6 August 2026
Announcing The State of Platform Disputes — the benchmark a whole function is missing
Every support org knows its CSAT. Every payments team knows its auth rate. Ask a dispute-ops team their fully loaded cost per dispute, their overturn rate, or whether two of their agents would rule the same case the same way — and the answer, almost everywhere, is silence. We are building the first annual benchmark for the function. Here is what we can already say from public numbers, what nobody can say yet, and how to put your queue in the dataset.
A cost center with no yardstick
Dispute resolution is a strange corner of the platform economy: it moves real money between users, it carries real legal exposure, it burns real headcount — and it has no published benchmarks at all. There is no industry number for what a dispute costs fully loaded. No number for how often an appealed decision gets reversed. No number for consistency between agents. Each ops lead calibrates against the only queue they have ever seen: their own.
The consequence is predictable. Budgets are defended by anecdote, tooling decisions are made blind, and when a CFO asks "is our dispute operation good?", the honest answer is "compared to what?"
The State of Platform Disputes is our attempt to build the "compared to what." First edition publishes when the participating cohort is large enough to anonymize credibly — the report will state its N on page one, and we will not publish until it is defensible.
What public numbers already tell us
Even before any survey data, the public anchors sketch the cost landscape — these are the only numbers in this post, and every one of them is external and checkable:
| Anchor | Public figure | What it tells you |
|---|---|---|
| Outsourced BPO ticket handling | ~$5–15 per ticket for simpler support work | The floor for human handling — before dispute work's evidence review, judgment calls, and two-sided anger premium |
| Card-network chargeback fees | $15–25 per case, before any labor | What the payments stack charges merely to administer a formal dispute, win or lose |
| DSA-certified out-of-court settlement bodies | Tens of euros per case, platform-paid | The statutory external benchmark European platforms are now obliged to engage with — at human-body prices (our Article 21 analysis) |
| Statutory ODS resolution window | 90 days, extendable for complex cases | The speed bar regulation considers acceptable — set by what human adjudication can promise, not what disputants need |
Put together: the market's reference prices for formally handling one dispute cluster in the $5–25+ range before labor, and the reference timelines run days to months. If those anchors feel comfortable, consider that a Tier 1 automated verdict on our published rate card lists at $4 with an SLA measured in minutes — not as a boast but as a category question: what should this function cost? Nobody can currently answer, because nobody has measured the inside of the queues.
What nobody can tell you (yet)
The report exists to produce the numbers that today live nowhere:
| Metric | Definition (measure yourself with this) |
|---|---|
| Fully loaded cost per dispute | (Handler minutes × loaded hourly cost) + escalation and review time + make-goods and goodwill credits + close-the-ticket refunds — the refunds issued not because the case merited them but to end the conversation. Divide by disputes closed. |
| Time to resolution | Median and p90, dispute opened → final decision communicated. Report both; the p90 is where the brand damage lives. |
| Overturn rate | Of decisions appealed through any second look (internal appeal, manager review, external body), the share reversed or materially modified. |
| Consistency score | Sample 20 closed cases, strip the outcomes, have a second agent rule them blind: the agreement rate. Few teams have ever run this experiment. Fewer publish it. |
| Acceptance rate | Share of rulings both parties accept without escalating, appealing, or churning within 60 days. |
| Refund-to-close rate | Share of disputes resolved by refunding without an actual adjudication — the quiet tax nobody books as a dispute cost. |
A self-test while you wait for the report: if your team can produce the first three numbers for last quarter within one week, you are running one of the most instrumented dispute operations we have encountered. We say that without irony — the function is under-measured everywhere, which is precisely why a benchmark is worth building.
How the data gets collected — and what participants get
- Contribute: a structured submission (roughly a 30-minute exercise against the definitions above) plus an optional 45-minute walkthrough of your queue with our team.
- Anonymity by construction: results publish only as aggregates and percentile bands by platform category and volume tier. No platform is named, ever, without written consent.
- What you get: early access to the full report, your own percentile placement on every metric against the cohort — the "compared to what" for your next budget conversation — and, if you want it, a free 30-day shadow pilot to measure a machine-cost baseline on your own queue alongside your team's.
Put your queue in the dataset — tell us your platform category and monthly dispute volume, and we'll send the submission kit.
Method notes will publish with the report: metric definitions as above, collection windows, cohort composition by category and volume tier, and known biases (self-selected cohort; self-reported inputs; our commercial interest in the category, which we mitigate by publishing definitions anyone can re-run against their own queue).