Most product data business cases die in the same meeting. They are written as a data project, so the finance director reads a cost with a vague benefit attached. The ones that get funded read as a change to three or four lines in the profit and loss. A measurement plan is bolted on. This article gives you that structure. The value lines that survive scrutiny, the ones that do not, the inputs to gather, and the eight objections you will get.
What your finance director is actually deciding
They are not deciding whether product data matters. They are applying three tests, usually without saying so out loud.
Is the money real? Can it be traced to a line they already report on, such as revenue, gross margin, returns provision or cost to serve? Anything that cannot be traced gets discounted to zero.
Is it attributable? If conversion rises next quarter, how will anyone know it was the content and not the pricing, the season or the new paid campaign? A business case with no attribution plan is a forecast, and forecasts lose to certainties.
Does it arrive inside the horizon? Most finance functions work to a twelve to eighteen month payback for operational investment. A benefit that lands in year three is worth very little to them today.
Write your case to those three tests and the conversation changes. The word “data” barely needs to appear.
The five value lines that stand up
Each one has a measurement route and an honest confidence level.
| Value line | How to measure | Confidence |
|---|---|---|
| Conversion | Cohort test | High |
| Returns | Reason codes | Medium |
| Discovery | Search logs | Medium |
| Cost to serve | Ticket deflection | Medium |
| Speed to market | Cycle time | High |
Take them in order.
Conversion on the pages you fix
This is the strongest line because it is the easiest to prove. Enrich a defined cohort of SKUs, hold back a matched control group, and compare conversion rate over the same period.
The number you need first is your current conversion rate on the pages in scope, split by category. Most ecommerce teams have this. What they usually do not have is a breakdown of conversion against content completeness, which is the analysis that makes the case.
Be careful with the uplift assumption. Any percentage you put in the model before the pilot is a guess, and your CFO will treat it as one. Present it as a range with a low case that still clears the hurdle rate. We cover how to measure the effect properly in our piece on product content performance.
Returns avoided
For anyone selling physical goods, returns are the value line finance understands fastest, because the cost is already sitting in their numbers. Every returned item carries shipping both ways, handling, restocking, write-down risk and the customer service contact.
The input you need is your returns rate by category, split by return reason code. If your reason codes include anything resembling “not as described”, “wrong item ordered” or “did not fit”, that subset is the addressable pool.
The honest caveat is that reason codes are unreliable. Customers pick whatever is quickest. Sample fifty returns manually and re-code them yourself before you quote a number, or the first person who checks will find the same weakness.
Discovery, on-site search and paid efficiency
Missing attributes remove products from filtered navigation. Products that cannot be filtered are effectively out of stock for a customer who filters. This is measurable without spending anything.
Pull your on-site search zero-result terms for the last quarter. A large share of them will be attribute values you do not hold, or synonyms you have not mapped. That report is the most persuasive single artefact we have seen in a product data business case. It shows demand you already paid for and could not serve.
On paid, poor feed data reduces the share of your catalogue eligible for shopping campaigns. It also pushes cost per click up on the products that do qualify. Your agency can tell you your current disapproval and eligibility rates in an afternoon.
Cost to serve
Every pre-sales question that a customer asks because the specification is missing is a cost you carry. In B2B distribution this is a real number, not a rounding error. The questions arrive by phone and by email into a branch or a contact centre.
Ask for a sample of pre-sales contacts tagged by reason. Count the ones that are answerable from a datasheet you already own. Multiply by your fully loaded cost per contact, which your service manager will have.
The reason this line is only medium confidence is that deflected contacts rarely reduce headcount. They absorb capacity that gets used elsewhere. Say that in the case rather than waiting to be caught. Claim it as capacity released, not as cash saved, unless a vacancy genuinely goes unfilled.
Speed to market and range expansion
If it takes eleven weeks to get a new supplier range live, every week of that is deferred revenue. Halving the cycle is worth the gross margin on the range multiplied by the weeks recovered. That arithmetic is easy for a finance team to check.
The input is your current cycle time from supplier file received to product live, measured across the last twenty ranges. Almost nobody measures this. Getting the number is often the single most useful week of work in the whole exercise.
This line also opens the growth argument. If enrichment capacity is the constraint on how many ranges you can list this year, then the investment is not a cost saving. It is a revenue enabler, and it competes for a different pot of money.
The value lines that do not stand up
Leading with these is why business cases fail. Each one gets discounted to zero by anyone numerate.
- “A single source of truth.” True, possibly useful, and worth nothing in the profit and loss. It is an architecture statement, not a benefit.
- Hours saved. Twenty hours a week saved across a team is not a saving unless a head comes out or a vacancy stays unfilled. Say which.
- Readiness for AI or for new channels. Real, but unquantifiable today. Put it in the strategic section, never in the financial model.
- Risk avoidance with no exposure attached. “We might get fined” is not a number. If there is a regulatory exposure, quantify the fine and the probability, or leave it out.
- Competitor comparisons. Nobody funds a project because a rival did. It supports urgency. It does not support the maths.
None of these are untrue. They belong in the last paragraph of the paper, not the first page.
The inputs to gather before you write anything
This is the part most people skip, and it is why the paper reads as opinion. Gather these first.
- Conversion rate by category, from analytics. Owned by ecommerce.
- Content completeness by category, from your PIM or a spreadsheet audit. Owned by you.
- Returns rate and reason codes, from the returns system. Owned by operations.
- On-site search zero-result terms, from the search platform. Owned by ecommerce.
- Feed disapproval and eligibility rates, from your paid agency or merchant account.
- Pre-sales contact volume by reason, from the contact centre. Owned by customer service.
- Cycle time from supplier file to live, measured across recent ranges. Owned by you.
- Fully loaded cost of current enrichment effort, including the people doing it informally.
