Cost to Collect: The Fixed Cost Problem | HL Hunt
Cost to Collect: The Fixed Cost Problem
A collections operation at $41,000 fixed and $3.80 per account worked costs $6.56 per account and 20.1% of recovery. Because 42% of that is fixed, extending the queue from 40% to 85% of accounts worked cuts cost per account by 28% whether or not the added accounts recover anything. The metric improves when the operation does the wrong thing, and there is exactly one metric that does not.
What you'll learn
What it costs to run a collections operation
Cost to collect is usually quoted as a percentage of recovery, and that single number hides the thing that actually governs the decision: the split between fixed and variable cost.
Take a portfolio of 24,000 delinquent accounts in a month, of which 62% get worked, meaning 14,880 accounts receive at least one contact attempt.
| Cost | Basis | Monthly |
|---|---|---|
| Fixed: staff, management, systems, compliance, premises | Does not move with volume | $41,000 |
| Variable: dialer minutes, letters, skip tracing, data, postage | $3.80 per account worked | $56,544 |
| Total | $97,544 | |
| Cost per account worked | $97,544 / 14,880 | $6.56 |
Against $486,000 recovered in the month, that is a cost to collect of 20.1%, and 42% of the total cost does not move at all when volume changes.
The incentive the fixed cost creates
Hold the fixed cost steady and change only the share of accounts worked. The per-account cost moves a long way.
| Accounts worked | Count | Fixed per account | Total per account |
|---|---|---|---|
| 40% | 9,600 | $4.27 | $8.07 |
| 62% | 14,880 | $2.76 | $6.56 |
| 85% | 20,400 | $2.01 | $5.81 |
Working more accounts always lowers the cost per account, whether or not those accounts recover anything. That is arithmetic, not performance. A manager measured on cost per account worked can improve the metric by 28% simply by dialing further down a list that was already ordered worst-last.
And the accounts added at the margin are the worst ones. Per our prioritization analysis, a sensibly ordered queue puts the highest expected-value accounts first, which means every account added beyond the current cut-off has a lower expected recovery than everything already being worked. So the cost per account falls while the cost per dollar recovered rises.
The metric to manage against is not cost per account
It is marginal cost per incremental dollar recovered. Per our early-stage analysis, the same confusion produces the self-cure problem: an operation that works accounts which would have paid anyway books the recovery, lowers its cost per account, and has spent money to no effect. Cost per account worked, cost to collect as a percentage of total recovery, and contact rate all improve when you do the wrong thing. Only the marginal figure does not.
Where the line actually falls
The right stopping rule is simple to state. Work an account while the expected recovery from working it exceeds the variable cost of working it, and not after.
The variable cost is $3.80. The fixed cost does not enter the decision at all, because it is incurred whether or not that account is touched. So the question for each account is whether the probability of recovery multiplied by the amount recoverable exceeds $3.80.
- An account with a 4% chance of recovering $300 has an expected value of $12.00. Work it.
- An account with a 1.2% chance of recovering $180 has an expected value of $2.16. Do not.
The second account is not worthless; it is worth less than the cost of touching it. Per our review economics analysis, this is the same structure as a manual review queue, where the relevant cost is the cost per changed decision rather than the cost per case examined.
The fixed cost re-enters the decision one level up. If the marginal rule leaves a lot of capacity idle, the fixed cost is too high for the portfolio, and the response is to resize it rather than to feed it work that loses money. Per our scale analysis, an operation sized for a portfolio it no longer has will find work to justify itself.
The cost per account falls when you do the wrong thing
Working further down an ordered queue lowers cost per account by 28% and raises cost per dollar recovered. HL Hunt AI Debt Collection scores accounts on expected recovery and stops where the expected value crosses the marginal cost, instead of where the list ends.
Cost by channel, which changes the ordering
The $3.80 average conceals a wide range, and the ranking by cost is not the ranking by effectiveness.
| Channel | Cost per attempt | Note |
|---|---|---|
| Under $0.01 | Cheapest and lowest response. Per our deliverability analysis, a stale address makes it free and useless | |
| SMS | $0.01 to $0.05 | Consent requirements are the binding constraint, not cost |
| Letter | $0.55 to $1.20 | Required in some notice sequences regardless of economics |
| Outbound call | $1.80 to $4.50 per connected call | Per our dialer analysis, the cost is per connection, not per dial |
| Skip tracing | $0.40 to $3.00 | Only worth it where the balance justifies locating the person |
| Legal placement | Hundreds, plus contingency | Per our placement analysis, the commission structure dominates the direct cost |
The call line is where most of the $3.80 comes from and where the measurement error is largest. A dialer that connects on 18% of dials produces a cost per connection five and a half times the cost per dial, and an operation that budgets on dials understates its own cost by that factor.
The cheap channels are not free, because the expensive part is not the send. Per our complaint analysis, a high-volume low-cost channel generates complaint and consent exposure that scales with volume, and the cost of one consent failure can exceed the entire month's SMS spend.
