Digital Corporate Finance, Analytics and AI

Introduction

Corporate finance is increasingly conducted through digital platforms, automated workflows, cloud systems, application programming interfaces (APIs), advanced analytics and artificial intelligence (AI). The finance function is therefore changing from a primarily retrospective reporting activity into an integrated decision system that supports forecasting, capital allocation, liquidity management, risk control and strategic execution in near real time.

Digital corporate finance does not mean simply replacing spreadsheets with software. It involves redesigning financial information flows so that data is captured once, governed properly, analysed intelligently and converted into decisions. This transformation creates significant opportunities: faster close processes, more accurate forecasts, improved working-capital management, continuous risk monitoring and more responsive investment decisions. It also introduces material risks involving data quality, cybersecurity, privacy, algorithmic bias, model instability, third-party dependence and regulatory compliance.

For a Chartered Finance Manager, the central issue is not whether a technology is fashionable. It is whether the technology improves decision quality, control effectiveness and stakeholder value while remaining explainable, secure and economically justified.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain the architecture and strategic role of digital corporate finance.
2. Construct and evaluate integrated financial models for forecasting and decision support.
3. Distinguish business intelligence, predictive analytics, machine learning and generative AI.
4. Apply forecasting, scenario analysis and performance metrics to corporate decisions.
5. Evaluate automation, fintech, digital assets and real-time finance applications.
6. Design controls for data quality, cybersecurity, privacy, bias and model risk.
7. Recommend a governed implementation approach for AI-enabled finance.

1. The Digital Finance Operating Model

A traditional finance function often consists of disconnected systems: an enterprise resource planning (ERP) system for transactions, spreadsheets for planning, separate treasury applications, manually prepared management reports and emails for approvals. Such fragmentation creates multiple versions of the truth, delayed information and high dependence on individual employees.

A digital finance operating model integrates five layers:

The model is valuable only when the layers are connected. A sophisticated dashboard based on incomplete or incorrectly classified transactions may provide a misleading appearance of precision. Similarly, an accurate forecast that is delivered after the investment committee has met has limited practical value. Digital finance must therefore be assessed on both analytical quality and decision latency.

Important characteristics

Integration means that financial, operational and external data can be connected through common definitions and controlled interfaces. Automation means that repeatable activities are executed by systems with limited manual intervention. Real-time or near-real-time finance means that information is updated frequently enough to support current decisions; it does not necessarily mean that every transaction is processed instantaneously. Self-service analytics allows authorised managers to explore information without requesting every report from finance. Embedded finance places financial analysis and controls directly into operational processes, such as credit limits within an order-entry system.

The finance leader remains responsible for interpretation, challenge and accountability. Technology can calculate, classify and recommend, but it does not possess organisational authority or professional judgement.

2. Financial Modelling for Digital Decision Support

A financial model is a structured representation of the economic relationships among revenues, costs, assets, liabilities, cash flows, financing and value. Digital tools make models easier to update and distribute, but they do not eliminate the need for sound assumptions.

2.1 Integrated model structure

A corporate model commonly links:

• an income statement;
• a balance sheet;
• a cash-flow statement;
• operating drivers such as volume, price, headcount and capacity;
• financing schedules;
• tax and depreciation schedules; and
• valuation and scenario outputs.

A fundamental cash-flow identity is:

The terms must refer to the same period and use the same currency basis. In an integrated model, net income is not equal to cash flow because accrual accounting recognises revenues and expenses when earned or incurred, whereas cash flow records changes in cash. A simplified free-cash-flow-to-the-firm relationship is:

where FCFF is free cash flow to the firm, EBIT is earnings before interest and tax, T is the applicable tax rate, D&A is depreciation and amortisation, Capex is capital expenditure, and ∆NWC is the increase in operating net working capital. The formula assumes that EBIT is measured before financing costs, tax is applied consistently, depreciation is non-cash, capital expenditure is a cash investment, and an increase in operating working capital consumes cash. If working capital decreases, ∆NWC is negative and the term increases cash flow.

