The Core Belief
The argument of this essay is simple: a company needs more than processes that work today. It needs the ability to redesign how it achieves its goals when conditions change.
Strategy provides direction. It becomes real through the work of finding customers, exploring opportunities, developing products, delivering on promises, making decisions and handling exceptions. A company advances when it changes that work in ways that produce better outcomes.
A workflow is a practical unit of this change: an organised way of moving from a need to a useful, verifiable result. It can combine human judgement, conventional software and AI agents. It need not be automated, linear or fully specified in advance.
The central hypothesis is that the ability to turn an understanding of what needs to change into a better, working practice can become a competitive advantage. Recognising a problem or building a prototype is not enough. The change must reach everyday use, demonstrate its value and contribute to the next improvement.
This capability can produce two benefits: better work today and a greater ability to change how work is done tomorrow. Neither replaces a product customers want or a sound strategy. Both can help turn those foundations into results.
From Software Features to Meaningful Work
A software-centred view describes capabilities: collect data, search, generate a report, send a message. These capabilities matter, but they do not explain the work they serve.
A workflow connects them to a purpose. What question should the data answer? Which decision should the report support? Who has authority to act? What should happen after the message is sent? APIs, databases, interfaces and agents become means of accomplishing something, rather than ends in themselves.
Workflows exist at different levels and intersect. A customer enquiry belongs to a service process; service affects retention; recurring enquiries may reveal a product problem. People and information move across these boundaries. Improving one part therefore requires understanding its relationship to the whole.
The capability that matters is not producing more workflow diagrams. It is organising work deliberately: defining the result, choosing who or what performs it, establishing authority, making progress visible and keeping the arrangement open to change.
Before arranging a route, ask what the business must be able to do. Delivering a promise also requires resources, preparation, observation and correction, not necessarily in one sequence. A polished route can still lack capacity or authority. Business analysis exposes such gaps without assuming that every problem needs new software.
A Longstanding Principle, Applied to New Possibilities
The underlying idea is not new. It draws on a longstanding tradition of treating the organisation of work as something to understand, question and improve. Two familiar examples illustrate complementary aspects of that tradition.
Toyota’s production system connects execution with visible problems and continuous improvement. Its account of jidoka includes stopping when an abnormality occurs, while kaizen develops people’s ability to improve the work itself. Amazon’s warning against treating process as a proxy for results addresses the other side of the same problem: following a procedure does not justify an outcome that fails the customer. These are useful illustrations, not a complete lineage or proof of a universal business formula. Toyota Production System; Amazon shareholder letter.
A standard is the current version of a way of working, not its final form. It provides a baseline against which a change can be understood. This relationship between standardised work and improvement is also central to lean practice. Lean Enterprise Institute.
For this to work, people must be able to expose failures, friction and unnecessary work without first having to defend themselves. A recurring workaround is worth investigating: it may reveal a broken process or a legitimate situation the process does not cover.
Continuous improvement does not mean continuous reorganisation. Effective practices need stability. Change should justify the cost of implementation, learning and maintenance.
Every Workflow Is a Hypothesis
A workflow embodies an assumption: organising work this way will produce a desired result. Completing every step correctly does not establish that the assumption is sound.
A company therefore needs three kinds of activity: carrying out the work, improving how it is carried out and reconsidering whether it is the right work at all. Sometimes the answer is to remove a delay. Sometimes it is to change responsibility. Sometimes the task no longer serves a useful purpose and should stop.
Creating a workflow, combining existing workflows and eliminating unnecessary work are all legitimate forms of improvement. The objective is progress towards business goals, not a growing inventory of processes.
From an Idea to an Adopted Practice
Consider a hypothetical team producing market research for clients. Its first idea might be to use an agent to write reports faster. Closer examination could reveal that most wasted effort occurs earlier: the client’s question is unclear, and previously verified information is repeatedly rediscovered. The more valuable change may be better intake and reuse of knowledge, not faster writing alone.
That explanation still needs checking. Repeated rewriting could also reflect genuinely changing client needs. The team should inspect actual requests and feedback before diagnosing a defect. Better intake, a changed review arrangement and a new research tool are different candidate changes, with different costs and consequences. Choosing the business arrangement is not the same as choosing its technology.
Recognising this still leaves the work of implementation. Someone must clarify the question. Relevant material must be accessible and checked for freshness. Each participant must know what to receive and pass on. The new approach must be tried on real cases and adjusted where it creates friction.
