Same AI, Two P&Ls: Where Economic Value Really Comes From

Two scientific studies explain why licences and training are not enough. Five tasks show, step by step, how fragmented use and an orchestrated approach produce different cycle times, error rates and financial outcomes.

by Andrea Viliotti

Scientific basis: Qing Xia and colleagues, CHIWORK 2026; Carsten F. Schmidt and colleagues, 2026 preprint.

AI adoption in business

For a CEO or CFO, the important question is not how many people have access to an AI assistant. It is what those people do differently when they receive a tender, prepare a management report, answer a customer, build a research brief or close a meeting.

When an employee uses a chatbot inside an unchanged process, they may become faster without the business being able to convert that advantage into lower costs, more saleable capacity or higher contribution margin. When the task is selected, redesigned, assigned, controlled and measured through a shared method, the same technology can produce an economically observable result.

The illustrative scenario in this article compares the two approaches in the same business: 60 users, the same volumes, the same six-month horizon and the same basic licences. In the central case, the fragmented approach ends at -€44,100, while the orchestrated approach ends at +€25,350. This is not a client case, a certified forecast or a guaranteed ROI. It is a transparent demonstration of how the result changes when workflows, errors, avoided costs and the ability to monetise released time all change.

AI creates potential capacity. The operating model determines how much of that capacity becomes economic value.

What the two papers tell us - and what they cannot tell us

The first study, by Qing Xia and colleagues, is based on 19 interviews with knowledge workers from different professions. The authors show that AI skills develop through direct experience and learning from colleagues. They also identify a paradox: some workers see the ability to remove the tell-tale signs of AI use as evidence of professional competence.

That behaviour may improve an individual output because it encourages the person to correct generic or poorly contextualised text. It may, however, weaken the organisation: prompts, sources, mistakes and verification criteria remain embedded in one person's working practice. Management sees the final document, not how it was produced. Knowledge becomes less transferable, and it becomes harder to understand where AI is creating value or risk.

The second study, by Carsten F. Schmidt and colleagues, examines a Microsoft 365 Copilot pilot in a large research organisation. A total of 550 licences were assigned; the two survey waves included 106 and 90 participants. The design is repeated cross-sectional and the measures are subjective, so the study is not an objective productivity measurement.

The findings are still useful. Among scientific staff, perceived usefulness rises from 0.42 to 1.09, while perceived ease of use rises from 0.74 to 1.31. Information gathering, analysis and text creation receive the strongest ratings. Perceived usefulness is weaker for socially complex, creative or difficult-to-verify activities. The pattern suggests that value increases as people learn to identify the right sub-tasks and embed them in routines.

The papers therefore support three mechanisms:

1.  value depends on the fit between the technology and the task;

2.  learning is social and organisational, not merely individual;

3.  control and routinisation evolve over time.

They do not provide the volumes, cycle times, efficiency rates or financial outcomes used in the pages that follow. All of those figures are illustrative assumptions, designed to make the model testable and replaceable with the company's own data.

The right comparison: same business, same task, different work system

By "fragmented approach", I do not mean that the market only offers software or superficial services. Fragmentation arises when licences, training, policies and proofs of concept are purchased separately and are not connected to the same workflow, the same owner or the same economic line item.

The comparison keeps the business, people, volumes, basic tools and time horizon constant. What changes is the organisational configuration.

Moment in the work	Fragmented approach	Orchestrated approach
The task arrives	The employee decides alone whether and how to use AI	The task has already been assessed for usefulness, risk and verifiability
Sources are gathered	Email, folders and personal materials	Approved repositories and sources for that workflow
AI is prompted	Personal prompts that vary from user to user	Shared prompts, templates and output criteria
Colleagues are involved	Email requests and multiple copies	Owners, deadlines and status in the same workflow
Quality is checked	Final review, when correction is most expensive	Progressive checks at the point where the error arises
The benefit is measured	Hours described as "saved"	Hours connected to overtime, avoided spend, saleable capacity or margin
Experience is reused	It remains part of an individual's practice	It becomes a workflow, checklist and organisational asset

Assumptions A1-A2: a business with 60 users; an average annual fully loaded cost of €60,000 per person; 54,000 available hours over six months; 30% of hours treated as AI candidates in the fragmented approach and 25% in the orchestrated approach. These percentages are not market benchmarks. In a real project, they must be replaced with a time audit and task inventory.