For valuation, when FCFF is expected to grow at a constant rate indefinitely and the discount rate exceeds the growth rate, the perpetuity-growth expression is:

Here V₀ is present value, FCFF₁ is next-period FCFF, r is the appropriate discount rate and g is the constant growth rate. The condition r > g is essential. This is a special case, not a universal valuation formula; firms with changing growth, finite lives or material financial distress require more appropriate models.

2.2 Driver-based modelling

Driver-based models express financial outcomes through operational causes. For example:

Such models are more useful than simply extrapolating last year's percentage change because managers can challenge the underlying drivers. A revenue forecast may distinguish customer acquisition, retention, product mix, foreign exchange and pricing. A workforce model may link headcount to activity volumes and productivity rather than applying a uniform growth percentage.

A sound model separates:

1. historical actuals;
2. management assumptions;
3. externally sourced assumptions;
4. formula-driven calculations; and
5. judgemental overlays.

This separation improves auditability and makes it possible to investigate why a forecast changed.

2.3 Scenario, sensitivity and simulation analysis

A sensitivity analysis changes one or a small number of assumptions while holding others constant. A scenario analysis changes a coherent set of assumptions, such as a recession scenario involving lower volume, higher bad debts, weaker currency and increased borrowing costs. A simulation assigns probability distributions to uncertain inputs and generates a distribution of outcomes.

For a project with uncertain cash flows, expected value can be expressed as:

where pᵢ is the probability of outcome i, Vᵢ is its value and the probabilities sum to one. Expected value is not necessarily the value that will occur, and it does not by itself capture risk aversion, liquidity constraints or extreme downside exposure.

Sensitivity analysis is particularly useful for identifying value drivers. If a small change in customer retention causes a large change in value, retention deserves management attention and potentially better data collection. Digital models can run thousands of scenarios, but computational speed should not be confused with economic understanding.

3. Business Intelligence and Predictive Analytics

Business intelligence (BI) primarily answers: What happened? What is happening? Predictive analytics asks: What is likely to happen? Prescriptive analytics goes further by asking: What action should be taken, subject to constraints?

3.1 Descriptive and diagnostic analytics

Descriptive analytics includes financial statements, variance reports, dashboards and key performance indicators (KPIs). Diagnostic analytics investigates causes through drill-downs, correlation analysis, reconciliations and process mining. A gross-margin dashboard may reveal that the overall margin fell, while diagnostic analysis identifies a product mix shift, a supplier price increase or a revenue-recognition issue.

A useful KPI must have a clear definition, owner, time period, unit, data source and decision purpose. For example, “cash conversion” is ambiguous unless the organisation specifies whether it means operating cash flow divided by EBITDA, or a working-capital cycle measure.

The cash conversion cycle is commonly expressed as:

where DIO is days inventory outstanding, DSO is days sales outstanding and DPO is days payables outstanding. A simplified formulation is:

The denominator must be chosen consistently with the organisation's policy and the number of days must match the reporting convention. These measures are indicators, not complete explanations of liquidity.

3.2 Predictive analytics and machine learning

Predictive models use historical data and explanatory variables to estimate future outcomes. Examples include:

• probability of customer default;
• sales-volume forecasting;
• cash receipts forecasting;
• fraud-risk scoring;
• equipment failure prediction; and
• employee attrition analysis.

A basic linear model can be represented as:

where y is the outcome, x₁ … xₖ are explanatory variables, β₀ is the intercept, β₁ … βₖ are estimated coefficients and ε is the unexplained error. The linear form assumes an additive relationship unless variables are transformed. It does not prove causation merely because a coefficient is statistically significant.

For a binary event such as default, logistic regression estimates a probability between zero and one:

where z = β₀ + β₁x₁ + … + βₖxₖ. The model is appropriate for probability estimation under its modelling assumptions, but its outputs require calibration, monitoring and a decision threshold. A threshold selected for fraud detection may prioritise catching suspicious cases, whereas a threshold selected for customer approval may prioritise avoiding false rejections.

