Post‑Opacity

AI-assisted systems are beginning to make the €2 check possible. AI can read a contract, locate the relevant records, and turn the written terms into structured rules. Calculation software can then apply those rules, while a person reviews unclear cases and prepares evidence. That sounds like a modest improvement to auditing. It may have a much larger effect: amounts on bills that customers once accepted because checking cost too much may now be questioned and corrected.

This essay follows that change from one bill to the wider economy. It answers four practical questions:

1. When does checking a bill become worth the effort? Checking one bill for a possible €2 error is rarely sensible. Checking thousands of bills for the same error can recover enough money to cover the cost of building and running the checking system (Section 1).
2. Why might a better checking system recover less money after its first few years? At first, the system finds errors left by the supplier's old billing process. Once the supplier knows that customers are checking, the supplier has a reason to fix that process. Fewer incorrect bills are then issued, so there is less money to recover (Section 3).
3. What changes when customers can check important facts for themselves? Customers may rely less on a familiar brand or an outside checker when they can directly test whether a bill follows a contract or a product meets a stated standard. Auditors and certifiers — organizations paid to inspect records or confirm that a standard has been met — remain useful when they provide expert judgment or accept responsibility for a wrong conclusion, such as paying the resulting loss. Easier comparisons can also push suppliers toward simpler contracts whose prices are easier to calculate (Section 4).
4. What can go wrong when software makes it cheap to claim that a company owes money? A person or business seeking easy settlements can send thousands of weak or false messages demanding payment. If a targeted company must spend more to disprove each demand than the amount being demanded, paying may be cheaper than defending itself — even when the company owes nothing (Section 4).

By the end, the reader will have a practical way to examine any market: identify who calculates the amount, which written rule governs the calculation, which records are needed to repeat the calculation, what the check costs, and who can turn the result into a correction. The essay ends with a dated 2026–2032 scenario that can be compared with what actually happens.

Section 1The checking cutoff: when is a charge worth checking?

Economists have studied costly checking in auditing and tax enforcement for decades. Townsend (1979); Border and Sobel (1987); Mookherjee and Png (1989); at state scale, Kleven et al. (2011) and Pomeranz (2015). AI adds a practical question: which charges become worth checking when checking becomes cheaper?

Four numbers determine the answer:

• The cost per check is the cost of examining one more bill.
• The setup cost is the one-time cost of adapting the checker to a contract and data format.
• The success rate is the chance that a real error will be found, proved, and corrected.
• The mistake rate is the share of correct bills that the checker wrongly flags.

Together, these numbers create a cutoff called the verification frontier in the formal model. Below the cutoff, the customer expects the check to cost more than the error is worth. Above the cutoff, the expected benefit is large enough to justify the check.Gill (2026a), Proposition 2 and equation (8).

Consider a company that receives 10,000 freight invoices each year. Adapting a checker to the freight contract costs €5,000. Running it costs €0.20 per invoice, or €2,000 a year. Suppose the same €2 surcharge may appear on half of the invoices. That puts €10,000 at stake. If a valid challenge has an 80% chance of producing a refund, the expected recovery is €8,000. The €7,000 first-year checking cost can therefore make sense, even though no one would build the system for a single €2 invoice.

Two conclusions follow.

High volume spreads the setup cost. A small possible error across millions of similar bills can justify a checker that would be irrational for one bill. High-volume billing, purchasing, and claims processing should therefore change early.

The two costs affect different cases. A lower cost per check matters most when a checker already exists and processes many bills. A lower setup cost matters most for custom contracts and unusual supplier formats that previously required a software project. Observing which cases change first can reveal which cost has actually fallen.

What becomes worth checking?

The amount of money near this cutoff may be large. U.S. federal agencies reported an estimated $186 billion in improper payments for fiscal year 2025, with cumulative estimates of about $3 trillion since 2003. Many government payments are determined by written eligibility rules and supporting documents, so at least part of this activity may be suitable for the checking process described here.GAO (2026), as discussed in Gill (2026a), Section 10. The classification of these payments is the paper's interpretation, not GAO's conclusion.