A change is implemented when everyday work actually changes, not when a diagram or demonstration is complete. It has an owner, the necessary resources and clear handoffs. The previous approach is deliberately replaced, retained for exceptions or kept alongside the new one for a defined reason. Its effects can be observed, and there is a way to correct the approach. Where consequences cannot be reversed, that limitation must shape the decision before the change is made.
The ability to complete this transition is as important as the ability to generate ideas.
People, Software and Agents
The nature of the work should determine the performer. Known rules and exact operations often belong in conventional software. Agents can search, interpret, compare and propose actions within defined limits. People set purpose, resolve consequential ambiguity and remain accountable for decisions requiring their authority.
This is not a ladder towards complete autonomy. A better process may simply give an expert well-prepared options and stronger grounds for a decision. If the expert must repeat the agent’s entire task to trust the result, the division of work is not yet useful.
Uncertain work does not require an invented sequence of predictable steps. Research and product development can have clear questions, access to relevant material, discussion of unfinished work and ways to test conclusions without prescribing the answer. Responsibility and boundaries can be clear while the route to a solution remains open.
Improvement should also be distributed. People doing the work notice problems and propose changes; leaders enable implementation and connect it to wider goals. The right to propose a change is not unrestricted authority to make it. Reversible experiments can happen close to the work, while decisions with substantial consequences require appropriate approval.
What AI Makes Possible
AI can reduce the effort involved in parts of the improvement cycle: analysing material, developing alternatives, producing prototypes and carrying out bounded tasks. More ideas can become practical candidates for testing.
But a cheap prototype is not necessarily a cheap operating process. Integration, quality checks, training, data access and ongoing support still matter. The opportunity is to shorten the whole path from an idea to a useful change, not merely to accelerate its most visible step.
Reusable components can help. A recurring agent task should have understood inputs and outputs, limited permissions, appropriate checks, human handoffs and defined behaviour when something fails. A team introducing an agent into another workflow should not have to recreate its entire operating environment.
These shared mechanisms do not replace understanding the business task. They should emerge from genuine recurring needs, rather than from an attempt to build a universal platform before learning how the work should be done.
Knowing Whether Work Has Improved
Measurement should reveal value, not substitute for it. Counts of agents, reports or completed tasks say little on their own.
Each significant change needs a clear account of what would indicate improvement: a customer receives a suitable solution, end-to-end time falls, rework decreases, total cost declines or risk is reduced. Not all quality fits a single number. Reviewing actual cases and applying informed judgement remain necessary.
The whole affected flow matters. Faster generation can overload reviewers. Lower costs in one department can shift effort into another. Freed capacity, cash savings and increased revenue are different outcomes and should not be confused.
Expected value needs an owner able to influence the operating changes, not just deliver software. Check whether the new practice is used and its expected effects appear. A missing benefit can reveal a wrong assumption, insufficient capacity or an unusable change; it is not automatically a failure of those asked to adopt it.
The capacity to improve also deserves scrutiny. Is it easier to move from a recognised problem to an adopted, tested change? Are mistakes recurring less often? Are useful solutions being reused? Producing diagrams and prototypes faster does not answer these questions.
What Should Accumulate
An improvement should leave behind more than a new procedure. The organisation should retain why it was chosen, what alternatives were tried, where it works and what would justify reconsidering it.
This memory helps avoid repeating failed experiments or copying a successful workflow into unsuitable conditions. Some knowledge becomes code, tools or agent instructions. Some remains in people’s experience and discussion. AI can make this knowledge easier to use, but storing a conclusion does not make it true indefinitely.
A competitor may copy a particular process. A deeper advantage may lie in the repeated ability to recognise what should change, choose a useful intervention, implement it and learn from the result.
This is a vision of business in which goals become working processes, experience becomes improvement, and improvement becomes a lasting organisational capability.
Sources And Perspective
The examples above link to public accounts of Toyota, Amazon and lean practice. Debra Paul, James Cadle and Donald Yeates (editors), Business Analysis, third edition, BCS Learning & Development, 2014, informs whole-business investigation, necessary activities, options and benefit ownership (chapters 1, 4, 6, 8-9 and 14). The vision, research-team example and AI applications are our synthesis, not the book’s cases or a proven formula for success.