3.3 Forecast evaluation

Forecast accuracy should be assessed using an out-of-sample or rolling-time evaluation rather than only fitting the historical data. Common measures include:

Here yₜ is the actual value, ŷₜ is the forecast and n is the number of observations. MAE is expressed in the units of the outcome and is less sensitive to large errors than RMSE. RMSE penalises large errors more heavily. Percentage-based metrics can be problematic when actual values are zero or close to zero.

A model with the lowest average error is not automatically the best management model. Forecast bias, interpretability, stability, computational cost and the consequences of errors must also be considered. In finance, a slightly less accurate but transparent model may be preferable when decisions are regulated or highly consequential.

4. Automation and Intelligent Workflows

Robotic process automation (RPA) uses configured software agents to perform repetitive, rule-based activities, such as extracting invoice data, matching purchase orders, preparing reconciliations and routing approvals. Intelligent automation extends this approach through optical character recognition, natural-language processing, machine learning and workflow orchestration.

Automation is most suitable when a process is stable, rules are explicit, inputs are sufficiently standardised and exceptions can be identified. Automating a poorly designed process can reproduce errors more quickly. Before implementation, finance should document the process, remove unnecessary steps, define control points and establish exception handling.

A robust automated control framework includes:

• segregation of duties between preparation, approval and release;
• access controls based on role and least privilege;
• complete logs of inputs, actions and outputs;
• reconciliation between source and destination systems;
• exception queues monitored by accountable employees;
• controlled changes to scripts, models and rules; and
• a tested fallback process for system outages.

Straight-through processing is not the same as control-free processing. The objective is to automate routine execution while preserving human oversight where risk, judgement or legal responsibility is material.

5. Generative AI in Corporate Finance

Generative AI systems produce text, code, summaries, classifications and other outputs from prompts and data. Finance applications include drafting management commentary, summarising contracts, explaining variances, assisting with spreadsheet formulas, generating scenario narratives and supporting internal-service queries.

Generative AI differs from conventional predictive analytics. A predictive model is generally designed to estimate a defined target, whereas a generative system produces probabilistic content. A fluent answer may therefore be factually incorrect, incomplete or unsupported. This is often described as hallucination, but the managerial implication is more important than the label: output must be verified against authoritative data.

Controlled use of generative AI

A professional implementation should include:

1. Purpose limitation: define approved use cases and prohibited uses.
2. Data classification: prevent confidential, personal or price-sensitive information from entering unauthorised systems.
3. Grounding: connect responses to approved corporate documents or databases where feasible.
4. Human review: require review before external reporting, accounting entries, regulatory submissions or consequential decisions.
5. Traceability: retain prompts, source documents, model version, output and reviewer evidence where appropriate.
6. Evaluation: test accuracy, completeness, bias, security and robustness before deployment.
7. Change management: reassess performance when the model, data or business context changes.

Generative AI can increase productivity, but accountability cannot be delegated to a model or a software vendor. The finance function must distinguish assistance from authorisation.

6. Fintech, Digital Assets and Real-Time Finance

Fintech refers broadly to technology-enabled financial products and services. Corporate applications include digital payments, embedded lending, open banking, automated treasury, electronic invoicing, supply-chain finance and alternative credit assessment.

Open banking and APIs can provide faster access to bank balances and payment information, subject to consent, authentication, data standards and regulatory requirements. Treasury teams may use centralised cash visibility to improve liquidity forecasts and reduce idle balances. However, API dependence creates operational and third-party risks, including service interruption, authentication failure and inconsistent data formats.

Digital assets

Digital assets include crypto-assets, tokenised securities, stablecoins and other blockchain-based representations of value or rights. Their potential corporate uses include programmable payments, automated settlement, tokenised ownership and transparent transaction records. Their risks include price volatility, liquidity constraints, private-key loss, fraud, uncertain legal rights, custody risk, sanctions exposure, accounting complexity and changing regulation.

