Aidar Akmaev
Disclosure: NOXI publishes this guide and is one of the tools it mentions; the method, labeling table and worked example work in any presentation software.
A slide says that next quarter's demand will be 12,483 requests. The figure looks exact because it has five digits. But perhaps it comes from a simple model using an uncertain conversion assumption and an incomplete pipeline. The calculation may be correct while the presentation gives the audience too much confidence in the outcome.
Quick answer: Present forecast uncertainty by showing what each number means, what it depends on and how it should be used. Label every figure as actual, forecast, target or scenario, and mark where observed data ends. Explain the basis of a range before giving it a statistical label, round to the precision the evidence supports, and date each forecast so you can show why it changed.
These habits let the audience evaluate the evidence without turning an uncertain estimate into a promise.
In short
- Name the numberSay whether it is a forecast, a target, a scenario or a projection, and state its conditions.
- Actual vs projectedMark the last observed period with a clear boundary and direct labels.
- RangesExplain what varies and how the limits were calculated before any statistical label.
- PrecisionRound headline figures to what the evidence supports; keep exact values in the appendix.
- RevisionsDate every forecast, keep the previous one and explain why it changed.
Name the number before discussing its size
Organizations sometimes use "forecast," "projection" and "plan" interchangeably. Define your terms in the deck so a reader can distinguish what is expected from what is desired or merely being explored.
A useful working distinction is:
- A forecast estimates a future outcome using available evidence and assumptions.
- A target states the outcome the organization wants to achieve.
- A scenario describes what could happen under a specified set of conditions.
- A projection calculates a future path from stated assumptions; explain whether it represents the team's current expectation.
The words alone are insufficient. Forecasts can themselves be conditional. The Bank of England's January 2026 Forecast Evaluation Report explains that its projections take certain input paths as given and may not correspond to the committee's single most likely expectation. Describe the conditions behind your number rather than relying on the label to do all the work. Bank of England: Forecast Evaluation Report — January 2026 (published 23 January 2026; checked 4 October 2026).
Let the verb carry the distinction: "we expect about 12,500" for a forecast, "we aim for 13,000" for a target, "could reach 15,000 if the campaign launches" for a scenario. If you use probability words such as "likely," define them once in the deck; otherwise each listener supplies a number of their own. Full-sentence headlines, as in our guide to consulting-style slides, make the verb hard to miss.
Draw an unmistakable boundary between actual and projected data
When a chart combines past results and future estimates, mark the last observed period. Use a vertical boundary, a clearly labeled projected area or a change in line style with direct labels. Do not depend solely on a subtle change of color. In a bar chart, give forecast periods lighter or outlined bars with a direct label; a legend alone is easy to miss.
Check partial periods carefully. A month-to-date count is an observed result for part of a month; the full-month estimate includes future activity. Joining them in a continuous series can make the estimate look like a completed result.
State units and coverage near the chart. Does "requests" include reopened cases? Does the measure cover every channel? Was the history revised? Those definitions may affect interpretation more than an additional decimal place.
Explain the range's origin
A range could describe variation across scenarios, sensitivity to one assumption, uncertainty estimated by a statistical model or simply a judgment supplied by the forecast owner. These are not interchangeable.
If low and high values come from changing one input, label the display accordingly: "Sensitivity to conversion assumption: 8–12%, with other inputs unchanged." Do not call it a "95% confidence interval" unless a qualified analysis supports that interpretation.
In regression, NIST distinguishes an interval for an estimated relationship from a prediction interval for a new observation; the latter includes additional variability. The relevant interval depends on the quantity and model. Confirm the analyst's terminology rather than relabeling a band while designing the slide. NIST: Prediction and prediction uncertainty.
Before placing a range on a slide, obtain four answers from its owner: what varies, how the limits were calculated, which assumptions remain fixed and whether any probability interpretation is justified. If those answers are unavailable, describe the range honestly as illustrative or remove it until its basis is clear.
Match the visual to the source of the range
Then choose a chart that shows that kind of uncertainty; the labeling table below gives the visual for each. Do not shade the area between scenario lines: a filled band suggests that every value in between has been assessed. When several inputs vary, a tornado chart, with one bar per assumption sorted by its effect, shows which one matters most. If the decision depends on the downside, label the low end with its value instead of leaving the audience to read it off a shaded edge.