When customers have historically accepted small errors, that does not prove the errors were unimportant. It may instead mean that challenging each error would have cost the customer more than the likely refund.

Section 2Finding an unusual number is not enough

Many "AI audit" products focus on detecting unusual numbers. Detection is useful, but it is only the first step after a customer decides to check a bill.

The complete process has four stages:

1. Detect: identify a possible error.
2. Prove: compare the charge with the governing rule and calculate the correct amount.
3. Prepare accepted evidence: assemble the contract, records, sources, and calculation in a form the decision-maker will accept.
4. Obtain a correction: secure a refund, corrected bill, revised contract, or other practical result.

Return to the simplified €43 electricity overcharge. Software may flag the number, but the customer still needs the correct contract clause, meter reading, market price, network fee, and calculation. The supplier must accept that package and correct the bill. If the supplier refuses, an arbitrator, regulator, or court must have the authority and willingness to act. Without those later stages, the alert has no financial effect.

How does an alert become a correction?

The papers call this the dashboard limit. A tool can display accurate alerts and still produce almost no corrections. When decision-makers rarely accept the evidence or rarely provide a remedy, almost no ordinary charge is worth pursuing, however cheap detection becomes.Gill (2026a), Corollary 1. A related regulatory result appears in Jin, Sokol, and Wagman (2026): AI-assisted monitoring can create enforcement work without reducing harm when institutional capacity is too weak.

This limit also explains why earlier promises of "trustless verification" changed less commerce than expected. Many blockchain systems required businesses to move activity onto a new technical platform. Machine-assisted checking can instead read contracts, invoices, ledgers, and meter feeds that businesses already use.Gill (2026a), Section 10, building on Catalini and Gans (2020). Existing records help only where a decision-maker will accept them and act on the result.

Section 3The possibility of checking can prevent errors

Suppose a large customer begins recalculating every energy bill. The supplier now faces a choice. It can keep a weak billing process and handle repeated challenges, or it can correct the process that produces the bills. Correcting the process may be cheaper.

If suppliers expect customers to check, suppliers have a reason to improve billing before the next invoice is issued. Customers then receive fewer incorrect charges. The checking process recovers less money because suppliers have overcharged less, not because the process has failed. The possibility of being checked changes the supplier's behaviour.Gill (2026a), Proposition 3, in the inspection-game tradition of Tsebelis (1989) and Graetz, Reinganum, and Wilde (1986). Related evidence includes third-party income reporting in Denmark (Kleven et al. 2011) and VAT paper trails (Pomeranz 2015).

A recent website example shows the same mechanism. After a German court ruled that dynamically embedding Google Fonts violated the GDPR, an Austrian lawyer used automated detection and claim generation across many websites. Within three months, non-compliance in Austria fell by 22.7 percentage points — nearly half. Most site owners changed their websites because a violation had become easy to detect and challenge, not because each owner had already lost a case.Merane and Stremitzer (2026), discussed in Gill (2026a), Section 8.1.

This behaviour changes how a checking system should be evaluated. At first, the system finds errors created under the old process, so recovered money rises. Later, suppliers improve their processes and issue fewer incorrect charges, so recovered money falls. Loss prevented can continue rising throughout.Gill (2026a), Proposition 4. Related field evidence includes U.S. tax audits (Boning et al. 2025), computerized VAT enforcement in China (Fan et al. 2020), and Medicare audits (Shi 2024).

Why can recovered money fall while prevention improves?

A dashboard that reports only recovered cash can therefore make a successful system look unproductive. Suppose annual recoveries fall from €100,000 to €20,000 after suppliers correct processes that would otherwise have produced €200,000 of new errors. The lower recovery is a success. Evaluation should track the number of bills covered, the frequency of incorrect charges, changes in supplier processes, and loss prevented — not recovered cash alone.