A digital asset should not be evaluated only by its technical design. A corporate finance assessment should consider:

• the legal nature of the asset and enforceability of rights;
• the identity and governance of the issuer or protocol;
• custody and recovery arrangements;
• market liquidity and valuation methodology;
• counterparty and smart-contract risk;
• financial reporting and tax treatment;
• cybersecurity and sanctions controls; and
• strategic necessity compared with conventional alternatives.

Real-time finance

Real-time finance combines frequent transaction feeds, automated reconciliations, continuous controls and rapid analytical updates. It can improve cash management and fraud detection, but it may also encourage excessive short-termism. A real-time alert is not necessarily a material event, and a continuously updated dashboard may distract from longer-term capital allocation. Finance leaders should define escalation thresholds and reporting frequencies appropriate to each decision.

7. Data Governance, Cybersecurity and Privacy

Data is an economic asset only when it is reliable, accessible for legitimate purposes and protected from misuse. A data-governance framework assigns ownership and establishes standards for data quality, metadata, retention, access and permitted use.

Data quality

Important dimensions include:

• accuracy: data correctly represents the underlying event;
• completeness: required fields and records are present;
• consistency: values agree across systems;
• timeliness: data is available when needed;
• validity: values conform to rules and formats; and
• uniqueness: duplicate records are controlled.

A practical data-quality control cycle is: define critical data elements, establish rules, measure exceptions, assign ownership, correct root causes and monitor trends. A dashboard showing a data-quality score without a remediation process has limited value.

Cybersecurity

Finance systems are attractive targets because they contain payment instructions, payroll data, customer information and commercially sensitive forecasts. Essential controls include multifactor authentication, privileged-access management, encryption, network segmentation, secure backups, endpoint protection, vulnerability management, phishing resistance, incident response and independent testing.

The principle of least privilege means that a user or service receives only the access required for its authorised task. Security should be layered: no single control should be assumed to prevent every attack. Payment controls should include independent verification of beneficiary changes, dual approval for high-risk transactions and monitoring for unusual patterns.

Privacy

Privacy governance requires a lawful and transparent purpose for collecting and processing personal data, data minimisation, appropriate retention, access controls and mechanisms for handling rights requests where applicable. A finance model should avoid using personal attributes merely because they improve predictive accuracy. Sensitive information can create legal, ethical and reputational exposure even when a model's statistical performance improves.

8. Bias, Explainability and Model Risk

Model risk is the possibility of adverse consequences arising from incorrect specifications, poor data, implementation errors, misuse or changes in the environment. It applies to spreadsheets, statistical models, machine-learning systems and AI tools.

A model-risk management framework normally includes:

1. an inventory of models and risk classification;
2. documented purpose, assumptions and limitations;
3. independent validation proportionate to materiality;
4. performance, stability and drift monitoring;
5. controlled implementation and access;
6. defined approval and escalation responsibilities; and
7. periodic review, retirement or redevelopment.

Bias may enter through historical decisions, sampling, measurement, missing data, proxy variables or the choice of outcome. Removing an explicitly sensitive variable does not necessarily remove bias if other variables act as proxies. Fairness also involves competing objectives: equal error rates, equal access, calibrated probabilities and financial sustainability cannot always be achieved simultaneously.

Explainability should be matched to the decision. A treasury forecast may require driver decomposition; a credit decision may require reasons that can be communicated to an applicant; a fraud model may require secure explanations that do not reveal detection logic. Explainability does not guarantee correctness, but lack of explainability makes challenge and accountability more difficult.

9. Implementation Framework for Finance Leaders

Digital transformation should be managed as a business and control programme, not as an isolated technology purchase.

Step 1: Define the decision problem

Specify the decision, users, timing, financial impact, acceptable error and required level of explanation. “Use AI in finance” is not a sufficiently defined objective.