The flagship case: preparing a tender response

The task

A bid writer must analyse the documentation, extract requirements, retrieve references and technical data, coordinate contributions, draft the response and verify that no mandatory requirement has been missed.

Before AI is introduced, the process takes an average of 24 hours per tender.

Fragmented approach: AI enters, but the process remains almost unchanged

The employee has an AI licence and has attended general training, but there is no company-specific workflow for the task.

1.  She reads the documents and copies the requirements into her own spreadsheet: 4 hours.

2.  She pastes extracts into the chatbot and tests different prompts to draft sections: 5 hours.

3.  She searches old bids, folders and messages for references and evidence: 3 hours.

4.  She emails technical and commercial colleagues, receives answers in different formats and reconciles them: 3 hours.

5.  She reviews the document at the end and discovers missing requirements, absent sources or inconsistent data: 3 hours.

6.  She formats and submits the response: 1 hour.

Total time: 19 hours.

AI has reduced drafting time, but research, coordination and verification remain largely individual. The employee is faster; the process is still fragile and difficult to replicate.

Orchestrated approach: the workflow changes

The same employee receives the same tender, but the task sits inside a designed workflow.

1.  She uploads the documents into the authorised environment. AI extracts the requirements into a matrix with page reference, mandatory status, required document and completion status. She validates the matrix: 2 hours.

2.  Each requirement is assigned to an internal owner; the system retrieves approved references and evidence: 2 hours.

3.  AI drafts the sections using shared templates and links claims to company sources: 4 hours.

4.  Technical, commercial and compliance colleagues progressively review only the parts within their remit: 3 hours.

5.  The bid writer runs the closing checklist, assembles the response and submits it: 3 hours.

Total time: 14 hours.

The advantage does not come from faster writing alone. The employee no longer rebuilds the process each time, no longer chases colleagues through email and no longer discovers all errors at the end.

For an urgent tender with 16 ordinary working hours available, the fragmented approach would require three overtime hours, while the orchestrated approach would remain within ordinary hours. At €65 per hour, the immediate effect would be €195 of avoided overtime. This micro-example makes the mechanism visible, but it is not added separately to the six-month model, in order to avoid double counting.

From released time to captured value

The business handles 60 tenders over six months and assigns a fully loaded labour rate of €50 per hour.

Fragmented approach:

60 tenders x 5 hours released x €50 = €15,000 of potential capacity

Of the 300 hours released, only 120 correspond to overtime or temporary cover that is actually cancelled:

120 hours x €50 = €6,000 of captured value

Orchestrated approach:

60 tenders x 10 hours released x €50 = €30,000 of potential capacity

Of the 600 hours released, 480 replace overtime or temporary support already included in the plan:

480 hours x €50 = €24,000 of captured value

Difference on the workflow: €18,000.

Assumption A3 - Tender/RFP: 60 tenders, average times of 24/19/14 hours, fully loaded rate of €50, and 120/480 monetised hours. Measured by: bid manager and finance/FP&A. Source systems: time sheets, overtime register, procurement and project management. The value is zero if the released hours neither cancel costs nor create saleable capacity.

Where the service portfolio intervenes

The portfolio I offer does not start with software. It starts with processes, data, people, responsibilities, expected return and the ability to control outcomes. The services are not a compulsory package: they answer different questions and should only be activated when they are needed.

Stage	What changes in the operating model	Related services	KPI and source system	Economic channel
Frame the decision	Projects without a real problem, owner or success threshold are stopped	AI Maturity Quick Check; AI Decision Check	initiatives stopped before spend; decision register	lower sunk costs
Assess readiness	Baselines, data inventory and informal AI use are mapped	AI Readiness Audit; Data & KPI Readiness Check	workflows with complete baselines; ERP, CRM, time tracking	fewer unmeasurable pilots
Select priorities	A small number of tasks are chosen with an owner, KPI and go/no-go threshold	AI Strategy Sprint	share of suitable tasks; use-case portfolio	better task fit
Define rules and controls	Sources, responsibilities, human oversight and escalation are assigned	Workshop AI Governance; AI Governance & AI Act Readiness Sprint	reopened incidents and remediation time; audit log	lower rework and risk
Redesign the work	Prompts, templates, checklists and integrations become shared assets	Prompt & Workflow Kit; AI Tender/RFP Accelerator; AI Automation Feasibility Lab	cycle time, first-pass quality, verification hours; workflow system	captured productivity and earlier stop decisions
Sustain and scale	Priorities, signals, policies and shared practices are updated	GDE Intelligence Radar; Advisory AI Continuativa; Enterprise AI Adoption Program	workflow reuse, adoption, benefit persistence	selective scale and lower dependence on individuals

The correct link is not "service purchased = saving achieved". It is:

service -> workflow step changed -> KPI -> source system -> economic event -> verification

That chain makes the portfolio falsifiable. If the KPI does not change, the service cannot claim the benefit.