Section 4How cheaper checking changes trust

Businesses already use several arrangements to create trust when direct checking is expensive:

• A buyer may accept a fixed price because verifying the supplier's underlying cost is impractical.
• A familiar brand reassures buyers who cannot inspect quality directly.
• Grades and standards make products easier to compare.
• Auditors examine samples, ignore discrepancies below a chosen threshold, and return only at intervals because checking every item would cost too much.
• Escrow accounts, letters of credit, and payment rules reduce the risk of trusting a party whose performance cannot be checked immediately.
• Brokers, auditors, and certifiers perform checks or accept responsibility for the result.Klein and Leffler (1981); Barzel (1982); Diamond (1984). Gill (2026b) models these arrangements together and studies the effect of falling checking costs.

When customers can check directly at lower cost, some of these arrangements lose part of their old purpose. Trading partners can accept that some facts remain difficult to see, rely on reputation, hire an outside certifier, or check directly with software and human review.

How do businesses create trust when checking is expensive?

Some certifiers can lose clients suddenly

A certifier pays the setup cost of a checking process and spreads that cost across many clients. If clients can run their own checker cheaply, some clients leave. The certifier must then divide its shared cost among fewer remaining clients, which raises the fee per client. The higher fee encourages more clients to leave.

This cycle can keep a certification market stable for years and then cause a sudden collapse. The market does not necessarily shrink at a steady rate. Certifiers can still survive where clients value an independent certificate, specialist judgment, insurance, or a certifier that accepts legal responsibility.Gill (2026b), Proposition 2 and Appendix A.2. Recovery-audit firms and benchmarking consultancies share parts of this cost structure.

Why can a certification market collapse suddenly?

Middlemen do not disappear. Their value changes. When a customer needs an independent name, specialist judgment, insurance, or a party that can be held responsible, an intermediary still has a clear role. The intermediary is no longer paid mainly to spread the setup cost; it is paid to stand behind the result.

Brands change, but do not disappear

A brand can provide three kinds of value. First, it reassures buyers about quality: a branded company risks damaging a valuable reputation if it cheats. Second, the brand helps buyers find and remember a product. Third, the brand expresses taste, identity, or status.

Cheap checking mainly affects the first role. A small company may be able to provide credible evidence that its product meets a standard instead of spending years building a trusted name. Cheap checking does not remove the brand's value as a shortcut, identity, or status symbol.

Which part of brand value does direct evidence replace?

A smaller version of this shift may already have occurred. Online reviews are noisy and incomplete, but they give buyers evidence that was previously unavailable. As reviews spread, the price premium for hotel-chain affiliation fell by more than half between 2000 and 2015. The decline was concentrated among lower-quality hotels and hotels in smaller markets, where the chain name provided more reassurance. Premiums at the high end, where taste and status matter more, largely persisted. Another study found that a one-star increase in a Yelp rating raised revenue for independent restaurants by five to nine percent, while chains gained no similar benefit.Hollenbeck (2018); Luca (2016). Gill (2026b), Section 6.

The prediction is not that brands die. Brand value moves toward recognition, identity, taste, status, and attention. Evidence-backed reassurance moves toward the company or certifier that supplies the evidence and accepts responsibility for it.

The optimality gap: a correct bill may still come from a poor contract

The first three sections mainly concern the conformity gap: whether a supplier followed the current contract correctly. A different problem remains. The supplier may follow the contract perfectly even though a better contract was genuinely available.

Suppose an energy customer's pattern of use has changed. A business may have moved from three shifts to two, or a household may now use more energy at a different time of day. The current tariff no longer fits that pattern. The supplier has committed no billing error and breached no contract, but an alternative available at the last renewal could have met the same needs for €300 less a year. Finding that alternative would have required a €1,000 study, so the customer had no economic reason to commission the comparison.