Forecast labeling table: how to show each kind of number
Use it to build or review any forecast slide. The example wording uses the fictional service team.
| Type of number | How to show it on the slide | Example wording | Avoid |
|---|---|---|---|
| Actual (final) | Solid line or solid bars up to the last observed period | "Actual requests through 30 September 2026" | Carrying the solid style into future periods |
| Actual (provisional or partial) | Same style as actuals, with a marker or note on the affected period | "September: provisional, as of 2 October"; "October to date (1–15 Oct)" | Joining a month-to-date count to full-month values |
| Forecast (central estimate) | Dashed line or lighter bars after a vertical boundary, labeled directly | "Forecast prepared 2 October 2026: about 12,500" | Five-digit precision; "will" in the headline |
| Target or budget | A horizontal reference line or marker, not a continuation of the data | "Capacity target: 13,000" | Drawing the target as if it were a forecast path |
| Scenario | A separate line or column named after its condition | "Campaign scenario: about 15,000 if the campaign launches on schedule" | Shading between scenarios; a headline that drops the condition |
| Projection | Dotted line or a note that states the rule used | "About 13,800 if September's daily rate continued" | Presenting a mechanical extension as the team's expectation |
| Sensitivity range | Range bar or text, with the varied input named | "11,500–13,500 if conversion is 8–12%, other inputs unchanged" | Calling it a confidence interval |
| Model-based interval | Shaded band around the central line, with type and level in the label | "80% prediction interval from the demand model" | A band with no stated type, level or source |
| Judgment range | A written range with the owner's role | "Operations lead's estimate: 11,000–14,000" | A smooth band that looks computed |
| Revised forecast | The new figure with the previous one beside it, both dated | "About 12,500 (was about 12,000 on 1 September)" | Replacing the old figure without a note |
If one chart needs more than three of these rows, split it: actuals and the forecast on one slide, the target and scenarios on the next.
Use precision that serves the decision
Keep the approved calculation and its original values in the source material. On the main slide, choose a display precision that matches the evidence and the audience's task.
For the fictional demand estimate, "about 12,500 requests" may communicate better than "12,483" if the model cannot support that apparent exactness. But a contractual threshold, a reconciliation check or a technical tolerance may need exact values. Rounding is a communication choice, not permission to alter the analysis.
A practical default for headline forecasts is two or three significant figures, such as "about 4.2 million" rather than "4,213,877." Round a range to the same step as its central figure: 11,500–13,500, not 11,483–13,483. Percentages follow the same rule: a growth forecast of 23.7% shown beside a range of 18–29% sends two different messages, while "about 24%" matches the range.
Apply it consistently. A total rounded to 12,500 and components shown to the nearest unit may look inconsistent. State when figures are rounded and preserve the underlying values for anyone who needs to reproduce the calculation.
Avoid smooth decorative bands around a forecast just because they look professional. A widening fan implies something about uncertainty over time. It needs a defensible basis and an explanation; a graphic designer or language model should not invent it.
Put limitations beside the claim they qualify
The audience should encounter a material condition when it sees the estimate. "Expected demand: approximately 12,500 requests, assuming the campaign launches as scheduled" is more informative than placing the assumption in tiny text on the final source slide.
Distinguish several kinds of uncertainty. Incomplete observations concern the data. An unconfirmed launch date concerns a future event. A simplified relationship between campaign activity and demand concerns the model. Each uncertainty has a different route to resolution.
Government Analysis Function guidance recommends explaining sources and limitations, showing possible outcomes and clarifying what the evidence can and cannot support. Its communication principles are useful here, although the 2018 page is marked as awaiting updates. Government Analysis Function: Communicating quality, uncertainty and change (published 17 December 2018; checked 4 October 2026).
Do not make every slide a list of caveats. Prioritize the limitations that change the audience's interpretation or action. Put detailed methodology in supporting material and make it easy to find.
Explain revisions without rewriting history
Attach an "as of" date to the forecast and identify its source version. When the number changes, retain the previous forecast in the working record and explain the revision.
A short revision note might say: "The estimate increased because approved campaign coverage expanded. Historical demand was unchanged; the conversion assumption remains provisional." That distinguishes new information from a correction or a changed method.
