Technology and analytics
Define the finance decision before choosing a dashboard, model or automation. Reconcile the input data, select an analytical method that fits the question, validate the result, and keep human accountability for the action.
Choose the analysis from the question
- What happened?
- Descriptive analysis summarizes reconciled historical data.
- What may happen?
- Predictive analysis estimates outcomes and tests forecast error.
- What should we do?
- Prescriptive analysis compares actions under objectives and constraints.
Diagnose causes separately. A dashboard correlation does not establish why an outcome occurred.
See the decision
A dashboard total that is too large
The ledger has 100 invoices totaling $250,000. Joining invoices to multiple shipment rows produces 125 rows and a dashboard sum of $312,500. The ledger is the trusted control total in this example.
USD reported total
- Ledger$250,000
- Incorrect join$312,500
Repair the grain of the join before analyzing sales. The $62,500 difference is a data-model error in this scenario, not revenue growth.
How to apply these ideas
1. Data quality
Check completeness, validity, consistency and timing.
Reconcile extracted records to a trusted control total. Inspect duplicate keys, missing values and inconsistent units before drawing conclusions. A polished chart cannot repair a join that double-counted invoices or omitted an entire business unit.
Watch for: Interpreting a trend before reconciling the underlying population.
2. Analytical question
Separate descriptive, diagnostic, predictive and prescriptive tasks.
A sales chart describes what happened; a driver investigation asks why; a forecast estimates what may happen; an optimization recommends an action under constraints. The decision and evidence determine the method, not the tool’s popularity.
Watch for: Presenting a historical correlation as proof of causation.
3. Model validation
Evaluate performance on data not used to fit the model.
Hold out a relevant period and compare with a simple baseline. Examine errors by segment and watch for leakage from future information. A small average error can conceal severe errors for the customers or products that matter most.
Watch for: Reporting training accuracy as expected future accuracy.
4. Automation controls
Assign access, exception handling and change approval.
Automate stable, defined steps with a clear audit trail. Monitor failures and preserve a safe escalation path. When an input schema or business rule changes, test the change before relying on the output for financial decisions.
Watch for: Automating an unreconciled process at higher speed.
Study the topic
Technology and analytics: explanations and applications
Choose finance information and analytical methods that fit a decision, then validate the data, model and operational controls. The FP&A application focuses on forecast evaluation.
By MiloPrep editorial team. Original teaching explanations and illustrative cases; official curriculum references are listed below.
Distinguish the transaction system, planning system and report
An accounting information system records financial transactions. ERP connects functions such as procurement, inventory, production and accounting. Enterprise performance management supports planning, consolidation and performance processes. A dashboard presents selected measures from these systems; it is not automatically the authoritative source for the underlying transaction.
Trace a reported number to its origin, transformations and calculation. Identify the grain of each table: one row per invoice, invoice line, shipment or customer. Joining a one-to-many table can multiply amounts even when every source record is correct. Reconcile both record counts and meaningful financial control totals before analyzing the result.
| Layer | Control question | Example evidence |
|---|---|---|
| Source | Which system and cut-off define the population? | Ledger control total and extraction time |
| Transformation | What was joined, filtered or converted? | Documented mapping and rejected-record log |
| Measure | What is the exact numerator and denominator? | Versioned definition and independent recalculation |
| Presentation | Can a filter hide a material segment? | Displayed scope, period and applied filters |
| Action | Who reviews and resolves exceptions? | Assigned owner and documented resolution |
Manage data quality through its life cycle
Define ownership for key data such as customer IDs, product hierarchies, exchange rates and account mappings. Validation should cover completeness, uniqueness, accuracy, consistency and timeliness. A field can be present and correctly formatted while still being wrong for the transaction. Reconcile it to an appropriate source rather than relying only on format checks.
Set rules for creation, authorized changes, retention and disposal. Access should follow the business need, with extra protection for sensitive records and privileged accounts. Preserve lineage so a reviewer can explain why a number changed after a source correction or mapping update. Governance is an ongoing operating responsibility, not a cleanup performed once before launch.
Match the analytical method to the question
Descriptive analysis summarizes what occurred; diagnostic analysis investigates possible drivers; predictive analysis estimates future outcomes; prescriptive analysis compares actions under objectives and constraints. A trend chart can reveal a change without explaining its cause. A regression can predict an association without showing that changing the predictor will cause the outcome.
Segment the question before selecting a model. A stable aggregate may hide losses in one product or repeated overforecasting in one region. For optimization, state the resource constraints and objective, such as maximizing contribution within available machine-hours. An answer is not useful if it violates the operating limits that make the decision difficult.
Separate fit, prediction and decision usefulness
Evaluate a model on observations it did not use for training. For time-series forecasting, preserve chronological order to avoid using future information. Compare against a simple baseline, such as last period or the same month last year. A sophisticated model that fails to outperform the baseline may add complexity without better decisions.
Choose an error measure appropriate to the business loss. Mean absolute error describes error size in original units; a signed mean can reveal directional bias but can cancel large opposite errors. Percentage error measures become unstable near zero actual values. Inspect errors by segment and time, and investigate whether the relationship changed after a policy or market shift.
Control automation before scaling it
Robotic process automation executes defined steps, so an unstable process can produce errors faster after automation. Specify inputs, business rules, exception handling, ownership and reconciliation. A failed item should have a clear status and recovery path, rather than silently disappearing or being processed twice when the job restarts.
