A Business Improvement Factory makes improvement a permanent operating capability, so your business can keep getting better while the day-to-day work keeps moving.
It brings together the problems teams encounter, the ideas they have, what customers say and what business data reveals. It connects those inputs to the organisation’s goals, workflows, systems and knowledge, then provides the strategists, developers and AI agents needed to design, build and implement the right response.
Every completed improvement deepens the organisation’s understanding of itself. That makes the next improvement faster, safer and more valuable.
Its promise is simple:
Build a business that gets better at getting better.
Most organisations do not have an ideas problem
Across more than 15 years working in digital transformation and product strategy, I have watched the same scene play out in organisation after organisation.
A group gets together to discuss how the business could work better. Within an hour, people produce 15 or 20 sticky notes describing genuine problems. When the group returns a fortnight or a quarter later, most of the notes are still there.
People usually know where the friction is. They know which report arrives too late, which system requires double entry, which customer complaint keeps recurring and which internal process makes no sense.
Your people know where the business could be better. They lack the capacity to do something about it while keeping the business running.
What they lack is a reliable way to turn that knowledge into implemented change without pulling people away from the work they already need to deliver.
Some improvements require authority the employee does not have. Others need technical skills, time or coordination across teams. Work that could make future delivery better continually loses to the pressure of delivering today.
This is the insight-to-implementation gap:
Problem noticed → idea discussed → insight documented → nothing changes
Producing more insight does not improve a business if implementation capacity stays fixed.
Why giving everybody an AI assistant is not enough
AI makes individuals faster at many tasks. That is useful, but it is not the same as changing how the organisation works.
One recent field experiment gave generative AI tools to 7,137 knowledge workers across 66 firms. Active users spent about two fewer hours on email each week, but the researchers detected no broader change in the quantity or composition of their tasks. Individual access produced individual time savings; it did not automatically redesign the work around them. (NBER, 2025)
That distinction matters because AI adoption is already widespread. The 2026 Stanford AI Index reports that 88% of surveyed organisations used AI in at least one business function in 2025 and 79% regularly used generative AI. Those self-reported figures are directional, but they show how quickly access is becoming normal.
Access will not be the durable advantage.
The advantage will come from how effectively an organisation can:
- identify a worthwhile improvement;
- supply the context required to understand it;
- decide whether it supports the strategy;
- implement it safely;
- put it into real work;
- measure what changes;
- and use the result to choose the next improvement.
Research on the economics of generative AI reaches a similar conclusion. Productivity gains depend on complementary investments such as redesigning business processes, integrating technology into workflows and developing workforce skills, not simply purchasing the tool. (NBER, 2024)
Technology change and organisational change have always arrived together. The technology usually moves faster.
What does a Business Improvement Factory actually do?
A Business Improvement Factory closes the complete loop between recognising that something could be better and producing a measurable result.
It needs five connected capabilities.
1. A shared place to capture what could be better
The inputs are scattered across meetings, business data, customer feedback, emails and people’s heads: problems people encounter, ideas they have, workarounds they repeat and evidence that something should change.
The platform gives people a low-friction way to submit them while the context and motivation are still alive. It can also surface potential improvements proactively by analysing existing work, patterns and feedback.
2. A living model of the organisation
Every request depends on context: goals, workflows, roles, systems, knowledge, data, permissions and previous decisions.
Instead of asking the organisation to explain itself from the beginning for every project, the platform builds a living Business Blueprint. Each improvement makes that Blueprint more complete.
3. A way to prioritise and govern change
The platform connects each potential improvement to strategic outcomes, estimates the likely effort and applies approval rules proportionate to its cost and risk.
It should make small, safe improvements easy without treating every request as a transformation programme.
4. Real implementation capacity
Capturing, scoring and tracking ideas is not enough. Someone or something must perform the difficult final work: edge cases, permissions, testing, security, integration, deployment and refinement.
A Business Improvement Factory includes developers, strategists and specialised agents that can get the change working.
5. A feedback and learning loop
Completion is not the same as improvement.
The platform needs to observe whether the result was adopted, what operational effect it had and whether it contributed to the intended business outcome. What it learns updates the Blueprint and informs the next decision.
How is it different from the software organisations already have?
Many existing categories solve one part of the problem.
| Category | Its primary job | Where the improvement loop can stop |
|---|---|---|
| Idea-management software | Capture, evaluate and track suggestions | The idea is approved but nobody can implement it |
| Project-management software | Coordinate known work | The opportunity still needs to be understood and designed |
| Process-mapping software | Document how work happens | The map describes the problem but does not change it |
| AI assistant | Help an individual think or produce work | The user reaches an 80% solution but cannot make it dependable |
| Automation or agent platform | Provide technical building blocks | Employees may lack the skill, time or organisational authority to deploy them |
| Consulting or strategy engagement | Diagnose and recommend change | The report arrives without ongoing production capacity |
| Traditional development agency | Build a defined solution | The agency repeatedly lacks situational context and requires lengthy discovery |
| Business Improvement Factory | Connect inputs, organisational context, decisions, implementation capacity and learning | The loop is not complete until the change is used and its effect is understood |
The distinction is not that the other tools are unnecessary. The Factory’s software layer can use or connect to them.