If you compare the revised forecast with actual results, use comparable definitions and periods. Do not present a favorable comparison against an old forecast while quietly using a new forecast elsewhere.
On the slide itself, a compact form works: "About 12,500 (was about 12,000 on 1 September)." If the forecast has been issued several times, a small table of earlier forecasts beside the outcomes helps the audience judge how much weight to give the current one. For results against plan, see our QBR presentation guide.
Worked example: a forecast section, slide by slide
The same sequence works as the forecast page of a board meeting presentation or an operations review.
| # | Slide | The question it answers | What goes on it (fictional service team) |
|---|---|---|---|
| 1 | Forecast summary | What do we expect, and how firm is it? | "We expect about 12,500 requests in Q4, or 11,500–13,500 if conversion is 8–12%" |
| 2 | Actual vs forecast | Where does the evidence end? | Solid "Actual requests through 30 September 2026", dashed "Forecast prepared 2 October 2026" after a vertical line; September flagged as provisional |
| 3 | What it depends on | Which assumptions drive the number? | About 7,500 requests from existing customers plus 50,000 leads at 10% conversion (provisional) |
| 4 | The range and its basis | How wide is it, and why? | Range bar: "Sensitivity to conversion, 8–12%, other inputs unchanged"; not a statistical interval |
| 5 | Target and scenario | How does it compare with what we want and what could happen? | Target of 13,000 as a reference line; "Campaign scenario: about 15,000 if the campaign launches on schedule (75,000 leads)" |
| 6 | What changed | Why is it different from last month? | "Up from about 12,000 (1 September): the pipeline grew from 45,000 to 50,000 leads. History unchanged" |
| 7 | How to use it | What should the audience do with the number? | "Staff to the 13,000 target; decide on surge cover when the campaign date is confirmed" |
| A | Appendix | Can a specialist reproduce it? | 12,483 = 7,483 + 50,000 × 10%; data cutoff; what counts as a request; version and owner |
The 15,000 scenario never appears without its condition, the target is never drawn as a forecast, the range is named after the input that varies, and the exact 12,483 sits in the appendix for anyone checking the model. In a short update, slides 3 to 6 can share one slide.
Review AI-generated slides for lost qualifications
When drafting with NOXI, supply the approved outputs, the meaning of each number, the observation cutoff, the forecast date and the required limitations. Review the headline as carefully as the chart. "Could reach 15,000 under the campaign scenario" must not become "Will reach 15,000."
Whichever tool drafts the slides, check that no probability, confidence band or extra precision has appeared that the source does not contain. NOXI's research-citation guide can support source tracing; its annual-plan and budget guide addresses the wider planning structure.
Ask a final reviewer to identify what is observed, what is estimated, what is conditional and what is a target. If they cannot tell, improve the labels and explanation before delivery. The audience needs a forecast it can interpret and revisit as new evidence arrives.
3 AI prompts for forecast slides
Replace the brackets. Each prompt works in NOXI after you attach your files, and the constraints work in any AI tool. For more, see our presentation prompts guide.
Using only the attached [forecast workbook] and [assumptions note], build a [7]-slide forecast section: summary, actual vs forecast, drivers, the range and its basis, target and scenario, what changed since [1 September], and an appendix with exact values. Label every figure as actual, forecast, target, scenario or projection. Label the chart "Actual through [30 September 2026]" and "Forecast prepared [2 October 2026]". Describe the range as [sensitivity to conversion, 8–12%], not as a confidence interval. Round headline figures to [the nearest 100]. Add no number, probability or range that is not in my files.
Use the attached [company template / last quarter's forecast deck] as the style reference and build the forecast section from my files in that style. Show actuals in one color and the forecast in a lighter shade of it, labeled directly. Show targets as a labeled reference line, never as a data series. Keep each condition next to the number it qualifies, not in a footnote.
On this slide, change the headline to "[Campaign scenario]: demand could reach about [15,000] if [the campaign launches on schedule]" and add our central forecast, [about 12,500], beside it. Rename the range "[Sensitivity to conversion, 8–12%, other inputs unchanged]" and remove the words "confidence interval". Keep every number as it is.
