What are the most important video marketing metrics in B2B?

Four carry real information: completion or sustained-watch rate, which tells you whether the film held anyone; the next-step rate, which is the share of viewers who took the one action the asset asked for; assisted pipeline, which is the value of deals where the asset is recorded on the account; and cost per influenced deal, which is what finance will actually ask. Views, impressions and average watch time belong in the diagnostic layer, not the report — they describe distribution and editing, not interest or outcome.

Why are video views a misleading metric?

Because a view is a measure of distribution rather than of interest, and because the definition varies by platform — some count a view at two or three seconds of autoplay in a muted feed. Two assets with identical view counts can behave in completely opposite ways once you look at whether anyone finished watching or did anything afterwards. A view count rises when you spend more on distribution, which is why it is the easiest number to improve and the least informative to report.

What is a good completion rate for a B2B video?

There is no defensible universal figure, because completion is a function of length, placement and audience intent as much as quality — a 90-second explainer on a demo page and a 12-minute customer story on a resource hub are not comparable, and a benchmark that ignores that is describing someone else’s asset mix. Use your own library as the baseline: compare like with like, watch the direction over time, and treat a sharp drop in the first third as the signal worth acting on.

How do you measure video that generates no clicks?

With leading indicators and by asking, rather than by inference. Branded search volume, direct traffic to the asset page, and the share of buyers who mention the asset unprompted in a first call all move on the same timescale as the influence itself. None is precise, and all three are more honest than attributing a months-long memory effect to a last-touch model that never saw it.

Which video metrics should you stop reporting?

Impressions, raw view counts and average watch time as headline figures. Impressions measure what you bought. Views measure distribution and vary by platform definition. Average watch time collapses a split distribution into one figure — where most people leave in seconds and a few watch to the end, the average describes nobody. Keep all three for diagnosis; stop putting them in front of anyone deciding a budget.

How many video metrics should a report contain?

Four or five, with the untraced share named alongside them. A report with fifteen metrics is not more rigorous, it is unreadable, and it invites the reader to select whichever number supports the view they arrived with. Pick the small set that maps to decisions you would actually take differently, and report the gap between influence you can evidence and influence you believe in as a quantity that should be shrinking.

Field guide

The video marketing metrics that predict pipeline

Views are a measure of distribution, not of interest.

By Paul Joseph · Updated 24 Aug 2026

Four video marketing metrics carry real information in B2B: sustained-watch rate, which says whether the film held anyone; next-step rate, the share of viewers who took the one action the asset asked for; assisted pipeline, the value of deals the asset is recorded against; and cost per influenced deal, which is the number finance will actually ask for. Views, impressions and average watch time are diagnostic instruments that have been promoted to headline figures, and the promotion is why most video reports are simultaneously long and uninformative.

Every metric below assumes a view can reach a known contact. If it cannot, the top two are all you will ever have — the Dark-Funnel Video Audit scores that in two minutes, free, and tells you which half of this page currently applies to you.

Why the default metrics mislead

The metrics that arrive by default arrive from platforms whose interest is in demonstrating reach. They are not dishonest, but they answer a question about distribution when the question in the room is about outcome.

A view is not a unit of attention. The definition varies by platform, and on several it is satisfied by two or three seconds of autoplay in a muted feed that the person never chose to start. That makes view counts non-comparable across channels and trivially improvable by spending more on distribution — which is exactly what makes them attractive to report and useless for deciding anything.

Average watch time describes nobody. Where retention splits — a large group leaving within seconds, a smaller group watching most of it — the average of those two populations lands in a middle where almost no actual viewer sits, and it moves for reasons — a change in traffic mix, an autoplay placement — that have nothing to do with the film.

The practical test for any metric on a report: name the decision you would take differently at a different value. A number that survives that question belongs there; one that does not is describing activity.

Impressions measure what you bought. They are a spend report wearing a performance report's clothing. Keep them for media planning; take them out of anything a budget-holder reads.

The four that carry information

Each answers a different question, and they are ordered by how far down the chain they sit — the first tells you about the film, the last about the business. A report with all four can survive a sceptical reading.

01 · Sustained-watch rate — did it hold anyone?

The share of starts that reach a meaningful point — completion on a short asset, or a fixed marker such as 50% on a longer one. Use one definition and keep it. This is the only metric on the list that is about the film itself rather than about what surrounds it, and a sharp fall in the first third is the most actionable signal in the whole set.
02 · Next-step rate — did it move anyone?