Its job is to make sure the organisation does not stop at the boundary between them.
For the building blocks and the commercial alternatives in more detail, see AI loop vs AI agent vs automation, Business Improvement Factory vs AI automation agency and developer vs software agency vs Business Improvement Factory.
AI has changed the minimum economically viable improvement
AI has not merely made software cheaper. It has changed the minimum size of an improvement worth implementing.
In the old model, every dashboard, internal tool or integration competed for scarce specialist capacity. Even a straightforward change carried the overhead of meetings, briefing, prioritisation, development, testing and handover.
That forced organisations to favour a small number of large initiatives. Minor customer frustrations and internal paper cuts accumulated because none was individually large enough to justify the machinery required to fix it.
AI can now assist across research, analysis, specification, design, coding, configuration, testing, documentation and project coordination. Expertise is still required, but each expert can ship considerably more.
The evidence is not uniform across every kind of technical work. In a controlled GitHub experiment, developers completed a defined coding task 55% faster with Copilot. In a different 2025 randomised trial, experienced open-source developers working in their own mature repositories took 19% longer with the AI tools being tested. (GitHub; METR)
The useful conclusion is not that coding is universally free or fast. It is that task shape, context, tools and implementation method determine whether AI creates leverage.
That is precisely why a Factory matters.
A directional review of close to 3,000 requests handled by Busy Work over a little more than five months reinforces the long-tail opportunity. Around 60% required less than $200 of build effort and a further roughly 20% required between $200 and $1,000. Work ready to use within five days was around three times more likely to reach production than work that took longer.
These are approximate operating figures, not a controlled academic study. But they suggest that much of the improvement organisations need is smaller than their existing project and procurement systems are designed to handle.
Why AI-literate employees still get stuck
Someone who understands their work and knows how to collaborate with AI can often create an impressive first version. Shaping the idea, describing the process and proving the central capability can feel quick and invigorating.
Then the final implementation work begins.
There are permissions to constrain, edge cases to resolve, error states to handle, integrations to finish, security questions to answer, tests to run and a reliable definition of done to enforce. The apparent final 20% can consume more time than the employee has available. It can also require expertise they should not be expected to develop.
AI has democratised prototyping faster than it has democratised dependable implementation.
The Business Improvement Factory creates a useful division of labour. The employee contributes the problem, intent and process knowledge. The organisational specification supplies shared context and constraints. A specialist build team uses its own AI-enabled production tools to complete the difficult implementation work efficiently.
Two kinds of improvement traditional projects miss
Perishable opportunities
In one organisation, somebody proposed analysing the last actions customers took before becoming dormant in a product.
The idea was recognised as valuable, but it had no project or owner. It survived through repeated suggestions and hallway conversations until a senior leader eventually took it directly to the data team.
The analysis found four common exit points where the organisation could potentially intervene. By then, months had passed, and the team still lacked development capacity to test the interventions.
The value of the original insight was perishable. It depended on moving from question to analysis to experiment while the problem remained strategically alive.
Inside a Business Improvement Factory, the same opportunity could be captured immediately, enriched with product and organisational context, analysed, reviewed and turned into small tests within a much shorter cycle.
The metric is not merely time to build a dashboard. It is:
Time from management question to informed action.
Compounding friction
Other problems do not expire. Their cost grows every time the work repeats.
An HR employee I worked with used two legacy systems that had no APIs and were both expected to be replaced. For two days each month, that person manually moved information between them. A conventional integration was quoted at around $15,000, which was too much for a connection between temporary systems.
We built a browser-using agent in an afternoon. It signed into the existing systems through the employee’s accounts, collected the required information from one and entered it into the other. That avoided a new system-level integration, but it did not eliminate the need for controls: the agent was constrained to the employee’s existing access and its early results were checked by that employee.
The appropriate solution was not a permanent integration. It was transitional automation: a lightweight bridge whose cost and lifespan matched the problem.
Do not build a permanent bridge between systems you already intend to replace.
The agent did not run unsupervised from day one. It was onboarded progressively, much like a new employee. The person demonstrated the process, observed early runs, handled exceptions and helped refine the instructions.
The most interesting result was not the initial time saving. Within a week, that employee identified another five processes where a similar approach might help. One small implementation changed what they believed was possible.
Agents should be onboarded, not switched on
Giving an agent access to business systems without context, supervision or a definition of done is not responsible delegation.
