Technology and Analytics

An accounting information system collects, processes, stores and reports financial and non-financial data. It is organized into transaction cycles that all feed the general ledger.

• ERP integrates transaction processing for all functions (finance, purchasing, production, sales, HR) in one database, so an event is recorded once and is visible everywhere. Benefits: consistent real-time data, standardized processes, fewer reconciliations. Risks: high cost, long implementations, business-process change, and a single point of failure.
• EPM (enterprise or corporate performance management) sits on top of ERP and other sources to support planning, budgeting, forecasting, consolidation, and performance reporting (scorecards, dashboards).
• OLTP (online transaction processing) systems such as ERP handle many small updates quickly; OLAP (online analytical processing) supports multidimensional analysis of historical data (slice, dice, drill down).

ETL (extract, transform, load) moves data from source systems into a warehouse: extract it, clean and standardize it, then load it. Data lakes often use ELT, transforming only when the data is used.

Data governance sets who owns data, the policies and standards that apply to it, and how quality, security and compliance are monitored. Data owners (business) decide access and use; data stewards maintain quality; IT custodians operate the systems.

1. Capture: data is created or acquired (entry, devices, purchase).
2. Maintenance: cleaning, enriching and moving data so it is usable.
3. Synthesis: creating new values from existing data (e.g. a credit score).
4. Usage: applying data in operations and decisions.
5. Publication: sending data outside where it is used (reports, customer statements).
6. Archival: storing data no longer in active use, retrievable if needed.
7. Purging: securely deleting data at the end of its retention period.

Data quality dimensions: accuracy, completeness, consistency, timeliness, validity and uniqueness.

COBIT (ISACA) separates governance (the board evaluates, directs and monitors) from management (executives plan, build, run and monitor). Its 40 objectives fall into five domains:

• GDPR (EU): applies to anyone processing EU residents' data; lawful basis, consent, data minimization, rights to access and erasure, breach notification within 72 hours, large fines.
• CCPA/CPRA (California): rights to know, delete and opt out of the sale or sharing of personal information.
• HIPAA (US health data) and GLBA (financial institutions) protect sector-specific data.
• Good practice: collect only what is needed, anonymize or pseudonymize data used for analytics, restrict access, and set retention limits.

flowchart LR
  P[Planning and feasibility] --> A[Systems analysis]
  A --> D[Design]
  D --> B[Development]
  B --> T[Testing]
  T --> I[Implementation]
  I --> M[Operation and maintenance]
  M -. new needs .-> P

1. Planning / feasibility: define the problem; assess economic, technical, operational and schedule feasibility.
2. Systems analysis: study the current system and gather user requirements.
3. Design: specify outputs, inputs, data, processes and controls (conceptual, then detailed).
4. Development: build or buy and configure the software.
5. Testing: unit, integration, system and user-acceptance testing.
6. Implementation: training, data conversion, go-live, post-implementation review.
7. Operation and maintenance: fixes and enhancements under change management.

Waterfall completes each phase before the next and suits stable, well-understood requirements. Agile (e.g. Scrum) delivers working software in short iterations with continuous user feedback, and suits changing requirements.

• Robotic process automation: software bots mimic user actions on rules-based, repetitive, high-volume tasks with structured inputs (invoice entry, bank reconciliation matching, report distribution). Bots follow rules; they do not learn. Risks: bots break when screens change; bot credentials need access control; errors repeat at scale.
• Artificial intelligence / machine learning: systems that learn patterns from data. Supervised learning uses labeled outcomes (fraud / not fraud) to predict; unsupervised learning finds structure without labels (clustering customers). NLP reads text (contracts, emails); generative AI drafts text and code.
• AI risks: biased or poor training data, lack of explainability ("black box"), overreliance, and data privacy. Human review remains a key control.

Deployment: public (shared provider infrastructure), private (dedicated to one organization), hybrid (mix). Benefits: scalability, pay-as-you-go (capex becomes opex), fast deployment. Risks: vendor dependence, data location and privacy, availability, and the need to review the provider's controls (a SOC 1 or SOC 2 report).

• Blockchain: a distributed ledger shared by many nodes; blocks are linked by cryptographic hashes and added by consensus, so recorded transactions are practically immutable. Public (permissionless) vs private/permissioned (known participants, e.g. a supply-chain consortium).
• Smart contracts: code on a blockchain that executes automatically when conditions are met (release payment when a shipment is confirmed).
• Limits: immutable does not mean accurate (garbage in stays in); scalability, energy use (proof of work), regulation, and the "oracle" problem of getting reliable external data.

• Big data: volume, velocity, variety, veracity (and value).
• Data mining techniques: classification (assign to known categories, e.g. fraud/not fraud), clustering (find natural groups without labels), regression (predict a number), association rules (items bought together), anomaly detection (outliers such as unusual journal entries).
• Benford's law: in many natural datasets the first digit is 1 about 30% of the time and 9 under 5%; departures can flag fabricated numbers.

• Correlation is not causation: a third factor may drive both variables.
• Multicollinearity: independent variables highly correlated with each other make individual coefficients unreliable.
• Extrapolation: predictions outside the range of the data are unreliable.
• A high R² does not prove the model is right; check the logic, the residuals and the significance of each variable.

• Sensitivity (what-if) analysis changes one input at a time to see which drivers move the result most.
• Scenario analysis changes several inputs together (best, base, worst case).
• Monte Carlo simulation draws every uncertain input from a probability distribution thousands of times, producing a distribution of outcomes: the mean, the spread and the probability of a loss.

• Start bar-chart axes at zero; truncated axes exaggerate differences.
• Remove clutter (3-D effects, heavy gridlines); highlight the message; label directly.
• Know the audience: executives want a few KPIs on a dashboard with drill-down; analysts want detail.