Of the people who watched, the share who took the single action the asset asked for. This requires the asset to have asked for exactly one thing, which is why it is the metric most often unavailable — and why its absence is itself a finding about the brief rather than about the film.
03 · Assisted pipeline — did it touch money?

The value of open and closed deals where this specific asset is recorded on the account. Record at account level rather than per person: a buying committee contains people who will never identify themselves, so person-level counting undercounts video structurally, and worst on the assets that circulate best.
04 · Cost per influenced deal — was it worth it?

Production cost divided by the deals the asset is recorded against, over the asset's life rather than the quarter. This is the one that translates into a language finance already speaks, and the only one on the list that improves by doing nothing — every additional month in service lowers it.

Report a fifth quantity alongside them: the untraced share — deals you believe video influenced minus deals you can evidence. Naming it is what makes the other four credible, and it should shrink quarter over quarter. The instrumentation that shrinks it is in video attribution for B2B.

A worked example: two assets, identical views

Illustrative arithmetic, not a benchmark. The numbers are round so the mechanism stays visible.

A company publishes two assets in the same quarter: a brand film and a customer story. Both cost $20,000. Both report 40,000 views. On the dashboard the quarter looks like two equal successes, and the natural conclusion is to commission more of both.

Two assets with identical view counts, compared on the four metrics
Per asset	Brand film	Customer story
Views	40,000	40,000
Sustained-watch rate	6%	34%
Watched to the marker	2,400	13,600
Next-step rate	0.5%	4%
Next steps taken	12	544
Deals it is recorded against	3	29
Cost per influenced deal	$6,667	$690

Identical on the metric that gets reported; an order of magnitude apart on the one that decides the next commission. The brand film was distributed into a paid feed where autoplay counts a view before anyone chooses to watch; the customer story sat on a page a buyer arrived at deliberately. The films may both be well made — the difference is placement and intent, which the view count cannot see.

The dangerous reading of this table is that brand film is a waste. It is not what the numbers say. An asset built for buyers who are not in market is being judged here on a conversion measure it was never designed to move, and roughly 95% of business buyers are out of market at any moment — the 95-5 rule, from Professor John Dawes of the Ehrenberg-Bass Institute for the LinkedIn B2B Institute (2021). Judged on branded search and unprompted mentions instead, the same film may be the better performer.

That is the actual lesson: the metric has to match the asset's job. Applying conversion metrics to memory-building work defunds the thing that fills the funnel, and applying memory metrics to conversion work excuses assets that are simply not landing. The 95-5 rule and your video budget sets out which measures belong to which half.

Match the metric to the asset's job

Two asset classes, two measurement regimes. Choosing the regime at brief time rather than at review time is what makes the result arguable in your favour instead of against it.

Which metrics apply to conversion assets and which to memory-building assets
	Conversion assets	Memory-building assets
Primary measure	Next-step rate at a named step	Branded search volume over time
Secondary	Assisted pipeline, cost per influenced deal	Unprompted mentions in first calls
Horizon	Weeks to a quarter	A year and beyond
Attributable?	Largely, if instrumented	No, and a plan that assumes so will fail

The right-hand column is the one that gets cut in a hard quarter, because its measures do not look like revenue and its horizon outlasts most marketing tenures. Agreeing its indicators in the brief — before anything ships — is the only defence that holds, and it is decision five of the 60-Second Brief.

Capturing the four without buying anything

A common objection at this point is that measuring properly needs a platform. It does not, at least not to start, and buying one before the chain exists is how teams spend a year implementing a tool that reports on data nobody is capturing. Each of the four can be captured with what most B2B teams already pay for.

1. Sustained watch. Your video host already reports retention. The only decision is to fix one marker per asset class — completion under two minutes, 50% above it — and record it consistently. The discipline is worth more than the precision: a number defined differently each quarter is not a trend.
2. Next-step rate. Needs the asset to ask for one thing and that thing to be reachable from a page you control. The denominator is starts, not impressions, and the two must come from the same source or the ratio is meaningless. If the asset asks for nothing, this metric is not missing — it is telling you the brief was incomplete.
3. Assisted pipeline. One multi-value field on the account record, holding asset names. Do not fight for the lead-source field, which is already contested by three teams and cannot express several assets across nine months. A field nobody else uses is easier to get and more honest to read.
4. Cost per influenced deal. Production cost, which you know exactly, divided by the count from step three, over the asset's life rather than the quarter. The only judgement is the life, and stating the assumption — two years, three — is what keeps the figure defensible.
5. The untraced share. A free-text “how did you hear about us” on the form and the same question in the first call, kept verbatim and reviewed monthly. This is the cheapest instrument on the list and the only one that reaches influence the stack never saw.