Four more tasks, two ways of working

1. Preparing management reporting

Fragmented approach. The controller exports and cleans data from several systems for 2.5 hours; asks the chatbot for an initial analysis for 1.5 hours; verifies revenue, margin and cost definitions for 2 hours; revises the commentary after management feedback for 1.5 hours; formats and archives the report for 1 hour. Total: 8.5 hours per report.

Orchestrated approach. The report starts from a standard extraction and an approved KPI dictionary for 1.5 hours; automated rules flag anomalies for 1 hour; AI drafts the commentary using defined metrics for 1 hour; the controller checks exceptions and causes for 1.5 hours; the manager approves within the workflow for 1 hour. Total: 6 hours.

Across 100 reports, potential capacity is €14,000 in the fragmented approach and €24,000 in the orchestrated approach. The hours genuinely linked to cancelled overtime or backfill produce €4,900 and €16,800, respectively.

Assumption A4 - Reporting: 100 reports, 12/8.5/6 hours, €40 per hour. Measured by: controller and CFO. Source systems: ERP, BI, time sheets and overtime budget. The value is zero if the released time does not change costs, measurable decision capacity or saleable output.

2. Responding to customer requests

Fragmented approach. The service representative reads the request for 3 minutes; searches email, folders and old tickets for 7; asks AI for a draft for 4; checks policy and contract for 3; personalises, sends and logs the response for 3. Total: 20 minutes.

Orchestrated approach. The request is classified by product, issue and risk in 1 minute; the system retrieves the customer profile and approved response blocks in 3; AI drafts a response linked to the relevant sources in 2; the employee checks exceptions and tone in 4; sending and logging take 2 minutes. Total: 12 minutes.

Across 2,000 requests, potential capacity is €10,000 versus €18,000. The share that reduces additional shifts or temporary support produces €3,500 in the fragmented approach and €11,700 in the orchestrated approach.

Assumption A5 - Customer service: 2,000 requests, 30/20/12 minutes, €30 per hour. Measured by: service manager and finance/FP&A. Source systems: CRM, ticketing and workforce management. The value is zero if volumes, shifts and saleable capacity remain unchanged.

3. Building an internal research brief

Fragmented approach. The analyst searches for documents for 50 minutes; asks AI for a summary and structure for 25; returns to the sources to verify data and citations for 30; rewrites and archives for 27. Total: 132 minutes.

Orchestrated approach. The analyst searches an approved corpus for 15 minutes; AI extracts relevant passages with document references for 20; prepares a draft in the company template for 20; the analyst evaluates reliability, conflicting sources and implications for 25; archives the brief and its sources for 10. Total: 90 minutes.

Across 250 briefs, potential capacity is €8,000 versus €15,000. The share that replaces overtime or external analytical support produces €2,400 and €7,500.

Assumption A6 - Internal research: 250 briefs, 3/2.2/1.5 hours, €40 per hour. Measured by: functional leader and finance/FP&A. Source systems: document management, time tracking and supplier budget. The value is zero if the additional output is not used in decisions and does not replace work or expenditure.

4. Closing a meeting and moving actions forward

Fragmented approach. The project coordinator gathers the agenda for 15 minutes; uses AI transcription or summarisation for 10; reconstructs decisions and responsibilities for 20; sends emails with actions for 20; chases updates over the following days for 20. Total: 85 minutes.

Orchestrated approach. The agenda comes from the project system in 10 minutes; transcription and AI extract decisions, owners and deadlines in 10; participants validate commitments in 10; actions are synchronised in 10; the coordinator follows only exceptions and dependencies for 20. Total: 60 minutes.

Across 400 meetings, potential capacity is €7,000 versus €12,000. The hours linked to cancelled administrative support or overtime produce €2,100 and €4,350.

Assumption A7 - Meetings: 400 meetings, 2/1.42/1 hours, €30 per hour. Measured by: PMO and finance/FP&A. Source systems: calendar, project management and time sheets. The value is zero if actions still do not get completed and the released time is spent chasing them in another form.

