it to perform individual tasks. They might ask AI to draft an email, summarize a document, generate ideas, analyze data or prepare a first draft of a presentation.
The employee remained responsible for the overall workflow. AI was simply another tool.
But this model is rapidly becoming insufficient.
AI is increasingly capable of participating in larger and more complex workflows. A single AI system may help research a problem, analyze information, generate alternatives, prepare documents, monitor progress and recommend next steps. Multiple AI systems may eventually work together, with different systems performing specialized roles.
This creates a new management challenge.
The question is no longer simply:
“Can AI perform this task?”
Increasingly, companies must ask:
“What outcome are we trying to achieve, what workflow is required to achieve it, and how should responsibility be divided among humans and AI?”
This represents an important transition.
Traditional workflow management was largely based on assigning tasks to people.
A manager might say:
- John will prepare the analysis.
- Maria will review it.
- The finance department will approve the budget.
- The operations team will implement the decision.
The emerging model is more complex.
A workflow may instead involve:
- an employee defining the desired outcome;
- AI gathering and organizing information;
- another AI system analyzing alternatives;
- a human reviewing strategic assumptions;
- AI preparing the resulting work product;
- a human approving a high-risk decision; and
- an automated system monitoring execution.
The work is no longer entirely human.
But neither is it entirely autonomous.
It is a managed portfolio of human and artificial intelligence contributions.
For SMEs, this creates both an opportunity and a challenge.
The opportunity is that relatively small organizations may be able to perform work that previously required significantly larger teams.
The challenge is that unmanaged AI adoption can create fragmentation.
Different employees may use different tools, different prompts, different data sources and different levels of human review. The company may know that AI is being used without having a clear understanding of:
- where AI is being used;
- what work AI is performing;
- what level of autonomy AI has;
- what risks exist;
- whether AI is actually improving results;
- what the work costs; and
- whether different AI workflows are reinforcing or duplicating one another.
The emerging management challenge is therefore not simply AI adoption.
It is AI work management.
🎯 Key Challenge
Most companies are still approaching AI as a collection of isolated tools and experiments.
One employee uses AI for writing.
Another uses it for coding.
Another uses it to analyze spreadsheets.
A department creates an internal chatbot.
A manager experiments with an automated workflow.
These activities may all be useful.
But eventually the company faces a larger question:
How do we manage all of these AI-enabled activities as part of an integrated system of work?
Without a management framework, several problems can emerge.
AI may be applied to the wrong work
A company may automate low-value tasks while failing to apply AI to major organizational bottlenecks.
AI may receive inappropriate levels of autonomy
Some low-risk tasks may continue to receive excessive human review, reducing the economic benefits of automation.
At the same time, higher-risk work may receive insufficient oversight.
AI workflows may become invisible
Employees may increasingly rely on AI without the company understanding how important decisions or outputs are being generated.
Performance may not be measured
The company may know that AI is being used but not know whether the workflow is actually:
- faster;
- less expensive;
- more accurate;
- more reliable; or
- more valuable.
Human effort may not be redesigned
Perhaps the greatest risk is that AI simply becomes an additional layer of activity.
Employees continue performing their existing work while also spending time managing AI.
The organization may therefore become AI-assisted but not AI-optimized.
The objective should not be to insert AI into every task.
The objective is to redesign workflows so that:
Humans perform the work where human judgment, creativity, relationships and accountability create the greatest value, while AI performs the work where AI can materially improve speed, scale, consistency or analytical capability.
This requires a new approach to work management.
🥋 Dojo Solution
Treat AI workflows as managed organizational assets.
The central principle is:
Do not manage AI as a collection of tools. Manage it as a portfolio of workflows designed to achieve specific organizational outcomes.
This requires seven emerging principles.
1. Manage outcomes rather than individual tasks
2. Treat AI delegation as a first-tier form of organizational delegation
3. Assign each workflow an appropriate level of AI autonomy and risk
4. Redesign human involvement around value and risk rather than mechanical approval
5. Instrument AI workflows so that performance can be observed
6. Manage AI workflows as a portfolio rather than isolated experiments
7. Create continuous learning loops that improve the workflow over time
Together, these principles create the beginnings of an AI Work Management System.
