Blog · AI and self-service analytics
The four types of analytics answer different business questions. Compare a summary, a driver bridge, a forecast check and an action comparison using the same sales workbook, with explicit assumptions and reconciliations.
The descriptive predictive prescriptive distinction separates a summary of what happened, an estimate of what may happen and a comparison of what to do next. Diagnostic analytics adds the investigation of why a result changed. Choose the job from the decision facing the business, then supply the records and assumptions that job needs.
The distinction is about the question, not the appearance of the report. A workbook can contain historical totals, an explanation of a variance and a forward estimate on adjacent sheets. Those are different analytical jobs even though they use the same file.
IBM's explanation of predictive analytics distinguishes all four types and describes prediction as using historical data and statistical methods to estimate future outcomes.
| Type | Question | Typical output | Additional evidence needed |
|---|---|---|---|
| Descriptive | What happened? | Checked revenue and margin totals | Complete historical records and definitions |
| Diagnostic | What drove the change? | A reconciled bridge and tested explanations | Comparable records, relevant drivers and business evidence |
| Predictive | What may happen? | A forecast with an error assessment | A stated method, sufficient history and validation |
| Prescriptive | Which action best serves the objective? | A comparison of feasible actions | Response assumptions, costs, limits and an accountable decision owner |
A good descriptive answer may be enough for a reporting request. A decision to change prices needs more: expected demand response, the effect on profit and permission to carry out the action. Prescriptive analytics adds that recommendation step, using objectives, constraints and tradeoffs to compare choices.
These questions also guide the use of AI analytics: define the job and establish the source evidence before asking for an explanation.
Covirage is not affiliated with IBM.
Consider an illustrative US equipment supplier selling two product lines. The workbook includes completed units, realized unit price and direct unit cost for January and February 2026. There are no returns or additional variable costs in this simplified example. The unit cost remains unchanged between months.
| Month | Product | Units | Unit price (USD) | Direct unit cost (USD) | Revenue (USD) | Direct cost (USD) |
|---|---|---|---|---|---|---|
| January | Core | 100 | 1,000 | 650 | 100,000 | 65,000 |
| January | Premium | 40 | 2,000 | 1,300 | 80,000 | 52,000 |
| February | Core | 120 | 950 | 650 | 114,000 | 78,000 |
| February | Premium | 30 | 2,000 | 1,300 | 60,000 | 39,000 |
Revenue on each row is units multiplied by unit price. Direct cost is units multiplied by direct unit cost. All further results below come from these rows or from explicitly labeled action assumptions.
Before comparing months, match the workbook's revenue and costs to the selected source records. Confirm that the same products, statuses and posting dates appear in both periods. The four types cannot rescue an incomplete or inconsistent population.
Revenue = Sum of Units × Realized unit price
Gross profit in this example = Revenue − Direct cost
| Measure | January calculation | January result | February calculation | February result |
|---|---|---|---|---|
| Units | 100 + 40 | 140 | 120 + 30 | 150 |
| Revenue (USD) | 100,000 + 80,000 | 180,000 | 114,000 + 60,000 | 174,000 |
| Direct cost (USD) | 65,000 + 52,000 | 117,000 | 78,000 + 39,000 | 117,000 |
| Gross profit (USD) | 180,000 − 117,000 | 63,000 | 174,000 − 117,000 | 57,000 |
Revenue fell by 174,000 − 180,000 = −$6,000, or −6,000 / 180,000 = −3.3%. Gross profit also fell by $6,000 because total direct cost was unchanged. Percentage results throughout this example are rounded to one decimal place.
This answers the historical question. It does not explain the customer decisions behind the movement. For the narrower boundary between reporting and forecasting, see descriptive versus predictive analytics.