A useful progression is:
- Demonstrate: show how the process is performed, often by recording a task demonstration.
- Document: turn the demonstration into steps, rules, inputs and a definition of done.
- Supervise: let the agent perform the work while a person checks the result.
- Capture exceptions: add examples, checks and guardrails as unfamiliar situations appear.
- Delegate progressively: reduce routine oversight as the agent proves reliable.
- Improve continuously: use failures and exceptions to strengthen both the agent and the process.
This is also why routine approval is not the same as good governance: see AI loop vs human-in-the-loop.
Research from the OECD’s workplace AI case studies found that workers often made important contributions to implementation because developers needed their understanding of the actual work. Worker involvement also helped build acceptance and trust. (OECD)
This is not just change management around the edges. The employee holds essential process knowledge.
What is an organisational specification?
The Factory needs enough context to act coherently, but it should not spend months attempting to document the whole organisation before producing value.
It should build a progressive organisational specification.
The first version captures four things:
- Outcomes: what the organisation, team or individual is trying to achieve and how progress is measured.
- Systems: which tools and platforms are used to perform the work.
- Knowledge and data: where the required information lives, including documents, databases, messages and people’s heads.
- Loops and workflows: how work starts, moves between people and systems, creates an output and generates feedback.
Each implementation then adds more detail: owners, triggers, decision rules, permissions, exceptions, dependencies, quality standards and observed results.
There is no minimum organisational model that must be completed before improvement begins.
There is only a minimum amount of context required for the next safe, useful change.
What happens after an employee submits an improvement?
The complete Factory cycle looks like this:
- Capture: the employee describes the opportunity in ordinary language.
- Compare: the platform checks completed, active and proposed work for duplication and reusable components.
- Contextualise: it consults the Blueprint and locates relevant context in approved systems such as email, the CRM, knowledge stores or a codebase.
- Clarify: it creates the strongest specification it can, then asks the employee to resolve uncertainty and confirm assumptions.
- Estimate: it assesses likely size, complexity and risk.
- Authorise: it automatically approves work inside delegated cost and risk boundaries or routes it to the appropriate decision-maker.
- Implement: a developer or specialist agent completes the build, returning to the employee when real judgment is required.
- Test: the requester checks the result and flags missing requirements, defects or edge cases.
- Learn: the platform records the implementation, observed effect and new organisational knowledge.
The employee is not expected to write a technical brief containing information the organisation already knows.
The platform assembles the context, then asks only for what remains unclear.
How should improvements be prioritised?
Strategic alignment is the entry condition. The first question is not merely whether an idea sounds useful, but how it contributes to the outcome the organisation is pursuing.
After that, the Factory considers expected value relative to implementation and maintenance effort. Large opportunities are broken into smaller changes capable of producing value or learning sooner.
The person closest to the work should usually make the final judgment within the strategy, budget and risk boundaries established by the organisation.
Set direction centrally. Make improvement decisions as close to the work as possible.
Opportunities that cannot currently be implemented should be retained and grouped by their shared constraints. Ten requests might reveal that one data foundation, permission change or platform capability would unlock them all.
That creates demand-backed infrastructure:
Capture demand → implement what is possible → identify common constraints → fund foundations with known value
Instead of buying a large platform and then searching for use cases, the organisation gathers evidence for the foundational investments it genuinely needs.
How do you govern improvement without slowing it down?
The organisation should own strategic direction, operating structure, approved foundations and risk boundaries. Employees should be encouraged to identify friction and improve the work closest to them.
The governing principle is:
Freedom at the edge requires coherence at the core.
That does not mean every employee can deploy anything they can generate. It means the Factory applies organisational standards during specification, approval and implementation so governance feels like enabling infrastructure rather than a maze of gates.
A useful rule is:
Zero friction to suggest. Proportionate friction to deploy.
A reversible personal workflow using non-sensitive information may require almost no review. An agent that changes financial records, accesses sensitive data or communicates with every customer needs considerably stronger testing and approval, regardless of how cheap it is to build.
How do you know the business actually improved?
“Done” means the requested thing works. “Improved” means it contributed to a better business outcome.
Each improvement should begin with a simple hypothesis:
We believe changing X for these people will affect this behaviour, contributing to this business outcome. We expect to see this early signal within this period.
Measurement then operates at four levels:
- Business outcome: retention, revenue, margin, customer experience, employee capacity, delivery speed or risk.
- Operational effect: the behaviour or workflow expected to change.
- Adoption and quality signals: usage, successful completion, overrides, exceptions and abandonment.
- Delivery economics: implementation time, build cost, running cost, maintenance and reuse.
Business outcomes tell the organisation whether it is winning. The other signals help it understand whether a particular improvement is contributing.