That last one surprises people. In most businesses the work is already happening, badly, spread across product managers, branch staff and marketing. Costing what you already spend is often half the case on its own. Our overview of product data services sets out what that work involves when it is done deliberately.
The model, in five lines
Keep it simple enough to fit on one slide. A model nobody can follow is a model nobody approves.
- Baseline. Annual revenue and gross margin on the SKUs in scope. Your figure.
- Uplift. The percentage change per value line, expressed as low, expected and high. Use 5 to 15 per cent for observed enrichment uplift, and mark it clearly as an assumption until your pilot replaces it.
- Phasing. When each cohort goes live and when the benefit starts. Enriched content does not pay back the day it is published.
- Cost. Enrichment, tooling, internal time and any platform change. For programme cost at your catalogue size, use £15,000 to £50,000 below 10,000 SKUs, £50,000 to £250,000 between 10,000 and 100,000, and £250,000 upwards above that.
- Payback. Cumulative benefit against cumulative cost, by month, for twenty-four months.
Run three scenarios and show all three. The low case is the one that gets scrutinised, so make sure it still clears your hurdle rate. If it does not, the scope is wrong and you should cut it back to the cohort where it does.
Never present a single number. A single number invites an argument about that number. A range invites an argument about the assumptions, which is the conversation you want.
Prove it with a cohort, not a projection
The strongest move available to you is to stop projecting and start measuring. It also costs very little.
Pick 500 to 2,000 SKUs in one category. Enrich them properly against a defined schema. Hold back a matched control group in the same category, similar price points, similar traffic. Leave both alone for eight to twelve weeks. Then compare conversion, returns rate, search impressions and average order value.
That test converts your business case from a forecast into evidence. It also gives you a real uplift figure to replace the assumption in line two of the model. We would rather scope a pilot like this than write a proposal for a full catalogue. Our product content enrichment engagements usually start with one.
Two practical warnings. Choose a category with enough traffic to reach significance, or the result will be noise. And do not enrich the control group when someone inevitably asks you to, because that is how these tests get ruined.
Where external benchmarks help, and where they do not
Published research is useful for sizing the problem and useless as proof of your return. Use it in the first paragraph, never in the model.
Salsify’s 2025 Consumer Research surveyed 1,910 shoppers in the US and UK. It found 54 per cent had abandoned a purchase over inconsistent product information across sites. Another 53 per cent had abandoned over incomplete or poorly written titles and descriptions. The same study found 71 per cent had returned a product because it did not match the online description.
Baymard Institute’s product page research found that 10 per cent of large ecommerce sites carry product descriptions that are insufficient for users’ needs. Gartner puts the average annual cost of poor data quality at 12.9 million dollars or more. That research was first published in 2020.
Now the honest part. A consumer survey of US and UK shoppers tells you very little about a trade counter customer buying hydraulic fittings. Your CFO knows that. Quote the numbers to establish that the problem is real and widely measured, then move straight to your own cohort test. A business case that leans on somebody else’s survey is a business case with no evidence of its own.
Eight objections and how to answer them
“We tried this before and nothing happened.” Usually true. Ask what was measured last time. In most cases the answer is nothing, which is why nobody could tell whether it worked. Lead with the measurement plan.
“How do I know it was the content?” The control group. This is the whole reason to run a cohort test rather than enrich everything at once.
“Can we not just use AI?” Partly, yes, and the case should already assume AI does the first draft. What AI does not remove is the cost of verification and the cost of being wrong. A generated specification nobody checked is worse than an empty field.
“Why can the team not absorb this?” They are already absorbing it, at product manager day rates, in the gaps between other work. Show the fully loaded cost of the current informal effort.
“What stops it going stale again?” Nothing, unless the case includes ongoing maintenance and someone owns the schema. Put the running cost in the model from month one. A case that ignores maintenance is the reason the last one failed.
“Prove it on our own products first.” Agree immediately. That is the pilot. Negotiate the size of the pilot, not the principle.
“This is an IT project.” It is not, and letting it be reclassified as one is how it gets deprioritised behind the ERP upgrade. Product data investment sits with whoever owns revenue on the channel. If a platform decision is genuinely in scope, treat it separately. Our guide to whether you need a PIM at all beats a vendor demo as a starting point.
“What is the cost of doing nothing?” Answer with the zero-result search report and the returns reason codes. Both are current losses, already incurred, and both are visible in systems your business already runs.
How to present it on one page
One page, four blocks, in this order.
The ask. The number, the period, and what it buys. First line, no preamble.
The three scenarios. Low, expected, high, with payback month for each. A small table beats a paragraph.
The evidence. Your own cohort test if you have run one, your zero-result search data and your returns codes if you have not.
The measurement plan. What you will report, to whom, monthly, and the date at which the programme gets stopped if the numbers do not appear. Offering a kill criterion is the fastest way to be trusted with the budget.
Everything else goes in an appendix. If a platform selection is part of the scope, keep it in a separate paper. PIM selection has its own decision process and its own timeline.
Key takeaways
- A product data business case gets funded when it changes named lines in the profit and loss. Describing a better data model does not.
- Five value lines stand up: conversion, returns, discovery, cost to serve, and speed to market.
- Hours saved, a single source of truth, and readiness for AI get discounted to zero. Keep them out of the model.
- Gather your own inputs first. Conversion by category, returns reason codes, zero-result search terms and cycle time are the four that matter most.
- Replace assumed uplift with a cohort test on 500 to 2,000 SKUs, against a held-back control group.
- Present three scenarios, a measurement plan and a kill criterion. Never a single number.
We help clients build this paper more often than we run the enrichment that follows it. Want a second pair of eyes on your model, or help scoping the cohort test? Book a thirty minute discovery call. You can see the delivery side on our product data services and product content enrichment pages.