A system change should move through defined requirements, testing, approval and deployment controls. Separate development access from production authority where practical. Test edge cases, permissions and rollback as well as the happy path. For models, document versions and monitor drift so a once-accurate forecast is not treated as reliable indefinitely.
Make the chart support the decision
Use a line chart for a time sequence, a bar chart for category comparison and a table when exact values matter. Label units, periods and the denominator. An axis truncated on a bar chart can exaggerate differences; percentages with changing populations can hide changes in volume. Show uncertainty when it materially affects the action.
A useful finance explanation states the result, the evidence, the limitation and the next check. “Revenue increased” is descriptive. “Revenue increased while receivable days worsened; inspect customer aging before changing the credit policy” connects the pattern to a testable decision without claiming a cause that has not been established.
Work it through
Worked example: A dashboard total that is too large
The ledger has 100 invoices totaling $250,000. Joining invoices to multiple shipment rows produces 125 rows and a dashboard sum of $312,500. The ledger is the trusted control total in this example.
| Check | Ledger | Dashboard join |
|---|---|---|
| Rows | 100 invoices | 125 joined rows |
| Amount | $250,000 | $312,500 |
| Difference to investigate | $0 | $62,500 overstatement |
- 1
One invoice may appear on more than one shipment row. Counting rows is no longer counting invoices.
- 2
Aggregate shipment data to the invoice key, or use a properly defined invoice measure, and reconcile the corrected amount to $250,000.
Repair the grain of the join before analyzing sales. The $62,500 difference is a data-model error in this scenario, not revenue growth.
Work through the alternatives
More worked cases: technology and analytics
Original teaching scenarios. Each case states its assumptions, shows the working and explains the limits of the conclusion.
Low average bias can coexist with large errors
Actual sales are 100, 120 and 80 units; forecasts are 90, 130 and 80. Define signed error as actual minus forecast.
| Period | Signed error | Absolute error |
|---|---|---|
| 1 | +10 | 10 |
| 2 | -10 | 10 |
| 3 | 0 | 0 |
| Mean | 0 | 6.67 |
- Signed errors sum to zero, so mean bias is zero.
- Absolute errors sum to 20. Mean absolute error = 20 ÷ 3 = 6.67 units. Cancellation in the signed mean does not mean perfect prediction.
What the result means: Report error magnitude alongside bias, and compare the model with a baseline using the same evaluation periods.
A valid join can produce the wrong financial total
Invoice A is $600 and has three shipment records. Invoice B is $400 and has one shipment record. Joining invoices to shipments repeats each invoice total on each matching shipment row.
| Invoice | Repeated rows | Naive sum | Correct invoice total |
|---|---|---|---|
| A | 3 | $1,800 | $600 |
| B | 1 | $400 | $400 |
| Total | 4 | $2,200 | $1,000 |
- The join describes invoice-shipment relationships, but invoice total is not additive at that grain.
- Aggregate shipments to the invoice key before joining, or define a measure at invoice grain and reconcile to $1,000. A blanket SUM(DISTINCT amount) is unsafe because different invoices can share the same amount.
What the result means: Correct row relationships do not guarantee a correct sum. Establish the grain of the measure before using it in a chart.
Practice this topic
Check your understanding
Try each short application before opening its explanation. These are original learning checks, not official exam questions or a score prediction.
1. A forecast uses the final collection date to predict whether an invoice will be late at issuance. What is wrong?
Show answer and explanation
It uses information unavailable at the prediction time.
This is data leakage. Evaluate the model using only information that would have been known when the decision was made.
2. Absolute errors are 4, 8, 6 and 2 units. What is MAE?
Show answer and explanation
5 units.
The sum is 20 across four observations. This gives average magnitude, not the direction of bias or the size of the worst error.
3. Late deliveries and overtime rise together. Does that prove overtime causes late deliveries?
Show answer and explanation
No.
A common factor such as unexpected demand or machine failure could increase both. Investigate the process and timing before claiming causation.
Questions about technology and analytics
- Does a high R-squared prove a useful forecast?
- No. It measures in-sample fit under the model, not causal validity or future performance. Check assumptions, out-of-sample errors and stability of the relationship.
- What should an exception report contain?
- The failed rule, affected item, owner, review time and resolution evidence. An unassigned list of exceptions is not a completed control.
Sources and scope
Curriculum mapping: Part 1, Section F, Technology and Analytics. The official outline assigns 15% to the entire domain, not to this individual technique. These are selected study notes, not complete curriculum coverage.
Official curriculum references establish scope and expected skills. The explanations, diagrams and numerical cases are independent teaching material from MiloPrep, not official questions or source-validated solutions. Curriculum references checked 2026-10-10.
- IMA Content Specification Outlines (effective September 2024) ↗
Supports: Part 1, Section F, beginning on printed page 9. Domain names, weights and curriculum boundaries.
- IMA Learning Outcome Statements (effective September 2024) ↗
Supports: Part 1, Section F, beginning on printed page 14. The calculations, analysis and professional judgments candidates are expected to perform.
Independent US CMA study material. Not affiliated with or endorsed by IMA. This covers management accounting, not medical-assistant certification or the separate ICMAI qualification.