Step 2: Assess process and data readiness

Map systems, data lineage, ownership, controls, manual workarounds and exceptions. Resolve high-impact data defects before building advanced models.

Step 3: Select the appropriate analytical method

Use a rule when rules are stable and transparent. Use statistical forecasting when relationships are interpretable and data is adequate. Use machine learning when nonlinear patterns or high-dimensional data justify additional complexity. Use generative AI for controlled language and knowledge tasks, not automatically for numerical truth.

Step 4: Build governance into design

Assign a business owner, technical owner, risk owner and approval authority. Define access, validation, monitoring, incident response and retirement criteria.

Step 5: Pilot and evaluate

Compare the proposed approach with the existing process or a suitable baseline. Measure accuracy, cycle time, cost, control effectiveness, user adoption and unintended consequences.

Step 6: Scale selectively

Scale only when benefits are demonstrated and operational support exists. Vendor due diligence, service-level agreements, data-location review, exit plans and business-continuity arrangements are essential.

A useful investment test compares the present value of expected benefits with implementation and operating costs. Benefits should include measurable savings, reduced losses, improved working capital and better decisions, while recognising that some benefits are uncertain and should be scenario-tested.

10. Illustrative Case: Meridian Components Ltd

Background

Meridian Components Ltd is a fictional mid-sized manufacturer supplying industrial equipment producers. It operates in three currencies, holds significant inventory and offers credit to distributors. Its monthly cash forecast is prepared manually in spreadsheets and is often inaccurate.

Problem

The treasury team receives bank data with delays, sales forecasts are not reconciled to order data, inventory information is updated inconsistently and overdue receivables are analysed only at month-end. Forecast error has caused unnecessary short-term borrowing and occasional missed opportunities to invest surplus cash.

Analysis

A diagnostic review identifies four root causes:

1. inconsistent customer and product identifiers across systems;
2. lack of ownership for forecast assumptions;
3. manual aggregation of receivables, inventory and payment schedules; and
4. no formal measurement of forecast bias or accuracy.

The company considers an AI-based cash forecast but first establishes a controlled data foundation. It standardises master data, connects bank and ERP feeds through secure interfaces and creates a daily reconciliation between cash transactions and the general ledger.

Approach and solution

Meridian introduces a layered forecasting process:

• a driver-based baseline using confirmed orders, contractual payment terms, payroll and supplier schedules;
• a statistical model for expected customer payment timing;
• scenario adjustments for foreign exchange, late payments and production disruption; and
• a human treasury review for exceptional events.

Forecast performance is assessed using MAE and bias. A rolling 13-week horizon is used, and the model is recalibrated quarterly or sooner if performance deteriorates. Access is restricted through role-based controls, and changes to assumptions are logged.

Outcome

After six months, Meridian reports shorter forecast preparation time, earlier identification of liquidity pressure and fewer unexplained differences between forecast and actual cash. Management does not eliminate judgement; instead, judgement becomes visible through documented overrides and scenario explanations. The company also rejects a proposal to permit an external generative-AI tool to process raw customer data because privacy and vendor-control reviews are incomplete.

Learning points

The case demonstrates that the value of AI depends on process discipline, data quality and governance. A simpler baseline with reliable data may outperform a complex model built on inconsistent records. It also shows that a forecast is a management control, not merely a statistical output.

11. Limitations and Critical Evaluation

Digital finance has several structural limitations. Historical data may not represent a changed economic regime. Models can fail during crises, mergers, supply shocks or regulatory changes. Automation can reduce informal knowledge and create concentration risk in a small number of platforms or specialists. Cloud and software vendors may introduce lock-in, opaque updates and cross-border data issues. Continuous measurement can create surveillance concerns and encourage managers to optimise visible KPIs rather than underlying value.

Cost-benefit analysis must include implementation, integration, training, cybersecurity, validation, monitoring, model retirement and opportunity costs. The appropriate comparison is not between an ideal AI system and a poor manual process, but between realistic alternatives under the organisation's constraints.

