AI consulting, AI transformation, change management
AI consulting, AI transformation, change management
Successful AI transformation starts with how work gets done
Successful AI transformation starts with how work gets done
Businesses that turn AI into lasting change start with the work, the tools people actually use and the organisation’s ability to make decisions.
Businesses that turn AI into lasting change start with the work, the tools people actually use and the organisation’s ability to make decisions.
Some businesses are already using AI to make lasting changes, with gains their finance teams can measure and validate. Research is beginning to identify what these organisations have in common.
A BCG Henderson Institute and Columbia Business School survey of around 1,400 employees and leaders in August 2025 found a strong association between employee centricity and AI maturity. Organisations in the higher employee-centricity categories were seven times as likely to fall into the study’s highest AI-maturity category. Employee centricity explained more of the observed variation than company size or industry. These are survey-based associations, not proof of causation.
Comparable tools are available to many businesses. How an organisation listens to and supports its people deserves attention alongside its models and budgets.
Reported AI maturity does not show how work gets done
AI maturity describes an organisation’s ability to turn tools into lasting practices, better decisions and measurable value.
Yet many assessments still begin with a score produced in a workshop with a few managers. That exercise has value: it creates a shared vocabulary, brings ambitions into view and provides a reference point. Its limitation is that it captures perspectives shaped by each person’s position.
The sponsor describes a programme making progress. The tool owner describes a solution people have adopted. IT describes infrastructure that is available. Each sees a different part of the picture.
Two distinctions matter.
First, an activated tool is not necessarily useful in practice. A licence tells us little about the work it changes, the time it saves or the additional checking it creates.
Second, deployment does not establish value. A solution may be available, used and appreciated without yet improving turnaround, lowering costs, increasing capacity or strengthening decisions.
Deloitte’s global State of AI in the Enterprise 2026 research illustrates the gap. Only 25% of respondents reported moving at least 40% of their AI pilots into production. Access to tools and reliable use in production are different stages.
Much of the value still to be captured lies between making tools available and demonstrating their contribution.
Measure the gap between AI strategy and everyday experience
The BCG Henderson Institute and Columbia Business School survey found another useful difference. Some 76% of executives agreed that employees felt enthusiastic and optimistic about AI adoption, compared with 31% of individual contributors—a 45-percentage-point gap. Both groups were assessing employee sentiment. The study also found a gap in views about how well informed employees were about AI strategy and tools. Read the survey findings.
These differences do not automatically mean that leaders communicate poorly or that teams resist change. They show that strategy and work are observed from different positions.
Leaders see investments, programmes, governance and ambitions. Employees see the tools they can use, their limitations, missing rules, the time spent checking answers and the everyday decisions needed to keep working.
The gap between these views is rarely measured, yet it is useful management information. A large gap may indicate:
different interpretations of the strategy across levels;
uses that are more advanced than leaders realise;
overly optimistic reporting;
rules that exist on paper but are absent from daily work;
benefits visible to leaders while teams absorb the checking costs;
a decision process that prevents early warnings from reaching the right people.
Treat this gap as an organisational issue with causes, consequences and someone responsible for addressing it. To understand it, listen to both sides.
Recognise what unofficial AI use has already taught teams
‘Shadow AI’ means using AI tools for work without those tools being officially provided, approved or recorded by the company.
The risks are real: uncontrolled data sharing, unclear checking rules, reliance on personal accounts, unreported errors and gaps in traceability. But unofficial use can also reveal learning that has outpaced the organisation.
Generative AI has made first drafts much cheaper, often faster than usage policies have evolved. Employees have tried the tools on tasks that caused them real difficulty, without waiting for a central programme to identify every useful application.
These uses can therefore point to a gap between available tools and company rules. Once identified and assessed, they provide:
a map of tasks where AI is already helping;
an inventory of data that may be exposed;
practices to evaluate and potentially extend;
needs that official tools do not meet;
a starting point for governance and, depending on the systems involved, compliance work.
Employees need conditions in which they can describe these uses candidly. Few will volunteer information about a personal tool that may breach a rule they do not fully understand. Confidential conversations, with individual responses inaccessible to the employer, create a different basis for discussion.
The aim is to understand the AI already in use before deciding how to organise what comes next.
Reported AI maturity does not show how work gets done
AI maturity describes an organisation’s ability to turn tools into lasting practices, better decisions and measurable value.
Yet many assessments still begin with a score produced in a workshop with a few managers. That exercise has value: it creates a shared vocabulary, brings ambitions into view and provides a reference point. Its limitation is that it captures perspectives shaped by each person’s position.
The sponsor describes a programme making progress. The tool owner describes a solution people have adopted. IT describes infrastructure that is available. Each sees a different part of the picture.
Two distinctions matter.
First, an activated tool is not necessarily useful in practice. A licence tells us little about the work it changes, the time it saves or the additional checking it creates.
Second, deployment does not establish value. A solution may be available, used and appreciated without yet improving turnaround, lowering costs, increasing capacity or strengthening decisions.