From workflow to P&L

Productivity reconstructed from the bottom up

Workflow	Fragmented potential	Fragmented captured	Orchestrated potential	Orchestrated captured
Tenders and RFPs	€15,000	€6,000	€30,000	€24,000
Management reporting	€14,000	€4,900	€24,000	€16,800
Customer requests	€10,000	€3,500	€18,000	€11,700
Internal research briefs	€8,000	€2,400	€15,000	€7,500
Meetings and follow-up	€7,000	€2,100	€12,000	€4,350
Total	€54,000	€18,900	€99,000	€64,350

The aggregate percentages emerge from these values, not the other way around:

• fragmented: €54,000 / (€1,800,000 x 30%) = 10% potential efficiency; €18,900 / €54,000 = 35% capture;

• orchestrated: €99,000 / (€1,800,000 x 25%) = 22% potential efficiency; €64,350 / €99,000 = 65% capture.

Assumption A8: the percentages summarise the five workflows. They must be recalculated if volumes, cycle times, labour cost or the share of monetised hours changes.

The bridge that explains the difference

Step	Change	Cumulative result
Fragmented approach result	-€44,100	-€44,100
Higher direct cost of orchestration	-€15,000	-€59,100
Higher captured productivity	+€45,450	-€13,650
Lower rework	+€11,000	-€2,650
Net external spend avoided	+€8,000	+€5,350
Higher incremental contribution margin	+€20,000	+€25,350

The bridge reveals the most important point: productivity and lower rework are not enough, on their own, to make the orchestrated approach positive within six months. After those two levers, the result is still -€2,650. The sign changes because external expenditure is genuinely cancelled and an incremental order is genuinely won.

The remaining lines, with an event and a zero condition

Incremental contribution margin. In the fragmented approach, an additional order worth €20,000 with €15,000 of variable costs produces €5,000. In the orchestrated approach, an order worth €100,000 with €75,000 of variable costs produces €25,000. The hours required to prepare and deliver those orders are excluded from captured productivity. Assumption A9: if the order is not won or is not genuinely incremental, the margin is zero.

External spend avoided. In the orchestrated approach, four external research briefs worth €10,000 are cancelled; 40 internal hours at €50 cost €2,000; the net benefit is €8,000. Assumption A10: if the contract is not cancelled or the internal cost equals the supplier quote, the value is zero.

Rework. Across 300 deliverables, the fragmented approach generates 60 incidents requiring 6 hours each at €50, equal to €18,000. The orchestrated approach generates 28 incidents requiring 5 hours each, equal to €7,000. Assumption A11: only errors reopened after approval or delivery are counted; ordinary review is already included in workflow time.

Direct costs. The fragmented approach includes €18,000 of licences, €8,000 of training, €14,000 for a proof of concept and €10,000 of internal coordination: €50,000 in total. The orchestrated approach includes the same licences, €15,000 for decision/readiness/strategy, €22,000 for governance/workflow/pilot and €10,000 for measurement and change: €65,000 in total. Assumption A12: this is an illustrative envelope, not a quotation or price list.

The scenario P&L

Six-month line item	Fragmented approach	Orchestrated approach
Captured productivity	+€18,900	+€64,350
Incremental contribution margin	+€5,000	+€25,000
Net external spend avoided	€0	+€8,000
Rework	-€18,000	-€7,000
Direct costs	-€50,000	-€65,000
Net scenario value	-€44,100	+€25,350

The ratio of net value to direct cost is -88.2% for the fragmented approach and +39.0% for the orchestrated approach. This is an internal return within the illustrative scenario, not an empirically proven ROI.

Assumption A13: the result depends on the benefit channels being independent. If contribution margin and avoided external spend are already the monetisation of captured productivity, they cannot be added again.

The tougher test: break-even and counterfactuals

When contribution margin and external spend are treated as independent channels, the orchestrated approach breaks even at a capture rate of 39.39%.

Under the prudent reading, counting only captured productivity and subtracting rework and cost, the threshold rises to 72.73%. The assumed 65% is not enough:

€64,350 - €7,000 - €65,000 = -€7,650

The approach remains better than the fragmented one, but it does not break even within six months.

Event that does not occur	Orchestrated net value
The incremental order is not won	+€350
The order is not won and external briefs are not cancelled	-€7,650
Rework remains at the fragmented level	+€14,350
Capture falls to 50%, while margin and avoided spend remain	+€10,500
Capture at 50%, no order and no avoided external spend	-€22,500

This table is more useful than a promise: it shows which event changes the sign of the business case.