Professional judgement remains necessary because finance involves incomplete information, competing objectives, uncertain regulation and ethical trade-offs. The strongest digital finance function combines human accountability with machine-supported speed and consistency.

Contemporary Developments and Best Practices

Current developments include cloud-native ERP systems, event-driven architectures, embedded payments, tokenisation, sustainability-data analytics, privacy-enhancing computation, large language models and autonomous workflow agents. These technologies may improve access to information, but each requires careful assessment of reliability, control and accountability.

Best practices include:

• begin with material decisions rather than technology categories;
• maintain a single governed definition for critical financial measures;
• preserve data lineage from source transaction to reported output;
• use independent validation for material models;
• monitor drift, bias, errors and override patterns;
• keep humans accountable for consequential decisions;
• test cybersecurity and business continuity regularly;
• document vendor responsibilities and exit options; and
• stop or redesign systems that cannot meet required control standards.

Chapter Summary

Digital corporate finance integrates transaction systems, governed data, analytics, automation and decision processes. Financial models should be driver-based, transparent and linked across the income statement, balance sheet and cash flow. Business intelligence explains historical and current performance, while predictive and prescriptive analytics support future-oriented decisions. Forecasts should be evaluated out of sample using suitable accuracy and bias measures.

Automation improves speed and consistency when processes are stable and controls are designed into workflows. Generative AI can assist with language, analysis and knowledge work, but its outputs require grounding, verification and human accountability. Fintech and digital assets offer new payment, treasury and settlement possibilities while creating legal, custody, cyber and valuation risks.

Data quality, privacy, cybersecurity, bias and model risk are core finance responsibilities. Successful transformation begins with a defined decision problem, reliable data, proportionate technology, independent validation and continuous monitoring. Digital tools enhance finance only when they improve decisions without weakening control, ethics or professional judgement.

Key Terms

Digital corporate finance: The use of digital systems, data and analytics to conduct and govern corporate financial activities.

Business intelligence: Technologies and methods that report and analyse historical or current business performance.

Predictive analytics: Analysis that estimates future outcomes using historical data and explanatory variables.

Prescriptive analytics: Analysis that recommends actions while considering objectives and constraints.

Driver-based model: A financial model linked to operational factors that cause financial outcomes.

Scenario analysis: Evaluation of coherent alternative assumptions about future conditions.

Sensitivity analysis: Assessment of the effect of changing selected assumptions, usually while holding others constant.

Simulation: Generation of possible outcomes by repeatedly sampling uncertain inputs.

Robotic process automation: Software-based execution of repetitive, rule-based tasks.

Generative AI: AI that produces new text, code, summaries or other content from learned patterns.

Data lineage: The documented path of data from source through transformations to final output.

Master data: Core reference information, such as customers, products and suppliers, shared across systems.

Cash conversion cycle: The period between cash paid for operating resources and cash collected from customers.

Fintech: Technology-enabled innovation in financial products, services or processes.

Digital asset: A digitally represented asset, right or unit of value, often transferred or recorded electronically.

Model risk: Risk of adverse consequences from model errors, misuse, poor data or changed conditions.

Algorithmic bias: Systematic unfairness or distortion arising from data, design, variables or deployment.

Explainability: The extent to which a system's outputs and contributing factors can be understood and challenged.

Least privilege: Restricting access to only what a user or system requires for an authorised task.

Real-time finance: Financial information and controls updated frequently enough to support current decisions.

Review Questions

Short Answer Questions

1. Distinguish business intelligence from predictive analytics.
2. Why is data lineage important in digital corporate finance?
3. State two conditions under which robotic process automation is appropriate.
4. What is model risk?
5. Why must generative-AI outputs be independently reviewed in consequential finance activities?

