Pitch a Data-Led Commercial Without Turning the Animation Into Extra Evidence

Every figure in a treatment can be correct and the passage can still pitch a claim nobody approved. The values live in the table. The claim lives somewhere else: in the bar lengths, the axis range, the order two series appear on screen, the ease on a transition. Copy the numbers faithfully and a bar that grows sixfold can still be the loudest unsupported sentence in the frame.

The fix isn't a disclaimer in the corner. It's deciding what the graphic is allowed to carry before deciding how it moves. One supported sentence, one visual relationship held stable, one job for the motion. The rest of this piece is a way of doing that inside a treatment document, which is usually written before the data is final.

Write the supported sentence first

Before the first storyboard panel, write one sentence with four slots filled in: what is being compared, for whom, over which period, and in what units. Not a mood, not a category of chart — a comparison, stated plainly enough that a lawyer and an art director could both sign it.

Here is a fixture, invented for this article to make the arithmetic checkable. Two fictional categories, Kestrel and Meridian. The recorded figure is monthly active accounts, in thousands, for January through June of a fictional year. No real brand, dataset, or measured result appears here, and nothing in this exercise has been shown to an audience.

Three sentences are true of that table, and they are not the same finding.

1. Meridian's figure sat 20 thousand above Kestrel's in every month recorded.
2. Kestrel's figure rose 25 percent across the period; Meridian's rose about 17 percent.
3. Both rose by 2 thousand per month.

The first is about a gap that never moves. The second is about relative growth, and it favors the smaller category. The third is about rate, and the rates were identical. A graphic built for sentence 2 will look like a different and more dramatic story than one built for sentence 1, and both can be built from the same six pairs of numbers. So pick one, and say in the treatment which comparison you picked, in words a viewer would recognize.

Then separate what was observed from what you think it means. "Both rose by 2 thousand per month" is in the table. "Meridian's lead is holding" is a reading of the table. "The campaign is working" is not in the table at all and needs a different kind of support than a chart.

Treatments get written before data is locked, and that's normal. What matters is that the unverified slot stays visibly unverified instead of being filled with a number that sounds persuasive. A treatment page can carry a small block like this:

CLAIM     [pending: one sentence, units, population, period]
SOURCE    [pending: owner, dataset, date]
VISUAL    [relationship, baseline, axis range]
MOTION    [what the movement does, and what it must not imply]
TEXT      [one or two plain sentences, plus qualifications]

Filling those slots is the work. The storyboard is downstream of them.

Choose the relationship the graphic will carry

A bar chart's meaning is a promise: the length of the bar is proportional to the value. That promise holds only from a zero baseline. Break the baseline and the length is no longer representing the quantity — and length is the thing a viewer reads first.

Put the fixture's bars on a baseline of 38 and watch what happens. Kestrel's January bar is 2 units tall; its June bar is 12. The figure rose by a quarter. The bar grew six times. Meridian's bars run 22 and 32, about 1.45 times. On screen, Kestrel's June bar is 37 percent of Meridian's; in the table, 50 is 71 percent of 70. The gap between the two tops is still 20 thousand, but the picture now says the smaller category is exploding and the leader is a different order of thing.

A line chart encodes position rather than length, so it can legitimately start above zero. The cost is vertical exaggeration. Draw a 10-unit rise inside a window of 38 to 72 and the line climbs 29 percent of the plot height; draw the same rise from zero to 72 and it climbs 14. Same figure, twice the apparent movement. That's not automatically dishonest — if the approved sentence is "the figure rose by a quarter," a tight range shows the quarter. If the approved sentence is about how large that movement was next to the other category, the tight range is making an argument the data didn't supply.

Dual axes are the sharper version of the same problem. Set Kestrel's scale to 38–52 and Meridian's to 58–88, and Kestrel's line climbs about 71 percent of the plot height while Meridian's climbs 33. The small series appears to be pulling away. In the table, the gap is 20 thousand in January and 20 thousand in June. A dual axis lets you choose each series' apparent slope independently, which means you can produce almost any relationship you like from the same numbers — including relationships nobody can support.