The system does not need to be technologically complex at first.
An SME could begin with a structured spreadsheet or digital workflow board.
Over time, however, this information could become the foundation for a more comprehensive management platform.
🏗️ Putting It into Practice
Step 1. Start with outcomes, not AI capabilities
One of the most common mistakes in AI adoption is beginning with the technology.
A company asks:
“What can this AI tool do?”
This may lead to experimentation, but it does not necessarily lead to strategic value.
A better starting point is:
“What organizational outcome are we trying to improve?”
For example:
- reduce the time required to prepare client proposals;
- improve the quality of market research;
- respond to customer inquiries more quickly;
- reduce the cost of financial reporting;
- identify operational risks earlier; or
- accelerate product development.
The outcome may require multiple tasks.
Suppose the objective is:
Prepare a high-quality client proposal within 48 hours.
The workflow may involve:
- gathering information about the client;
- reviewing previous interactions;
- analyzing the client’s industry;
- identifying relevant internal experience;
- developing potential solutions;
- preparing a first draft;
- reviewing commercial assumptions;
- revising the proposal; and
- approving the final version.
Once the desired outcome is defined, the company can determine which parts should be performed by:
- humans;
- AI;
- humans assisted by AI; or
- increasingly autonomous AI workflows.
This changes the conversation.
Instead of asking:
“Can AI write proposals?”
the company asks:
“How should the entire proposal-generation workflow be designed to produce a better result?”
That is a much more powerful management question.
Step 2. Elevate AI delegation to a first-tier organizational assignment
Historically, delegation meant delegating work to another person.
A workflow management system might record:
Task: Prepare market analysis.
Delegated to: John.
Deadline: Friday.
Increasingly, companies will need to recognize AI delegation as a first-tier form of delegation.
A future workflow log may instead include:
Outcome: Prepare preliminary market analysis.
Primary executor: AI workflow.
Human owner: John.
Constraints: Use approved sources; identify uncertainty; do not make final recommendations.
Required review: Human review before external use.
This distinction is important.
Delegating work to AI should not mean simply entering a prompt and hoping for the best.
Like human delegation, AI delegation should involve clarity regarding:
- the desired outcome;
- scope;
- constraints;
- available information;
- authority;
- required standards;
- deadlines;
- escalation requirements; and
- review procedures.
In this sense, companies may eventually need to become as good at delegating to AI as they are at delegating to people.
This is likely to become a genuine management capability.
The quality of AI performance will often depend not only on the model itself but also on the quality of the work architecture surrounding it.
Step 3. Assign every AI workflow an autonomy level
Not all AI workflows should be treated equally.
A useful framework is to classify them according to their level of AI involvement.
Level 0. No AI involvement
The work is performed entirely by humans.
This may be appropriate where AI adds little value or where the activity requires a degree of human judgment that cannot reasonably be delegated.
Level 1. AI assistance
The human performs the work while using AI as a supporting tool.
Examples include drafting, brainstorming, research assistance or data organization.
Level 2. AI preparation with human review
AI performs the primary preparatory work.
A human reviews, revises and takes responsibility for the final result.
Level 3. AI execution with human approval
AI carries out the workflow and produces an executable result.
The human’s role is primarily approval rather than detailed re-performance of the work.
Level 4. AI autonomous execution
AI performs the work without routine human approval, operating within defined rules and constraints.
Human involvement occurs through monitoring, auditing or escalation.
Level 5. Multi-agent autonomous workflow
Multiple AI systems perform specialized functions and coordinate to achieve a broader outcome.
For example, one system may gather information, another analyze it, another test assumptions and another monitor execution.
The purpose of this classification is not to create a technological hierarchy.
It is to create an autonomy management framework.
A low-risk workflow may appropriately operate at Level 4.
A high-risk workflow may remain permanently at Level 1 or 2.
The key question is:
What level of AI autonomy is justified by the value created and the risks involved?
Step 4. Replace “human in the loop” with purposeful human involvement
The phrase “human in the loop” is useful but often too vague.
A human can technically be involved without adding meaningful value.
For example, if an employee receives an AI-generated report and spends two minutes approving it without meaningful review, the human is formally in the loop but may not actually be reducing risk.