The second clock of adoption: value does not appear in full in month one

The minutes spent on a single task belong to operational time. The organisation also has a second clock: the time required to learn, standardise, correct and increase the share of benefit that is genuinely captured.

In the central case, 65% is treated as the average capture rate across the six months. If it were only the level reached in month six, the result would change.

With linear growth from 30% to 65%, average capture would be 47.5%:

€99,000 x 47.5% + €25,000 + €8,000 - €7,000 - €65,000 = +€8,025

With growth from 20% to 65%, average capture would be 42.5%:

net result = +€3,075

This sensitivity explains why a pilot must measure the month-by-month ramp-up. Stating only "65%" without saying whether it is an average or an exit rate creates false precision.

From narrative to evidence: a six-month measurement contract

Turning the scenario into company evidence requires only a few objects, but they must be defined before the pilot begins.

Object	Formula or question	Owner	Source system
Time per unit	minutes/hours from task opened to task closed	process owner	workflow, ticketing, time sheet
First-pass quality	deliverables approved without reopening / total	quality owner	audit log
Escaped rework	post-approval incidents x hours x cost	finance/FP&A	incident register
Capture rate	monetised value / potential capacity	CFO/FP&A	budget, payroll, workforce management
External spend avoided	contracts cancelled - internal substitution cost	procurement/CFO	ERP and procurement
Incremental contribution margin	additional revenue - variable costs	commercial/CFO	CRM and ERP
Total cost	licences + suppliers + internal hours + change	project owner	project accounting

The minimum design should include:

1.  frozen baseline: at least four weeks of volumes, times, errors and costs before the change;

2.  comparison unit: comparable teams, processes or periods, avoiding comparisons between different activities;

3.  benefit-channel register: each euro can appear only once as avoided cost, margin, external spend or risk;

4.  monthly adoption curve: active-user share, workflow reuse, capture, errors and verification cost;

5.  decision thresholds: stop, modify or scale criteria defined before the result is known;

6.  ex-post verification: data owner, measurement date and reconciliation with the budget or management P&L.

The thresholds can be simple. For example:

• stop: no reduction in cycle time or unchanged rework after two correction cycles;

• modify: lower cycle time but economic capture below break-even;

• scale: quality not impaired, independent channels documented and positive value after full costs.

This is the point at which the portfolio gains financial credibility. Not because it promises an outcome, but because it builds the system through which the outcome can be confirmed or disproved.

The decision that remains with management

The fragmented approach can create faster employees, some excellent prompts and local benefits. The business may still not know which sources were used, how much verification costs, which errors emerge after delivery, which hours genuinely changed the budget or whether the practice can be replicated.

The orchestrated approach attempts to change that condition. It does not replace the employee: it shifts their judgement away from duplicated research and email reconciliation, towards exceptions, sources, accountability and decisions.

Before funding an AI initiative, management should be able to answer six questions:

1.  which employee performs which task;

2.  how the task is performed today, step by step;

3.  how it should be performed in the new workflow;

4.  which time, error or hand-off is removed;

5.  which economic line genuinely changes;

6.  which result will trigger continuation, modification or termination.

If one of those answers is missing, the business does not yet have a business case. It has a technology hypothesis.

Through-line: the licence makes AI available. Value emerges when workflows, measurement and accountability turn that availability into a verifiable economic decision.

Methodology and transparency note

The business, volumes, times, costs, orders and incidents described are illustrative examples. They do not represent a client, price list, quotation or certified forecast.

Productivity is treated as economically captured only when the released hours are connected to an avoided cost, saleable capacity or another documented line item. Contribution margin, avoided external spend and productivity may be added only when they arise from independent events and do not use the same hours.

Sources

1. Qing (Nancy) Xia, Marios Constantinides, Advait Sarkar, Duncan P. Brumby and Anna L. Cox, "If You're Very Clever, No One Knows You've Used It": The Social Dynamics of Developing Generative AI Literacy in the Workplace, CHIWORK 2026, DOI 10.1145/3808045.3808050. DOI

2. Carsten F. Schmidt, Sophie Petzolt, Wolfgang Beinhauer, Ingo Weber and Stefan Langer, Generative AI in Knowledge Work: Perception, Usefulness, and Acceptance of Microsoft 365 Copilot, preprint, version dated 20 February 2026, under review. arXiv

3. Andrea Viliotti, AI Consulting for Businesses: Strategy and Governance, official services page, accessed on 18 June 2026. Official page