So the treatment has to name the relationship it is carrying, not just the chart type. "Grouped bars, zero baseline, six months" is a specification. "Grouped bars showing a gap that stays flat while both series rise" is a claim — and it is the one the passage will actually make.

Give motion an explanatory job

Motion can do two honest jobs. It can reveal items one at a time, which organizes attention and asserts nothing about time. Or it can follow recorded time, which asserts something about the order and spacing of real observations. Both are useful. They are not interchangeable, and the difference decides what you're allowed to animate.

Here is the same fixture twice. Same six months, same twelve figures, same running time.

Sequence A — the draft.

• The vertical axis starts at 38 and carries no label while the bars are moving.
• Kestrel's bars animate up over 1.2 seconds with an ease-out, arriving fast and settling. Meridian's bars start 0.4 seconds later, so Kestrel's growth plays first and alone, before the second series moves.
• The caption lands after: "Kestrel's surge."

Every printed number in A is correct. The bar lengths are not: they've been converted from values into differences from 38. The ease-out adds a rate change that isn't in the table — the increments are flat at 2 thousand per month, five months running, and a bar that snaps upward reads as accelerating. The stagger points the camera at the sixfold bar growth. And "surge" names the impression the baseline and the easing produced, which is how a claim gets into a treatment without anyone deciding to put it there.

Sequence B — repaired, same data.

• The vertical axis starts at zero, runs to 80, and shows gridlines at 20.
• Bars are drawn to value from that zero line. The baseline is labeled once and stays on screen.
• Each month's pair enters together and holds. No series finishes before the other starts. The 20-thousand gap sits at 25 percent of the axis height in all six frames, at the same visual width every time.
• Bars arrive with a short uniform reveal. No ease-out, no stagger, no arrival speed that distinguishes one category from the other.
• A fixed line under the chart: "Recorded figure, in thousands, January–June. Bar length is proportional to the value."
• The last frame holds, and includes the six-row table at reduced size.

Nothing in B costs more than A. The difference is a baseline, a stagger, and a sentence. But B's motion has a job it can actually do: it walks through six recorded months in order and lets the viewer hold a constant offset. It reports nothing between the months and nothing about July.

That last point is where most data passages get into trouble, because animation time is not measured time. A four-second reveal of six months does not make one second equal to six weeks. If a progress bar runs beneath the animation, it needs to be labeled as animation position, not date. And the in-between is riskier still. A line segment drawn between January's 40 and February's 42 passes through every value a straight line would produce; those pixels are a convention, not a reading. That's usually fine, because line-chart readers know the convention. It stops being fine when a dot slides along the segment at constant speed, because now each frame presents a position that looks like an observation. The same test applies to the other standard moves. A particle field that accumulates during the passage reads as a running count, and if the table holds monthly totals, the particles are counting something nobody counted. A smooth morph from one month's graphic to the next implies everything in between, which is the same as inventing the months you didn't record.

Keep the essential explanation available without motion

If the passage only means something while it is moving, then everyone who arrives late, looks away, screenshots it, or reads a text description gets a different passage. Plan the held frame first, and write down what it contains: the comparison, the period, the units, and any qualification that changes the reading. A six-month window and a count of accounts are both conditions. So is a truncated baseline, if you decide one is warranted.

Alongside that held frame, write one or two plain sentences that carry the same relationship in words. This is the line you would use for a caption, a voiceover, or a description, and it is worth drafting early, because a sentence that can't state the relationship is a sign the graphic can't either.

Having that text also makes a nonvisual account possible. The W3C's Images Tutorial on complex images, at https://www.w3.org/WAI/tutorials/images/complex/ and checked on 8 September 2026, addresses describing the essential information in complex images such as charts and diagrams. That's guidance about describing images on the web. It is not a test of animated data, not a validation of any encoding, and not a certification that a sequence like this one is accessible. What it does support is the habit: the explanation has to exist somewhere other than the transition.