Conversely, a human may intervene only at specific points where judgment is genuinely valuable.
A better question is:
Where can human involvement materially improve the outcome or reduce meaningful risk?
Human involvement might be particularly valuable when:
- defining strategic objectives;
- establishing constraints;
- evaluating ambiguous information;
- making ethical or commercial judgments;
- managing important relationships;
- approving high-risk actions;
- resolving unusual situations; or
- redesigning the workflow itself.
This leads to a more sophisticated model.
Instead of placing humans mechanically at the beginning or end of every AI process, companies should identify Human Value Points.
For each workflow, ask:
- Where does human judgment create the greatest additional value?
- Where does human review materially reduce risk?
- Where is human involvement merely repetitive?
- What types of situations should automatically trigger human escalation?
This can allow companies to increase AI autonomy without eliminating meaningful human control.
Step 5. Instrument the workflow
An AI workflow should not be evaluated only by its final output.
The company should be able to understand how the workflow is performing.
A basic AI workflow record might include:
| Workflow Element | Example |
|---|---|
| Outcome | Prepare a preliminary client risk analysis |
| AI Autonomy Level | Level 2 |
| Human Owner | Head of Risk |
| Task Steps | Research → analysis → draft → review |
| AI Interventions | 4 |
| Human Interventions | 2 |
| Escalations | 1 |
| Execution Time | 3 hours |
| Direct Cost | $12 |
| Result vs. Target | Delivered within target time |
| Quality Assessment | 8/10 |
| Overall Value | High |
This creates what might be called workflow instrumentation.
Over time, the company can analyze:
- which workflows require excessive intervention;
- where AI produces recurring errors;
- which workflows are becoming faster;
- where costs are increasing;
- which tasks are suitable for greater autonomy; and
- where AI is not producing sufficient value.
This is important because AI workflows are likely to evolve.
A workflow that begins at Level 1 may eventually become reliable enough to move toward Level 2 or 3.
Another workflow may appear promising but consistently require so much human correction that it should be redesigned.
Without measurement, these patterns may remain invisible.
Step 6. Build an AI workflow portfolio
Individual AI workflows should eventually be managed collectively.
The company might have workflows for:
- market research;
- proposal preparation;
- customer support;
- financial analysis;
- internal reporting;
- risk monitoring;
- recruitment support;
- software development; and
- knowledge management.
Viewed individually, each workflow may appear useful.
But portfolio analysis can reveal larger organizational patterns.
For example:
- Are several workflows performing overlapping research?
- Are multiple departments paying for similar capabilities?
- Are there workflows that depend on the same information?
- Could several small workflows be combined into a larger process?
- Is one workflow creating bottlenecks for another?
- Are high-value workflows receiving sufficient resources?
- Are low-value experiments consuming disproportionate management attention?
This introduces a new management concept:
AI workflow portfolio management.
The company is no longer simply managing software licenses or individual AI projects.
It is managing an evolving portfolio of work systems.
Each system competes for:
- data;
- technology resources;
- employee attention;
- financial resources; and
- management oversight.
This portfolio should therefore be reviewed like any other important organizational asset.
Step 7. Create a continuous learning loop
An AI workflow should never be considered permanently finished.
Every workflow produces information about:
- what worked;
- what failed;
- where intervention was required;
- how much the work cost;
- how long it took;
- what risks emerged; and
- how the workflow might be improved.
This information should feed back into workflow redesign.
A simple learning loop might be:
Execute → Measure → Review → Identify Friction → Redesign → Test Again
Suppose an AI workflow prepares customer proposals.
After several months, the company discovers:
- 70% of first drafts require only minor changes;
- 20% require substantial commercial revisions;
- 10% contain recurring errors related to pricing.
The correct response is not necessarily to abandon the workflow.
Instead, the company can ask:
“Why are the pricing errors occurring?”
Perhaps the AI does not have access to current pricing information.
Perhaps pricing rules are unclear.
Perhaps this step requires a different approval process.
The workflow can then be redesigned.
Over time, the organization begins building something more valuable than an individual AI process.
It develops institutional knowledge about how to design and manage AI-assisted work.
That knowledge may become a significant competitive capability.