Sequence matters more than completeness. Do this on one asset, end to end, before applying any of it across a library — a single instrumented film exposes where your particular stack breaks, which no portfolio-wide policy written in advance can predict. A platform is worth buying once you are doing this across enough assets that the manual version is the bottleneck, and not a quarter earlier.

One caution on all four: none of them is a quality judgement. A film can score badly because it was placed on a page nobody arrives at with intent, and well because it sat at the end of a well-qualified funnel. The metrics test the chain the asset sits in as much as the asset itself, which is why a poor number is a prompt to ask which link failed rather than a verdict on the work. Read them as diagnostics, and the conversation stays about the system instead of becoming about whose film underperformed.

What a usable video report looks like

Four or five numbers, the untraced share named, and one sentence per number saying what you would do differently if it moved. A report nobody would act on differently at any value is decoration, however carefully assembled.

Keep the diagnostic layer — views, impressions, average watch time, platform retention curves — in a place you can reach when a headline number moves and you need to know why. That is what those metrics are genuinely good for. What they are not good for is being the report, because they invite a reader to pick whichever figure supports the position they walked in with.

Two habits make the difference over a year. Compare like with like — a 90-second explainer on a demo page and a twelve-minute customer story on a resource hub share no useful baseline, and averaging them produces a number about neither. And use your own library as the benchmark rather than a published industry figure, which describes an asset mix, a length distribution and a placement pattern that are not yours. The arithmetic that turns these metrics into a return figure is in how to measure video marketing ROI, and the Frame to Funnel Method names the three links each of these metrics is testing.

Video metrics, answered

What are the most important video marketing metrics in B2B?: Four carry real information: completion or sustained-watch rate, which tells you whether the film held anyone; the next-step rate, which is the share of viewers who took the one action the asset asked for; assisted pipeline, which is the value of deals where the asset is recorded on the account; and cost per influenced deal, which is what finance will actually ask. Views, impressions and average watch time belong in the diagnostic layer, not the report — they describe distribution and editing, not interest or outcome.
Why are video views a misleading metric?: Because a view is a measure of distribution rather than of interest, and because the definition varies by platform — some count a view at two or three seconds of autoplay in a muted feed. Two assets with identical view counts can behave in completely opposite ways once you look at whether anyone finished watching or did anything afterwards. A view count rises when you spend more on distribution, which is why it is the easiest number to improve and the least informative to report.
What is a good completion rate for a B2B video?: There is no defensible universal figure, because completion is a function of length, placement and audience intent as much as quality — a 90-second explainer on a demo page and a 12-minute customer story on a resource hub are not comparable, and a benchmark that ignores that is describing someone else’s asset mix. Use your own library as the baseline: compare like with like, watch the direction over time, and treat a sharp drop in the first third as the signal worth acting on.
How do you measure video that generates no clicks?: With leading indicators and by asking, rather than by inference. Branded search volume, direct traffic to the asset page, and the share of buyers who mention the asset unprompted in a first call all move on the same timescale as the influence itself. None is precise, and all three are more honest than attributing a months-long memory effect to a last-touch model that never saw it.
Which video metrics should you stop reporting?: Impressions, raw view counts and average watch time as headline figures. Impressions measure what you bought. Views measure distribution and vary by platform definition. Average watch time collapses a split distribution into one figure — where most people leave in seconds and a few watch to the end, the average describes nobody. Keep all three for diagnosis; stop putting them in front of anyone deciding a budget.
How many video metrics should a report contain?: Four or five, with the untraced share named alongside them. A report with fifteen metrics is not more rigorous, it is unreadable, and it invites the reader to select whichever number supports the view they arrived with. Pick the small set that maps to decisions you would actually take differently, and report the gap between influence you can evidence and influence you believe in as a quantity that should be shrinking.

Related: the dark funnel is why the untraced share exists, when a B2B video strategy is not working covers the case where the metrics are fine and the programme is not, and the glossary defines every term used here.