Too Small to Fight: The Harms Below the Cost of Contesting | HL Hunt
Too Small to Fight: The Harms Below the Cost of Contesting
Every dispute has a price — time on hold, a form, a follow-up, some chance of losing anyway. Below that price, being right stops mattering. A $14 fee charged in error is not worth forty minutes during a working day to someone paid by the hour, so it stays charged. That creates a threshold under which errors persist regardless of merit. And it produces something more interesting than persistence: because firms correct errors that cost them and customers must correct errors that cost them, the errors that survive lean toward the firm — with nobody intending it at all.
In this report
The contest threshold
Stated as simply as it can be: a person contests a harm when the expected recovery exceeds the cost of contesting.
| Component | What it includes |
|---|---|
| Expected recovery | The amount, multiplied by the chance of getting it back |
| Time cost | Hours on hold, writing, following up — valued at what they displace |
| Direct cost | Postage, fees, travel, occasionally advice |
| Uncertainty | Not knowing whether you're right, or what the process will demand |
| Friction | Finding the process, getting through, repeating yourself |
Everything below the line where these balance goes uncontested. Not because people don't care, and not because they don't notice — because acting on it is a loss.
Per our consent analysis, not reading terms is an accurate response to a document you can't negotiate. This is the after-the-fact version of the same observation: not disputing is an accurate response to a harm smaller than the cost of disputing it. Both look like consumer passivity and both are correct calculations.
Below the threshold, the question of who was right has no practical answer, because nobody asks it.
The arithmetic of not disputing
A stylized case — a $14 fee charged in error, to someone paid hourly:
| Amount | |
|---|---|
| Fee charged in error | $14.00 |
| Chance the dispute succeeds | 70% |
| Expected recovery | $9.80 |
| Time to dispute, including hold and follow-up | 40 minutes |
| Value of that time at $22 an hour, unpaid | $14.67 |
| Net result of disputing | −$4.87 |
Disputing a charge that is plainly wrong makes this person worse off. The figures are illustrative; the structure isn't. For an hourly worker whose calls must happen during their own working hours, the threshold sits well above most fees — and per our time analysis, that's the population with the least schedule control.
For someone salaried with a flexible day, the same forty minutes costs close to nothing in foregone income. Same error, same merit, opposite correct decision. The threshold isn't a property of the harm — it's a property of the person experiencing it.
What the firm sees
The same $14 error, applied across a customer base.
| Figure | |
|---|---|
| Customers affected | 150,000 |
| Total charged in error | $2,100,000 |
| Share who contest | 8% |
| Share of contests that succeed | 70% |
| Refunded | $117,600 |
| Retained | $1,982,400 — 94% |
Every individual decision not to contest was rational, and together they left 94% of an erroneous charge in place. The harm that isn't worth fighting individually is substantial in aggregate — and only the firm ever sees the aggregate.
That's the core structural point. The customer experiences $14; the firm experiences $2.1 million. One side of the transaction has the incentive to act and the other doesn't, and it's the side with the larger number that doesn't need to.
Asymmetric correction
The part of this analysis that doesn't require anyone to behave badly.
Errors happen in both directions. Systems misapply fees and miss fees, overcharge and undercharge, credit accounts late and early. Assume, generously, that they happen equally often each way.
Now look at how each direction gets corrected:
| Error against the firm | Error against the customer | |
|---|---|---|
| Detected by | The firm's reconciliations | The customer, if they notice |
| Corrected by | The firm, routinely | The customer, if they contest |
| Cost of correcting | Borne by a process built for it | Borne by the person harmed |
| Illustrative correction rate | 95% | 8% |
Start with 1,000 errors, 500 each way. After correction, 25 errors remain against the firm and 460 remain against customers. The surviving error population is 95% in the firm's favour — from a starting point of perfect symmetry, with no intent anywhere in the system.
That's the finding worth holding onto. A firm can be run by honest people, make mistakes at random, and still end up with a book of residual errors that leans heavily one way, because correction is a selection process and the selection is lopsided. Per our measurement analysis, it's also invisible from inside: the firm sees a low complaint rate and reads it as a low error rate.
And it compounds over time. A system that ships with a small error in the firm's favour has no internal pressure to find it, while an error in the other direction triggers a fix in the next release. Over years, the system drifts toward whichever errors nobody is paid to catch.
Where the threshold sits for whom
The distributional finding, and it's the same pattern this desk keeps reaching.
- Time cost is highest for hourly and shift workers, whose disputing happens in unpaid or lost time — per our time analysis.
- Uncertainty is highest for people with least product familiarity, who can't tell whether a charge is wrong.
- Friction is highest for people with language, access, or disability barriers.
- The amount matters most to people with least slack — which should lower their threshold, but per our time preference analysis, a household under pressure has its attention fully committed elsewhere.
Which produces a perverse outcome: the people for whom $14 matters most face the highest cost of recovering it. A salaried customer contests the fee and gets it back; the hourly customer absorbs it. Per our defaults analysis and our place analysis, it's the same population that bears the cost of defaults and of the lost escalation path — the fourth mechanism in this library that sorts on capacity rather than on merit.
Who sets the threshold
The strategic observation, and it's the uncomfortable one.