Step 8. Establish an AI workflow register
SMEs do not need to build a sophisticated AI management platform immediately.
A useful first step is simply to create an AI Workflow Register.
For every significant workflow, record:
- workflow name;
- desired outcome;
- business owner;
- AI systems involved;
- autonomy level;
- human involvement points;
- key risks;
- escalation triggers;
- estimated cost;
- execution time;
- performance indicators; and
- next review date.
Over time, this register becomes a map of the company’s emerging human-AI operating system.
It allows management to see not only where AI is being used but how the organization’s work itself is changing.
📌 Key Takeaways
- AI is expanding from individual task assistance toward participation in increasingly complex workflows.
- Companies should increasingly manage outcomes and workflows, rather than simply assigning isolated tasks.
- Delegating work to AI should become a first-tier form of organizational delegation with defined objectives, constraints, authority and accountability.
- Different workflows require different levels of AI autonomy.
- A useful autonomy framework can range from no AI involvement to multi-agent autonomous workflows.
- “Human in the loop” should be replaced with purposeful human involvement at points where people materially add value or reduce risk.
- AI workflows should be instrumented and measured, including time, cost, interventions, escalations and results.
- Companies should manage AI as a portfolio of interconnected workflows rather than isolated tools or experiments.
- Continuous measurement and redesign can gradually improve workflow speed, quality and cost.
- An AI Workflow Register can provide SMEs with a practical starting point for managing their emerging human-AI operating model.
- The ultimate objective is not to automate as much work as possible. It is to design a better system for combining human and artificial intelligence to create greater organizational value.
🌿 Reflection
The arrival of AI is often discussed as a technological event.
But its deeper impact may be managerial.
For more than a century, companies have developed increasingly sophisticated ways of managing people.
They created:
- organizational charts;
- job descriptions;
- reporting lines;
- KPIs;
- performance reviews;
- project management systems; and
- delegation frameworks.
These systems were designed around a relatively simple assumption:
Work is primarily performed by people.
That assumption is now changing.
The future organization may increasingly consist of work performed by combinations of people and AI systems.
Some AI systems will assist.
Some will prepare.
Some will execute.
Some will monitor.
Some will coordinate other AI systems.
This means that companies will need to answer questions that traditional management systems were never designed to address.
Who owns an AI workflow?
How much autonomy should it have?
When should it escalate?
How should its performance be measured?
When is human review necessary?
How do multiple AI workflows interact?
When should a workflow be expanded, redesigned or discontinued?
These are not merely technical questions.
They are questions about the future architecture of the organization.
The companies that benefit most from AI may therefore not simply be those that adopt the most advanced technology.
They may be those that become best at managing the interaction between human and artificial intelligence.
The competitive advantage may not lie in having access to AI.
Increasingly, many companies will have access to similar models and tools.
The differentiating capability may instead become:
The ability to design better workflows, assign work intelligently, manage autonomy appropriately and continuously learn from the results.
That is why AI work management deserves to be considered an emerging management discipline in its own right.
⚔️ Dojo Mission
Choose one recurring business outcome that currently requires significant employee time.
Do not begin by asking what AI tool you should use.
Instead, complete the following exercise.
1. Define the outcome
What is the result you are trying to achieve?
For example:
“Prepare a high-quality client proposal within 48 hours.”
2. Map the workflow
List every major step currently required to produce that outcome.
3. Assign responsibility
For each step, determine whether it should be performed by:
- a human;
- a human assisted by AI;
- AI with human review;
- AI with human approval; or
- AI autonomously.
4. Identify Human Value Points
Ask where human involvement genuinely:
- improves quality;
- provides necessary judgment;
- reduces meaningful risk; or
- manages an important relationship.
5. Measure the first version
Track:
- time;
- cost;
- number of interventions;
- number of escalations; and
- quality of the result.
Then ask:
“If we redesigned this workflow from the beginning today, knowing that humans and AI could both perform different parts of the work, would we design it the same way?”
The answer may reveal the beginning of a much larger transformation.
The future challenge is not simply learning how to use AI. It is learning how to manage work when intelligence itself becomes an organizational resource that can be delegated, measured, governed and continuously improved.
Leave a Reply