The number of improvements shipped is not itself success. Improvement velocity matters because it shortens the learning cycle, not because more software automatically creates a better business.
Can you build a Business Improvement Factory internally?
Building one does not necessarily mean developing every piece of software yourself.
The organisation must own:
- its goals and strategic choices;
- its operating model and decision rights;
- its understanding of the work;
- its standards and risk appetite;
- and the outcomes it wants to create.
It can use an external platform and implementation partner for the technical system, specialist capability and production capacity.
The employee should not be forced to complete the difficult final 20% alone. The partner should not be forced to rediscover the organisation through hundreds of meetings for every request.
The Business Blueprint connects those two sides: internal context and external implementation capability.
What should happen in the first 30 days?
The first month should establish three things: belief, direction and momentum.
Belief
People understand that improvement is part of the work, feel safe contributing their knowledge and see agents as capabilities they can progressively onboard.
Direction
The sponsor defines the mandate, intended outcomes and boundaries. The initial Blueprint captures the most relevant tools, knowledge sources and loops.
Momentum
The team experiences at least one complete cycle from noticing friction to using a working change. Early wins do not need to be enormous. They need to prove that worthwhile ideas now have somewhere to go.
The first 30 days are successful when people understand what they are optimising for, believe they have permission to improve it and have experienced proof that good ideas can become working changes.
When will a Business Improvement Factory fail?
It is not suitable everywhere.
It needs a sponsor with authority over the starting scope. If a department head has permission to optimise one department, begin there rather than pretending the initiative is organisation-wide.
It needs psychological safety. People will not teach an organisation how their work operates if they believe success means removing them from it or monitoring them more closely.
It needs direction, or leaders willing to make directional choices. The Factory can expose strategic ambiguity, but it cannot substitute for decisions leadership refuses to make.
It also needs a technically viable path to action. Legacy systems are not necessarily a barrier, but if the organisation will not permit approved AI tools, data access, testing or any form of small change, the implementation bottleneck remains.
A useful readiness test is:
- Is there a sponsor with authority to improve the chosen scope?
- Is the participating team willing and safe enough to experiment?
- Is there a goal, or leadership willing to establish one?
- Is there a practical route to AI-enabled implementation?
- Can the team ship and learn from small changes?
Why should you build one now?
Organisations that treat AI as a collection of individual tools may produce scattered productivity gains without changing their overall capacity to adapt.
The larger opportunity is to redesign how the organisation notices, decides, implements and learns.
Businesses that develop that capability will not simply add more features. They will be able to respond to customers faster, remove operational friction sooner, personalise experiences more economically and learn through a much larger number of real changes.
Smaller organisations may use this to outperform larger competitors by generating more learning and customer value per employee. Larger organisations that retain slow, project-based implementation cycles may find that additional headcount cannot compensate for a structurally slower learning system.
You cannot brute-force your way to the gains created by a business that continually improves how it improves.
Frequently asked questions
What is a Business Improvement Factory?
A Business Improvement Factory makes improvement a permanent operating capability, so the business can keep getting better while day-to-day work continues. It combines a software platform, living organisational context, proportionate governance and the strategists, developers and AI agents needed to implement and learn from the changes that matter.
Is a Business Improvement Factory the same as a business improvement platform?
No. A business improvement platform is the Factory’s software layer: it captures inputs, assembles context, develops specifications, applies approvals and records results. The Business Improvement Factory combines that platform with strategy, governance and access to the specialist production capacity required to implement agents, workflows, internal tools, product changes and operating-model decisions.
Is it the same as continuous-improvement software?
They overlap, but many continuous-improvement platforms focus on capturing ideas, managing initiatives and measuring results. A Business Improvement Factory also maintains persistent organisational context and includes the human-and-agent capacity required to get changes working.
Is it an AI agent platform?
No. An agent platform supplies technology for creating or operating agents. A Business Improvement Factory helps the organisation decide what should improve, supplies organisational context, governs implementation and may use many different agents, tools and systems to produce the result.
Does every organisation need to build its own platform?
No. The organisation needs to own its goals, processes, context and decisions. It can use an external platform and implementation partner to supply the technology and build capability.
Where should an organisation start?
Start at the highest level with genuine authority and participation. That might be the whole organisation, a department, a product or one important process. Establish the intended outcome, capture the relevant context and complete a small but meaningful improvement loop.
How is success measured?
The ultimate measure is contribution to a business outcome. Adoption, quality, operational effects, implementation speed and cost are signals that help explain whether a particular change is contributing to that outcome.
Bottom line
AI will not realise its full value inside organisations that retain the same slow path from insight to implementation.
A Business Improvement Factory creates a permanent operating capability for recognising where the business could be better, understanding the organisation, choosing what matters, implementing changes and learning from the results.
The goal is not to build more technology.
It is to build an organisation that gets better at getting better.