Deloitte’s global State of AI in the Enterprise 2026 research illustrates the gap. Only 25% of respondents reported moving at least 40% of their AI pilots into production. Access to tools and reliable use in production are different stages.
Much of the value still to be captured lies between making tools available and demonstrating their contribution.
Measure the gap between AI strategy and everyday experience
The BCG Henderson Institute and Columbia Business School survey found another useful difference. Some 76% of executives agreed that employees felt enthusiastic and optimistic about AI adoption, compared with 31% of individual contributors—a 45-percentage-point gap. Both groups were assessing employee sentiment. The study also found a gap in views about how well informed employees were about AI strategy and tools. Read the survey findings.
These differences do not automatically mean that leaders communicate poorly or that teams resist change. They show that strategy and work are observed from different positions.
Leaders see investments, programmes, governance and ambitions. Employees see the tools they can use, their limitations, missing rules, the time spent checking answers and the everyday decisions needed to keep working.
The gap between these views is rarely measured, yet it is useful management information. A large gap may indicate:
different interpretations of the strategy across levels;
uses that are more advanced than leaders realise;
overly optimistic reporting;
rules that exist on paper but are absent from daily work;
benefits visible to leaders while teams absorb the checking costs;
a decision process that prevents early warnings from reaching the right people.
Treat this gap as an organisational issue with causes, consequences and someone responsible for addressing it. To understand it, listen to both sides.
Recognise what unofficial AI use has already taught teams
‘Shadow AI’ means using AI tools for work without those tools being officially provided, approved or recorded by the company.
The risks are real: uncontrolled data sharing, unclear checking rules, reliance on personal accounts, unreported errors and gaps in traceability. But unofficial use can also reveal learning that has outpaced the organisation.
Generative AI has made first drafts much cheaper, often faster than usage policies have evolved. Employees have tried the tools on tasks that caused them real difficulty, without waiting for a central programme to identify every useful application.
These uses can therefore point to a gap between available tools and company rules. Once identified and assessed, they provide:
a map of tasks where AI is already helping;
an inventory of data that may be exposed;
practices to evaluate and potentially extend;
needs that official tools do not meet;
a starting point for governance and, depending on the systems involved, compliance work.
Employees need conditions in which they can describe these uses candidly. Few will volunteer information about a personal tool that may breach a rule they do not fully understand. Confidential conversations, with individual responses inaccessible to the employer, create a different basis for discussion.
The aim is to understand the AI already in use before deciding how to organise what comes next.
Understanding how work gets done is a practical instrument of transformation.
Understanding how work gets done is a practical instrument of transformation.
Perspective
Perspective
Understanding how work gets done is a practical instrument of transformation.
Perspective
Three realities to understand before deciding
An assessment needs to look beyond ambitions, investment and licence counts to examine three aspects of the organisation.
How AI is actually used
Which tools do people use, whether provided by the company or chosen individually? For which tasks, how often, with what data and with what level of confidence?
Measure the time spent checking, correcting or reworking results. A tool may speed up a first draft while shifting work into review. The useful unit of analysis is the task as a whole: its result, risks and total effort.
How the work is actually done
Procedures never capture every detail of work. Teams manage exceptions, look for scattered information, compensate for system limitations and create shortcuts to keep things moving. Some practices create friction; others are useful solutions the organisation has never formalised.
These real tasks provide the starting point for relevant use cases: a slow decision, inaccessible data, repeated entry, knowledge that is hard to find or checks that take too much time. Observe the situation and estimate its impact before choosing a solution.
That may lead to removing a task, simplifying a rule, automating without AI or preserving a human responsibility entirely.
The ability to make and review decisions
A solution needs more than technical performance. It needs an owner, an objective, feedback and a route for resolving decisions.
Who authorises deployment? Who checks results and receives error reports? Who can change the rules, suspend use or stop the system? What happens when quality falls, data changes or an employee challenges the tool’s recommendation?
In our view, this decision-making capacity is essential to sustained use. Without clear responsibilities and feedback, a promising pilot is difficult to correct, govern and maintain.
Sociotechnical thinking made this point long before generative AI: social and technical systems need to be designed together. A technical improvement can underperform when its organisational setting is overlooked.
Technology changes work when it is connected to responsibilities, rules, skills and decisions.
From reported maturity to demonstrated maturity
Four practices help turn intentions into evidence.
First, invite everyone in scope to contribute, including people outside management and tool ownership.
Second, compare their experience with leaders’ objectives, committed budgets, accepted risks and intended level of maturity.
Third, measure the differences in perceptions, practices, available data and understanding of responsibilities.
Fourth, assess each finding according to the strength of the evidence behind it.
A statement, a document, a record of execution and a verified result provide different levels of assurance. A policy shows that a rule was written; it does not establish that it is followed. An activated tool shows availability; it does not establish value. A completed pilot shows that a system worked under particular conditions; it does not yet establish that it can be integrated, adopted and maintained.
Spentia’s AI maturity assessment is built on this principle: compare objectives with the work itself, and give important decisions a stronger basis than self-reported claims alone.
