The percentage in your business case has no denominator. A content tool that promises to cut review cycle time by 57% is quoting a figure with no sample behind it, no baseline underneath it, and no written definition of the thing being cut. Before you buy a tool that promises to move a number, spend one month measuring that number yourself. Your own figure is the only version that will survive contact with your own team.
There is one product marketer where I work, and there are seventeen products. That is the position these numbers get sold into: one person, no analyst, a stack of tools to evaluate, and no time to check the arithmetic on any of them. So I went and read one vendor's features brief.
What makes a number real
A measured number is one where somebody can hand you four things: the population that was counted, the state before, the state after, and the written definition of the thing being counted. Miss any of the four and you cannot tell whether the figure describes your situation or somebody else's.
Ask a vendor for all four. Then ask which documents were counted. The slide does not carry an answer to the second question.
A decorative number is a claim about direction wearing a decimal point.
Direction is often true. Software does speed up review. The question is whether anybody counted, and by how much, against what.
The document
Veeva sells Vault PromoMats, a system pharmaceutical companies use to run the sign-off that promotional material has to clear before it goes out. Its features brief carries four figures in a block: 57% reduction in review cycle time, 55% reduction in time spent in the medical, legal and regulatory meetings the brief calls MLR and PRC, 25% reduction in time spent on compliance procedures, and 88% reduction in time spent on agencies initiating and uploading jobs. Veeva Vault PromoMats Features Brief
The only methodological statement attached to those four numbers is the sentence that introduces them. Most customers, it says, have recognised a six month return on investment and experienced significant improvement in other core content management metrics.
Read that sentence for what it commits to. Some unstated majority of customers improved by some unstated amount, over an unstated period, against an unstated starting point. Then four decimals arrive anyway.
The file itself sits at an upload path dated May 2014 and carries a copyright line reading 2025. Whatever that says about the document's history, those four figures are what it serves today.
Where the clock starts was decided in 2020
The missing sample is not the deepest problem here. The missing definition is.
In September 2020 the same company published an overview of what it called Standard Metrics. Nine document fields were added across every PromoMats vault so that a handful of process questions could be answered at all, for anyone running sign-off inside that system. The deck says the feature will, for the first time, allow customers to compare their business processes against industry average anonymised benchmarks, and that Veeva would aggregate the data and share industry averages in 2021. Vault PromoMats Standard Metrics Overview, 20R3
Number of review cycles became the cumulative number of times a document enters the In Review state before reaching its initial steady state date. Days to approval became the cumulative time between the document's lowest minor version and that same steady state date.
Notice how many arbitrary choices sit inside those two sentences. When does the clock start? At the lowest minor version, because somebody decided that is when the asset exists. When does it stop? At steady state, because somebody decided that is when it is done. Neither choice is obvious, and both had to be written down and switched on across every vault before one company's figure could be set beside another's.
You cannot compare one company's 57% against another's before anyone has agreed where the clock starts.
This is the part that travels past pharma. Every workflow metric you might want has the same problem waiting inside it. Time to approval, campaign cycle time, rounds of revision, time to first draft. Each one is a stopwatch, and somebody has to say when to press it. Two people on the same team will time the same asset differently and neither of them is wrong, because nobody wrote the rule down. In a regulated market the process itself tends to force the writing down, which is why the security questionnaire ends up deciding the deal. Outside one there is no such pressure, so I do not know whether a definition survives its second quarter. I have watched a related failure inside my own tooling, where a quality score turned out to be measuring the wrong thing entirely while reporting a confident 9.5 out of 10.
What the number was when somebody counted
In April 2021 the benchmarks arrived. Drawn from anonymised system data across more than 350 pharmaceutical companies, the average for delivering content from design to approval was 21 days, and 74% of digital content was approved in a single cycle. Leading Your Content Strategy with Industry Benchmarks, Veeva, April 2021 The source is careful to note significant differences between the simplest content types and the most sophisticated, so 21 days is an average across work of different weights. These figures also come from the same company that sells the tool, and they are worth something here only because they are derived from timestamps rather than from opinion.
Two years later a European cut of the same programme put average approval at 20 days, with an average of 1.3 review cycles. Using Benchmarks to Speed and Scale Life Sciences Content, Veeva, 2023
Now set that beside what the industry says about itself. In November 2024 the same company published a medtech benchmark release reporting that marketing assets undergo an average of three to five rounds of reviews. Report Reveals 70% of Medtechs Use Manual Processes, Veeva MedTech, November 2024 That figure is a survey finding, drawn from over 130 commercial medtech leaders, so it has a population and a disclosed sample. What it does not have is a document behind it. It records what experienced people say happens, rather than what any system recorded.