How to build forecast slides in NOXI from your approved files
NOXI builds the presentation around a forecast your team has already approved: your workbook and notes supply the content, and your company template or last cycle's deck sets the style. Our guide to AI presentations in your own template compares the routes. NOXI is free; uploading files needs a free account, and your files are not used to train AI models.
Step 1: Gather the approved inputs
Export the forecast workbook or CSV, and write a one-page assumptions note in Word: what each number is, the data cutoff, the forecast date, the basis of the range, the conditions and the reason for any revision.
Step 2: Set the style with Match reference
Choose Match reference and attach your company PowerPoint template, last cycle's deck as a PPTX or PDF, or a screenshot of a slide you like. NOXI designs the new deck in that style: colors, typography, layouts and visual rhythm.
Step 3: Upload the files and review the outline
Attach the workbook and the note with Prompt 1. When you upload your own files, NOXI works from your files. Review the outline first, then generate and watch each slide being built.
Step 4: Fix labels and headlines with AI
Click the chart or a headline and tell the AI what to change, as in Prompt 3, or ask for a deck-wide edit such as "round every headline forecast to the nearest 100". On desktop, sketch on the slide, for example an arrow where actual data ends, and the AI follows your drawing.
Step 5: Check every number with the forecast owner
Tick off each figure against the workbook; edit text on the slide and correct table and chart numbers in NOXI. Invite the forecast owner by email as an editor, and restore an earlier version from version history if needed.
Step 6: Export, share and present
Download an editable PowerPoint (.pptx) with real text boxes and embedded fonts, or a PDF for the pre-read, free and with no watermark. Share a view-only link or present full-screen. Translate a deck creates a separate Arabic version with right-to-left slides.
Frequently asked questions
- How do you present forecast uncertainty in a presentation?
- Label every figure as actual, forecast, target or scenario, mark where observed data ends, and show a range only with its basis stated: what varies and how the limits were set. Round headline figures to the precision the evidence supports, put the main condition beside the number and date the forecast. Keep the method and exact values in an appendix.
- What is the difference between a forecast, a target and a scenario?
- A forecast estimates what is expected to happen given the evidence and assumptions; a target states what the organization wants to achieve; a scenario describes what could happen if specified conditions hold. They answer different questions, so label each one and let the verb show which it is: "we expect", "we aim for" or "could reach X if".
- How do you show actual vs forecast on a chart?
- Mark the last observed period with a vertical line or a labeled boundary, change the line style for the forecast and label both parts directly, for example "Actual through 30 September" and "Forecast prepared 2 October". Flag provisional or month-to-date data separately, and do not rely on a change of color alone.
- Should I show a confidence interval on a forecast slide?
- Only if the analyst's model produced that interval and you can name it correctly. A spread from changing one assumption is a sensitivity range, a spread across scenarios is a scenario range, and an owner's estimate is a judgment range. For a single new observation, NIST notes that a prediction interval is wider than the confidence interval for the regression function.
- How do you avoid false precision in forecasts?
- Show only as many digits as the evidence supports, which for a headline forecast is often two or three significant figures: "about 12,500", not "12,483". Round the range to the same step, say that figures are rounded and keep exact values in the appendix. Exact figures still belong on slides about contractual thresholds, reconciliations and technical tolerances.
- Can AI build a forecast presentation?
- Yes, for the presentation layer; the forecast itself comes from your model and its owner. Upload the approved workbook and assumptions note to NOXI, use Match reference to follow your company template, fix labels and headlines by chat and export an editable PowerPoint. Then check that no probability, range or extra precision appears that your files do not contain.
Sources
All sources were checked on 4 October 2026. The service-team examples are fictional.
- [1] Projections that take certain input paths as given and may not match the single most likely expectation — Bank of England, Forecast Evaluation Report — January 2026 (published 23 January 2026).
- [2] Interval for an estimated relationship vs prediction interval for a new observation; the prediction interval is wider — NIST/SEMATECH e-Handbook of Statistical Methods, Prediction and prediction uncertainty (no publication date shown).
- [3] Explaining sources and limitations, showing possible outcomes, stating what the evidence can and cannot support — Government Analysis Function, Communicating quality, uncertainty and change (published 17 December 2018; marked as awaiting updates).
Build your next forecast presentation in NOXI, free: attach your approved workbook and notes, match your company template, refine labels with AI and export an editable PowerPoint or PDF.