The threshold isn't fixed. The firm largely controls it, through the design of its dispute process:
- Hours — business-hours-only raises the time cost for exactly the people with least flexibility.
- Channel — phone-only, with hold times, versus a form that resolves on submission.
- Evidence demanded — asking the customer to prove an error the firm could check.
- Repetition — being passed between departments, restating the problem.
- Outcome certainty — a process with unclear rules raises the uncertainty cost.
Every one of these raises the threshold, and a higher threshold means more errors stay uncorrected. So a firm can reduce refunds either by making fewer errors or by making disputes harder — and the second is usually cheaper, which is the incentive problem stated plainly.
Per our complexity analysis, this rarely needs to be deliberate. A dispute process optimized for handle time and cost will drift toward a higher threshold on its own, because every friction-reducing investment shows up as cost and every friction-increasing one shows up as savings.
The aggregation mechanisms
The ways small harms get combined into something worth pursuing — and what each has lost or kept.
| Mechanism | How it aggregates | Status |
|---|---|---|
| Class actions | Many claims, one case | Frequently waived by contract |
| Regulators | Supervise and enforce across a book | Capacity-limited |
| Complaint databases | Make patterns visible | Depend on individual complaints |
| Card network disputes | Lower the cost of contesting to near zero | Working, for card payments |
| Automatic remediation | Firm refunds without being asked | Rare, usually after enforcement |
Row one is the subject of our arbitration analysis. That report was about forum; this one is about why forum matters so much — a class waiver removes the mechanism that makes a sub-threshold harm worth pursuing, which is a larger consequence than moving disputes from one venue to another.
Row four is the instructive one. Per our reversibility analysis, a card dispute costs the customer a few taps and puts the burden of proof on the merchant — so it's the one domain where the threshold is genuinely low, and it's the domain where merchants complain most loudly about customers disputing too readily. That's what a low threshold looks like from the other side.
Row five is the only mechanism that doesn't depend on anyone contesting, and it's the one this analysis points toward.
The strongest objections
"Most small errors are genuine mistakes." Agreed, and the argument assumes so. The asymmetric correction finding holds precisely when errors are random — it's a property of the correction process, not of anyone's intent. Intent would make it worse; its absence doesn't make it go away.
"Firms have reputational reasons to correct errors." For visible errors, yes. The threshold concerns errors too small for the customer to act on, which means they're also too small to generate reputational cost — a customer who didn't dispute a $14 fee usually doesn't write about it either. Reputation disciplines the errors people notice and bother with, which is the set that was already being corrected.
"You've assumed the correction rates." Correct, and it's the central limitation. The 95% and 8% are illustrative, and the size of the lean depends entirely on them. The direction of the finding holds for any pair where firm-side correction exceeds customer-side correction; the magnitude needs data, and the test below is how to get it.
Testable implications
- Residual errors in any firm's systems should lean in the firm's favour, measurable by auditing a random sample of accounts for errors in both directions — the direct test.
- Dispute rates should fall sharply as dispute cost rises, with little change in underlying error rates.
- Contest rates should vary by customer schedule flexibility more than by amount at stake.
- Removing class waivers should raise remediation of small systematic errors disproportionately to large ones.
- Automatic remediation should recover far more per affected customer than any complaint-driven process.
- Error persistence should be longer for firm-favourable errors across system releases.
The first is the one any firm could run and almost none do. Take a random sample of accounts, audit them for errors in both directions, and count. If the residual errors are roughly symmetric, the correction process is working. If they lean heavily toward the firm, the process is asymmetric whatever anyone intended — and the fix is to correct errors the firm detects in the customer's disfavour the same way it corrects errors in its own, automatically and without waiting to be asked.
The conclusion we'd hold: the cost of contesting creates a zone where merit is irrelevant, and because correction runs differently in each direction, that zone accumulates errors in one party's favour without anyone deciding it should. The remedy is not persuading people to dispute more. It's making correction symmetrical, so that being right doesn't depend on being able to afford to say so.
Frequently asked questions
Because contesting costs more than it recovers. When expected recovery is smaller than the value of the time involved, not disputing is correct — even when the charge is plainly wrong.
Firms correct errors that cost them through their own reconciliations; customers must correct errors that cost them by contesting. Even with symmetric error rates, the survivors lean toward the firm.
Related but broader. Class waivers remove one aggregation mechanism; the contest threshold exists regardless of forum and applies to harms that would never reach any tribunal.
Lower contest costs, aggregation of small claims, and correction that doesn't depend on the customer acting — refunding detected errors in the customer's disfavour as routinely as errors in the firm's.
Key takeaways
- Below the cost of contesting, merit stops mattering — and not disputing is the correct individual decision.
- The threshold is a property of the person, not the harm: the same error is worth fighting for one customer and not another.
- Random errors plus lopsided correction produce residual errors that lean toward the firm with no intent required.
- The people for whom a small amount matters most face the highest cost of recovering it.
- Firms largely set the threshold through process design, and raising it is usually cheaper than reducing errors.
- Audit a random sample for errors in both directions; the lean is measurable and almost nobody measures it.
This report presents an analytical framework and the authors' interpretation; it is not legal or policy advice. Worked figures, including correction rates, are stylized illustrations rather than estimates, and the magnitude of any effect described depends on rates not measured here. The implications identified as testable are hypotheses.