The costs that are not in the budget
The $97,544 covers what someone invoices for. Several real costs do not appear there.
- Working the wrong balance. Per our balance accuracy analysis, the cost of a dispute arising from a wrong figure runs several times the cost of the contact that carried it.
- Furnishing errors. Per our furnishing analysis, a credit report correction costs more to resolve than the account was worth.
- Re-default. Per our re-default analysis, an arrangement set on the wrong basis consumes the full cost of setting it and recovers a fraction.
- Turnover. Per our turnover analysis, collections has high turnover and the ramp cost is the largest line in it, which sits inside the fixed cost and is rarely attributed.
- Quality monitoring. Per our monitoring analysis, a sample large enough to detect a rare defect is a real cost and omitting it is not a saving.
- Alert handling. Per our alert analysis, a flagging system tuned for coverage generates review work whose cost is not charged to the system that generated it.
Items one through three share a structure: they are costs created by the collections activity and paid somewhere else. That is why cost to collect measured at 20.1% is a floor rather than an estimate, and why an operation that reduces its measured cost by cutting data quality or monitoring has moved cost rather than removed it.
Fixed cost, contingency, and the make-or-buy question
Placing accounts with an agency on contingency converts the entire cost to variable. At a 28% commission, recovering $486,000 costs $136,080 against the $97,544 of running it internally, so in-house is cheaper on this month's volume.
That comparison reverses when volume falls. At a third of the volume, in-house cost is $41,000 fixed plus roughly $18,800 variable, about $59,800, against recovery near $162,000 and a contingency cost of $45,360. Contingency wins, because the fixed cost no longer has enough volume to spread across.
Which is why the honest answer is usually both
The structural argument is not that one model is better. It is that in-house carries fixed cost and therefore wants volume, while contingency carries no fixed cost and therefore costs more per dollar at high volume. An operation that keeps the segment with enough predictable volume to cover its fixed base and places the rest is making the right trade. Per our benchmarking analysis, the comparison is frequently made on headline recovery rates between portfolios of different quality, which answers a question nobody asked.
What to measure
| Metric | Why |
|---|---|
| Marginal cost per incremental dollar recovered | The only metric that does not improve when you do the wrong thing |
| Fixed and variable cost, reported separately | The stopping rule uses only the variable figure |
| Cost per connected contact, not per attempt | An 18% connect rate makes the two differ by 5.5x |
| Expected value at the current queue cut-off | If it is below the variable cost, the line is in the wrong place |
| Idle capacity against the marginal rule | Tells you the fixed base is the wrong size for the portfolio |
| Self-cure share of recoveries | Spend attributed to recoveries that needed no spend at all |
| Downstream cost of collections-caused errors | Created here and paid by another department entirely |
The fourth row is the one to put on a dashboard. Expected value at the cut-off is a single number that says whether the operation is working accounts it should not, and unlike cost per account it gets worse rather than better when the queue is extended past the point where it pays.
Cost to collect is a floor, not an estimate
Wrong balances, furnishing corrections and re-defaults are costs that collections activity creates and other departments pay. HL Hunt AI Debt Collection works from a single verified balance and records every contact, so the cost of the operation is the cost of the operation.
Frequently asked questions
About $6.56 per account worked and 20.1% of recovery on a stylized portfolio, with 42% of the cost fixed.
Because it falls by 28% as the queue extends even if the added accounts recover nothing, while cost per dollar recovered rises.
Where expected recovery drops below the $3.80 variable cost. The fixed cost does not belong in that decision at all.
In-house is cheaper at high volume and more expensive at low volume, because of the fixed base. Most operations should do both.
Key takeaways
- A stylized operation at $41,000 fixed plus $3.80 per account worked costs $6.56 per account and 20.1% of recovery.
- 42% of total cost is fixed, which is what makes cost per account worked a manipulable metric.
- Extending from 40% to 85% of accounts worked cuts cost per account 28% regardless of what those accounts recover.
- The correct stopping rule uses variable cost only: work an account whose expected recovery exceeds $3.80.
- If the marginal rule leaves capacity idle, the fixed base is too large and should be resized.
- Cost per connected contact differs from cost per attempt by 5.5x at an 18% connect rate.
- Wrong balances, furnishing errors, re-defaults and alert handling are collections costs paid by other departments.
- Contingency beats in-house at low volume and loses at high volume, which is why most operations should run both.
This guide is educational and does not constitute financial, legal, or compliance advice. All figures are stylized illustrations using assumed portfolio sizes, fixed and variable cost levels, channel costs, connect rates, recovery amounts, and commission rates, and do not describe any particular operation or portfolio. Channel costs, consent requirements, notice sequences, and permissible contact practices are governed by federal and state debt collection law and by telecommunications rules that vary by jurisdiction and change over time; consult qualified counsel and compliance advisers before changing collections practices.