The €300 difference is the full optimality gap, not automatically €300 of opacity rent. The comparison must use only information available when the contract was chosen and must hold service, flexibility, credit exposure, and market risk constant. Documented customer preferences and the cost of switching — including notice periods, exit fees, and paperwork — must also be separated. Only the remaining supplier benefit that persisted because a reliable comparison was too costly is an estimate of optimality opacity rent.

When comparison becomes cheap, the customer can show the supplier that a genuinely equivalent contract would cost less. A supplier that can still earn an acceptable return may reduce the price rather than lose the customer. Prices can therefore change before switching rates change.Gill (2026b), Proposition 4.

Once a €300 saving has been identified, the customer, supplier, and adviser may divide it. The customer's portion is the pass-through share. If the supplier reduces the annual price by €180, the customer receives 60% of the identified saving. The supplier retains the remaining €120, before any adviser fee.

This does not predict that electricity and gas become cheaper overall. The cost of energy, buying energy in advance to limit exposure to changing market prices, protecting against other risks, serving customers, and acquiring customers does not disappear when checking becomes cheaper. If an opacity-related gain shrinks, legitimate costs and required margins may be charged more explicitly elsewhere. The careful prediction is a change in where and how money is charged, not the disappearance of every cost or margin.

The danger: false challenges can become cheap

Tools that help customers check bills can also help challengers generate demands. The cost of defending against a weak or false demand may remain high because the target must find documents, involve managers, and sometimes hire lawyers. Accusing can therefore become cheap faster than answering. The papers call this a rebuttal asymmetry.

Suppose a challenger can generate a false €500 demand for a few euros. The targeted company knows that proving the demand false would cost €1,000. Paying €500 may be cheaper than defending the case, even when the company did nothing wrong. That difference gives the challenger a reason to target honest companies.

When can a false demand become profitable?

Cheap checking benefits customers when honest companies can also answer false demands at low cost. It can cause harm when challengers can send demands at scale to targets for whom defense costs more than settlement.Gill (2026b), Proposition 5, building on Rosenberg and Shavell (1985) and Bebchuk (1988). The Google Fonts episode produced both a genuine compliance response and judicial pushback against abusive claims. Possible responses include requiring specific evidence, giving targets inexpensive tools to assemble an answer, and making a failed challenger pay reasonable defense costs.

Automated rejection is not a complete response. A supplier might answer every machine-written demand with the same automated denial. For a genuine billing dispute, the denial leaves the faulty charge or unclear calculation unchanged, so similar disputes continue. For a false demand, the denial provides no supporting evidence and may only delay an expensive escalation.

A stronger response makes both the charge and the answer traceable. An energy supplier can identify the contract version, meter reading, market index, averaging period, network fee, and final calculation behind every bill. The customer can check a correct bill quickly, a genuine error is easier to correct, and the supplier can answer a false demand with the same records. The goal is not faster denials. It is fewer disputes.

Difficulty moves to what remains hard to check

Cheap checking does not make the entire economy transparent. When customers can verify weight, price, origin, or billing accuracy, suppliers have more reason to shift marketing — and any misleading behaviour — toward features that remain difficult to measure, such as comfort, judgment, style, or long-term quality. A contract may become easier to read while the quality of the product or service remains uncertain.Gill (2026b), Proposition 6; Gill (2026a), Section 6.3. Related evidence in taxation appears in Carrillo, Pomeranz, and Singhal (2017).

Section 5A post-opacity scenario: 2026–2032

The theory predicts an order: high-volume charges become affordable to check before custom charges, and business arrangements respond after customers adopt the new checks. The dates below are informed guesses, not model results. The order comes from the theory; the dates form a scenario. If a market lacks accepted evidence or effective enforcement, the predicted change may arrive later or not at all.

This scenario will be publicly assessed here every August, including the misses. Its credibility will come from those later assessments, not from publishing confident dates today.

Section 6What can stop the transition?

An argument this broad should state how it can fail.

A good alert may still produce no correction. A supplier or court may reject the evidence. A dispute may take years. The available remedy may be too small to justify the effort. When a checking process disappoints, researchers should identify which stage failed: detection, proof, accepted evidence, or correction.