Use a sequential bridge: change units at January prices, then change prices on February units. This convention assigns the interaction between volume and price to the price step. Other conventions can also work, but the report must name its convention and reconcile.
| Bridge step | Calculation | Revenue effect (USD) |
|---|---|---|
| Additional Core units at January price | (120 − 100) × 1,000 | 20,000 |
| Fewer Premium units at January price | (30 − 40) × 2,000 | −20,000 |
| Lower Core price on February units | 120 × (950 − 1,000) | −6,000 |
| Premium price unchanged | 30 × (2,000 − 2,000) | 0 |
| Total change | 20,000 − 20,000 − 6,000 + 0 | −6,000 |
The proof is 180,000 − 6,000 = $174,000, matching February revenue. Higher Core volume exactly offset lower Premium volume at January prices; the remaining decline is the lower Core price.
That is an arithmetic explanation. It does not establish why Premium demand fell or whether the Core discount caused additional sales. Sales needs customer and pricing evidence for those claims. Descriptive versus diagnostic analytics covers that boundary separately.
A simple baseline carries January's product units and prices into February. It predicts $180,000, compared with the actual $174,000 above. Its absolute revenue error is |180,000 − 174,000| = $6,000; dividing by actual revenue gives 6,000 / 174,000 = 3.4%.
The baseline was wrong at the product level even though some errors offset in the total. Keep product results available when reviewing accuracy. A forecast that misses the mix can lead to poor purchasing decisions despite an acceptable company total.
Carrying February forward would give a March baseline of 120 × 950 + 30 × 2,000 = $174,000. That is a transparent reference case, not a validated demand forecast. These two months alone do not establish a reliable seasonal pattern or uncertainty range. The predictive analytics guide covers prediction methods and validation in more depth.
Now suppose the commercial owner wants to maximize next month's gross profit within approved pricing options. The owner supplies a new assumption: restoring the Core price would reduce Core demand to 110 units, while Premium demand would remain at 30 units. This demand response is an assumption to test, not a conclusion from the historical rows.
| Action assumption | Core units | Core price (USD) | Premium units | Premium price (USD) |
|---|---|---|---|---|
| Keep February pricing | 120 | 950 | 30 | 2,000 |
| Restore Core price | 110 | 1,000 | 30 | 2,000 |
Apply the direct unit costs of $650 and $1,300 from the source table:
| Action | Revenue calculation (USD) | Direct cost calculation (USD) | Gross profit (USD) |
|---|---|---|---|
| Keep February pricing | 120 × 950 + 30 × 2,000 = 174,000 | 120 × 650 + 30 × 1,300 = 117,000 | 57,000 |
| Restore Core price | 110 × 1,000 + 30 × 2,000 = 170,000 | 110 × 650 + 30 × 1,300 = 110,500 | 59,500 |
Restoring price produces $2,500 more gross profit under the supplied assumptions: 59,500 − 57,000. It also produces lower revenue. That difference shows why the objective must be explicit. The prescriptive analytics guide covers action comparison separately from this choice of analysis type.
The decision owner still needs to confirm contractual limits, customer response evidence and whether other costs change. A computed ranking does not authorize a price change.
Bring a sales workbook with periods, products, units, prices and matching costs, plus any separate forecast or action assumptions you want to compare. Covirage maps the columns, its tools compute and check the supplied figures, and the external AI model explains the results with every figure traced to its rows. Create a free Covirage account and bring a sample of your own data.
See self-service analytics for the supporting approach to working from business files.
Predictive analytics estimates an outcome under stated conditions. Prescriptive analytics compares available actions and their expected consequences against an objective and constraints. A sales forecast estimates demand; a pricing decision also needs margin, customer response assumptions and the authority to change prices.
Diagnostic analytics can isolate the arithmetic drivers of a change and test possible explanations. It does not automatically prove the underlying cause. A price-volume bridge explains how revenue moved, while evidence about why customers changed their buying behavior requires further investigation.
No. A reconciled summary may fully answer a month-end reporting question. Add a driver investigation, forecast review or action comparison when the decision needs it. More elaborate analysis is useful only when its additional evidence changes the decision or reduces uncertainty.
Yes, if its historical records are consistent and sufficiently detailed. Predictive work also needs a stated forecasting method and a way to test it. Prescriptive work needs candidate actions, costs, constraints and assumptions about how business outcomes respond to those actions.