Review the implication, not only the transcription

The check that catches baseline problems isn't checking numbers against the table. The numbers match. The check is to watch the storyboard and write down what it asserts, as sentences, then hold those sentences next to the supported one.

Sequence A asserts: Kestrel's figure grew sixfold; Kestrel is surging. Neither sentence appears in the table, and neither came from a data error. They came from an axis and an easing curve. Once the assertions are written down, the review has something to argue with.

Some of what you find will be intended, and that's fine — the question is only whether the evidence covers it. What the review can't do is let copy repair a visual. "Illustrative only," set small beneath a bar that grew sixfold, doesn't cancel the bar. Two statements compete in the same frame, and the bar is bigger. If the visual says surge, the fix is in the visual.

Two restraints on this review, both worth keeping in the treatment itself. First, it is a judgment about the artifact, not a measurement of viewers. Calling a baseline misleading is an editorial reading of a chart, and a defensible one, but it isn't an audience test and shouldn't be described as one afterwards. Second, the treatment's own language is subject to the same check. "Data-driven," "real-time," "surging" — each is a claim with an empty slot in it.

Then the closing question, which is the one the reader's job turns on: what does the motion add? If the answer is "it reveals the comparison," check that nothing about the reveal changes the comparison. If the answer is "it follows six recorded months," check that no marker, ease, or interpolation claims a seventh. If the answer is that it makes the passage more exciting, that's an honest answer — and it means the excitement is coming from somewhere other than the evidence.

For a real commercial, the fictional fixture has to give way to the actual data, permissions, and claim conditions, inspected one by one. Until then, placeholders do the job better than confident numbers.

Ship one supported sentence, one stable visual relationship, and one clear statement of what motion adds. A stronger implication needs different evidence or a different animation. It never needs more reassuring copy.

Frequently asked questions

What separates a faithful table from a graphic that implies an unsupported claim?

The figures can all be correct while the passage pitches a claim nobody approved. The claim lives in bar lengths, axis range, series order, ease on a transition, and other visual choices. The fix is to decide what the graphic is allowed to carry before deciding how it moves: one supported sentence, one stable visual relationship, one job for motion.

Why can the same six monthly values support different-looking stories?

The article's Kestrel and Meridian fixture has three true but different findings: Meridian's figure sat 20 thousand above Kestrel's every month; Kestrel rose 25 percent while Meridian rose about 17 percent; both rose by 2 thousand per month. A graphic built for relative growth will look different and more dramatic than one built for a gap that never moves. The treatment should pick one comparison and state it in words a viewer would recognize.

What changes when a bar chart uses a nonzero baseline?

A bar chart's meaning is a promise that bar length is proportional to value, and that promise holds only from a zero baseline. Break the baseline and length no longer represents the quantity. In the fixture, with a baseline of 38, Kestrel's January bar is 2 units and its June bar is 12: the figure rose by a quarter, but the bar grew six times. Line charts encode position and can start above zero, but at the cost of vertical exaggeration. Dual axes let each series' apparent slope be chosen independently, so almost any relationship can be produced from the same numbers.

What can motion honestly do in a data passage?

Motion can reveal items one at a time, which organizes attention and asserts nothing about time, or it can follow recorded time, which asserts something about the order and spacing of real observations. The two are not interchangeable. Animation time is not measured time: a four-second reveal of six months does not make one second equal six weeks. Interpolation, particles that look like a running count, and smooth morphs between months can all imply observations nobody recorded.

How should a treatment review implications rather than only transcriptions?

The check is not whether the numbers match the table; they often do. Watch the storyboard and write down what it asserts as sentences, then hold those sentences next to the supported one. In the article's Sequence A, the visual asserts that Kestrel grew sixfold and is surging, though neither sentence appears in the table. Copy cannot repair a visual: 'Illustrative only' set small beneath a bar that grew sixfold does not cancel the bar. This review is an editorial judgment about the artifact, not a measurement of viewers.

Further reading

• Read a Brand-Lift Report Before Revising the Treatment — When the treatment's claim rests on a client's brand-lift headline, this shows which details of the study have to travel with the number to make it mean anything.