Long Answer Questions

1. Explain how an integrated financial model links operational drivers, financial statements and cash flow.
2. Compare sensitivity analysis, scenario analysis and simulation, including their managerial uses.
3. Discuss the principal cybersecurity and privacy controls required for a digital finance function.
4. Evaluate the opportunities and risks of digital assets for corporate treasury.
5. Design a governance framework for a machine-learning model used in credit-risk assessment.

Analytical / Application Questions

1. A company has accurate sales data but unreliable customer payment dates. Propose a phased approach to improve its 13-week cash forecast, including data, modelling and control measures.
2. A fraud model reduces losses but rejects a disproportionately high share of transactions from one customer segment. Analyse possible causes and recommend validation and governance actions.
3. A finance director proposes replacing all monthly management reporting with real-time dashboards. Critically evaluate the proposal, considering decision relevance, control, short-termism, data quality and organisational behaviour.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a driver-based financial model?

A. It extrapolates total profit without identifying operational causes.

B. It links financial outcomes to operational variables such as volume, price and headcount.

C. It records only historical transactions in the general ledger.

D. It replaces all management judgement with a statistical algorithm.

Correct Answer: B. It links financial outcomes to operational variables such as volume, price and headcount.

Question 2

In the cash conversion cycle formula CCC = DIO + DSO − DPO, what does DPO represent?

A. Days payable outstanding

B. Days production output

C. Debt payment obligation

D. Deferred profit outcome

Correct Answer: A. Days payable outstanding

Question 3

Which measure is expressed in the same units as the forecasted outcome?

A. Mean absolute error

B. A probability threshold

C. A correlation coefficient

D. A percentage growth rate

Correct Answer: A. Mean absolute error

Question 4

What is the most appropriate reason to retain human review of a generative-AI output used in a regulatory submission?

A. Generative AI always operates too slowly for finance.

B. The output may be fluent but factually incorrect or incomplete.

C. Human review prevents all cybersecurity incidents.

D. Regulatory submissions never contain numerical information.

Correct Answer: B. The output may be fluent but factually incorrect or incomplete.

Question 5

Which control best reflects the principle of least privilege?

A. Giving every employee administrator rights for convenience.

B. Allowing access only to data and functions required for an authorised role.

C. Sharing one finance password among the treasury team.

D. Removing all access logs to protect privacy.

Correct Answer: B. Allowing access only to data and functions required for an authorised role.

Question 6

A model has a lower RMSE than another model but is highly unstable during economic shocks. What is the best conclusion?

A. It must always be selected because RMSE is the only relevant criterion.

B. It should be assessed further for stability, stress performance and decision consequences.

C. It should be used without monitoring because shocks are impossible to model.

D. It is necessarily unbiased.

Correct Answer: B. It should be assessed further for stability, stress performance and decision consequences.

Question 7

Which activity is most clearly a data-quality control?

A. Assigning ownership for critical data elements and monitoring exceptions.

B. Increasing the number of dashboard colours.

C. Allowing users to overwrite master data without logging.

D. Deleting records that do not fit an expected trend.

Correct Answer: A. Assigning ownership for critical data elements and monitoring exceptions.

Question 8

What is a major risk specific to holding digital assets in corporate treasury?

A. They cannot ever be transferred electronically.

B. Private-key loss and custody failure can impair access to the asset.

C. They eliminate all valuation uncertainty.

D. They are automatically free from sanctions requirements.

Correct Answer: B. Private-key loss and custody failure can impair access to the asset.

Question 9

Which approach is most appropriate when selecting an AI use case for finance?

A. Begin with a clearly defined material decision and assess data and control requirements.

B. Select the most complex available model regardless of the process.

C. Prioritise novelty over auditability and business value.

D. Deploy first and define accountability after implementation.

Correct Answer: A. Begin with a clearly defined material decision and assess data and control requirements.

Question 10

If EBIT = 100, T = 0.25, D&A = 20, Capex = 30 and ∆NWC = 10, what is FCFF using FCFF = EBIT × (1 − T) + D&A − Capex − ∆NWC?

A. 45

B. 55

C. 65

D. 75

Correct Answer: B. 55