Demonstrated maturity broadens your options. You can accelerate where the evidence is strong, prepare data and processes where conditions are incomplete, or defer where value is still unproven. The ability to defer matters.
A project can succeed technically without contributing economically. It can meet deadlines and satisfy pilot users while leaving an important business indicator unchanged.
Assess value after accounting for the full cost:
licences and model usage;
integration;
maintenance;
human supervision;
checking outputs;
handling errors;
supplier dependence;
the cost of withdrawal if needed.
Time theoretically saved does not automatically become a financial saving. It may improve quality, increase capacity, shorten turnaround or allow more requests to be handled. Define the intended outcome before deployment, then measure it afterwards.
Understanding the work helps focus deployment
Starting with the work may appear slower than selecting a tool or completing a workshop scorecard. It can, however, help concentrate investment on the situations that matter, avoid pilots without owners and give teams solutions grounded in their needs.
It also reveals what already works. Employees test, correct, compare and find ways to work with the tools available. A mature organisation makes that initiative visible, controls the risks and turns the strongest practices into shared capabilities.
An AI maturity score is therefore only a starting point. The more useful question is: what can we demonstrate about our uses of AI, the work they change and our ability to govern them?
Understanding the work is a practical instrument of transformation.
Explore further: our approach to AI consulting.
Three realities to understand before deciding
An assessment needs to look beyond ambitions, investment and licence counts to examine three aspects of the organisation.
How AI is actually used
Which tools do people use, whether provided by the company or chosen individually? For which tasks, how often, with what data and with what level of confidence?
Measure the time spent checking, correcting or reworking results. A tool may speed up a first draft while shifting work into review. The useful unit of analysis is the task as a whole: its result, risks and total effort.
How the work is actually done
Procedures never capture every detail of work. Teams manage exceptions, look for scattered information, compensate for system limitations and create shortcuts to keep things moving. Some practices create friction; others are useful solutions the organisation has never formalised.
These real tasks provide the starting point for relevant use cases: a slow decision, inaccessible data, repeated entry, knowledge that is hard to find or checks that take too much time. Observe the situation and estimate its impact before choosing a solution.
That may lead to removing a task, simplifying a rule, automating without AI or preserving a human responsibility entirely.
The ability to make and review decisions
A solution needs more than technical performance. It needs an owner, an objective, feedback and a route for resolving decisions.
Who authorises deployment? Who checks results and receives error reports? Who can change the rules, suspend use or stop the system? What happens when quality falls, data changes or an employee challenges the tool’s recommendation?
In our view, this decision-making capacity is essential to sustained use. Without clear responsibilities and feedback, a promising pilot is difficult to correct, govern and maintain.
Sociotechnical thinking made this point long before generative AI: social and technical systems need to be designed together. A technical improvement can underperform when its organisational setting is overlooked.
Technology changes work when it is connected to responsibilities, rules, skills and decisions.
From reported maturity to demonstrated maturity
Four practices help turn intentions into evidence.
First, invite everyone in scope to contribute, including people outside management and tool ownership.
Second, compare their experience with leaders’ objectives, committed budgets, accepted risks and intended level of maturity.
Third, measure the differences in perceptions, practices, available data and understanding of responsibilities.
Fourth, assess each finding according to the strength of the evidence behind it.
A statement, a document, a record of execution and a verified result provide different levels of assurance. A policy shows that a rule was written; it does not establish that it is followed. An activated tool shows availability; it does not establish value. A completed pilot shows that a system worked under particular conditions; it does not yet establish that it can be integrated, adopted and maintained.
Spentia’s AI maturity assessment is built on this principle: compare objectives with the work itself, and give important decisions a stronger basis than self-reported claims alone.
Demonstrated maturity broadens your options. You can accelerate where the evidence is strong, prepare data and processes where conditions are incomplete, or defer where value is still unproven. The ability to defer matters.
A project can succeed technically without contributing economically. It can meet deadlines and satisfy pilot users while leaving an important business indicator unchanged.
Assess value after accounting for the full cost:
licences and model usage;
integration;
maintenance;
human supervision;
checking outputs;
handling errors;
supplier dependence;
the cost of withdrawal if needed.
Time theoretically saved does not automatically become a financial saving. It may improve quality, increase capacity, shorten turnaround or allow more requests to be handled. Define the intended outcome before deployment, then measure it afterwards.
Understanding the work helps focus deployment
Starting with the work may appear slower than selecting a tool or completing a workshop scorecard. It can, however, help concentrate investment on the situations that matter, avoid pilots without owners and give teams solutions grounded in their needs.
It also reveals what already works. Employees test, correct, compare and find ways to work with the tools available. A mature organisation makes that initiative visible, controls the risks and turns the strongest practices into shared capabilities.
An AI maturity score is therefore only a starting point. The more useful question is: what can we demonstrate about our uses of AI, the work they change and our ability to govern them?
Understanding the work is a practical instrument of transformation.
Explore further: our approach to AI consulting.