The two sets are not directly comparable -- one covers medtech, the other pharma. Set that aside and the interesting part survives. One method counted documents and the other counted recollections, and the recollections are what travel.
That is also how folklore acquires authority. A number that began as what everyone remembers picks up a sample size, a vendor logo and a publication date, and from there it gets quoted as a measurement.
Why the unmeasured number is the one that spreads
Veeva is the company that went and fixed the definition problem. It wrote the rules down, switched nine fields on across every vault, aggregated the data and published benchmarks built from timestamps rather than opinion. I have not come across another enterprise vendor that has done that work in its own category. And the undefined 57% is still on the front of the brief.
The two numbers are built for different jobs. A percentage reduction is a sales object, designed to be lifted into somebody else's slide. An average of 1.3 review cycles is a level rather than a gain, and a level cannot be lifted into anything. Even at the company that built the measurement, the number on the brochure is the undefined one.
Folklore also flatters your memory, which is the second reason it spreads. The five-round assets are the ones you remember. Single-pass approvals leave no trace. A number that matches what you already feel gets repeated without anyone checking it, and it keeps travelling long after the thing it described has changed, which is what happens to a product story after its launch.
Feeling right and selling well are the two properties that carry a number into a business case. Neither of them tells you anything about whether it is true of your team, which is the only question you are actually asking.
Instrument the number you do not have
Here is the month of work that replaces the brochure figure. Run it before you sign anything.
- Write the definition first, before you collect a single data point. Pick the moment the clock starts and the moment it stops, and write each in one sentence. Copy Veeva's if you like, since they are public and specific. If two people on your team would time the same asset differently, you have not finished this step, and everything you collect afterwards is noise.
- Count everything in flight for thirty days, not a sample you chose. If nothing in your process emits timestamps, one shared sheet is enough. Give it one row per handoff rather than one per asset, because an asset that goes round three times needs three rows: the asset, who received it, the date it reached them, the date they opened it, and the date they sent it back. Note the clock-start and clock-stop dates against the asset itself. Whoever moves the asset fills the row. That is worse than system data and far better than memory.
- Record the shape, not the average. Count the rows each asset collected, then note how many cleared in one pass and how many took more than three. Say the average comes back at 2.4 rounds. That figure hides the difference between a slow, even process and a fast one with a few disasters in it, and those two need opposite fixes.
- Separate waiting from working, using the columns you already have. The gap between arrival and opening is queue. The gap between opening and return is work. Look at which of the two is larger before you buy anything, because a tool that shortens the queue does nothing to the second. Some of your queue will also have been created upstream, at the point where requests arrive with no triage attached.
- Read the vendor's number again with your own figures in hand. Ask which of them it claims to move, and against which definition of the clock. A vendor who cannot answer has not measured it either, and that tells you more than the answer would have.
The first step is the one people skip, because a week of thinking about definitions produces nothing anyone can look at. Skipping it makes the remaining four unusable, since counting against a definition two people read differently gives you a number nobody will defend in a room.
If you do not have a month
Sometimes the tool is already committed, the discount expires at quarter end, or the decision was made above you. The compressed version is two steps, and it costs one afternoon.
Run step 1 on your own, which takes a few hours: write down where the clock starts and stops. Then run step 5 in the vendor meeting you already have booked, and ask which definition their number moves against. You now hold the two most valuable steps for the price of a single conversation, and the answer to step 5 tends to settle the question on its own.
Then start the thirty-day count in parallel with the pilot rather than ahead of it. A pilot with no baseline can only be judged on how it feels, which is the same trap as asking whether people liked the deck.
When the answer comes back boring
Plan for the likeliest result. You may find your work clears faster than the folklore says, and that the bottleneck you were about to spend a budget on is not where you thought it was.
Price it honestly in both directions. A thirty-day count costs a shared sheet and about twenty minutes a week from whoever moves the assets, which is cheap against a workflow tool and the implementation behind it. Against a bottleneck that is genuinely costing you launches, though, a month of counting is a month of that continuing. If you already know the queue is the problem, run the count alongside the pilot instead of in front of it.
If the number does come back bad, you are in a far stronger position than the business case you started with. You know which part is queue and which part is work. You know your own starting point. You will know within a quarter whether the thing you bought moved it, because you can run the same count again against the same definition. That comparison is unavailable to anyone arguing from a brochure. It is also what separates a real workflow change from automating the part of the job that was never the bottleneck.
The tool might still be worth buying. Many of them are. You will simply be buying it against a number you produced rather than one a vendor produced.
Nobody else is going to count it for you.