Some suppliers will make checking harder. A supplier may split data across systems, bundle fees, add contract complexity, or provide large amounts of low-value documentation. The aim is to make checking cost more than the customer expects to recover. Difficulty can move from reading the contract to interpreting what the contract means.Gill (2026a), Section 6.3; Gill (2026b), Section 9.1; Biais and Landier (2020); Carlin (2009); Ellison and Wolitzky (2012).

Not every question has a formula and a complete record. Preferences, long-term strategy, artistic value, and much medical and legal judgment cannot be settled by the same process used to recalculate a bill. The checking cutoff moves only among charges governed by clear rules and recorded data.

An observed gap is not automatically opacity rent. A billing difference may be underbilling or may be corrected through the normal reconciliation process. A more expensive contract may provide different service, flexibility, or protection from risk, or may reflect a documented customer preference. Research must report the full gaps separately from the portion estimated to persist because verification and correction are too costly. That estimate depends on explicit assumptions about the available alternatives and the reasons customers chose them.

Private incentives can produce too much checking in one place and too little in another. An opportunistic challenger may profit from demands that transfer money without correcting harm, while targeted companies bear much of the response cost. At the same time, checking paid for by one customer can make a supplier more accurate for every customer. The paying customer cannot capture that full benefit. Private incentives may therefore encourage wasteful demand campaigns while leaving useful prevention underfunded.Gill (2026a), Proposition 5, building on Landes and Posner (1975), Shavell (1982, 1997), and Hirshleifer (1971).

Possible safeguards include ignoring discrepancies below a reasonable value, requiring specific evidence, making a failed challenger pay reasonable response costs, and protecting charges that have already passed a recognized independent check.

Section 7What should researchers measure?

The research program separates measurement from institutional change:

A proposed study of household and business electricity and gas contracts addresses the first question. Energy is a useful setting because contracts combine written formulas, metered use, published market indices, conversion factors, network charges, and later reconciliations. The proposed study has three measurement tasks:

1. Measure conformity. Recalculate bills from the contract and records. Report total errors, overbilling, underbilling, the net amount benefiting the supplier, and the amount left after the normal reconciliation process — the routine later correction of estimated or provisional bills.
2. Measure optimality. Compare the existing contract with genuinely feasible alternatives that were available when the customer made the choice. Report the full difference separately from the estimate that remains after service, flexibility, risk, preferences, and switching costs are considered.
3. Measure the checking process. Report the setup cost for each kind of contract, the cost of checking one more bill, mistaken flags, missed errors, cases in which the system cannot reach a conclusion, and cases that require a person to decide.

These measurements can estimate current gaps and checking costs. They do not establish how an institution behaves after checking becomes cheaper. A later study must examine that response—for example, by introducing checking at different times for comparable groups—and measure changes in customer choices, contracts, billing processes, prices, margins, errors, and disputes.Gill (2026a), Section 8; Gill (2026b), Section 10.

The theory makes eight predictions that evidence can prove wrong:

1. Custom checks spread. Customers adopt checking in relationships that previously relied on reputation. Middlemen valued only for sharing setup cost lose clients before middlemen that provide independent judgment or accept legal responsibility.
2. Some certifiers collapse suddenly. As clients leave, fees rise for those who remain, followed by a rapid loss of a market segment rather than a smooth decline.
3. Brand value shifts. Price premiums fall most where a brand mainly reassured buyers about measurable quality. Premiums based on recognition, taste, and status remain.
4. Recoveries rise, then fall. Error frequency declines throughout, but recovered cash first increases as old errors are found and later decreases as suppliers prevent new errors.
5. Optimality gaps shrink. Regular comparison reduces the difference between what a customer pays and what a genuinely equivalent, feasible contract would cost. It does not eliminate justified differences in service, flexibility, or risk.
6. Prices move before customers do. Existing suppliers improve offers before switching statistics change.
7. False demands depend on response cost. Weak or false demands against honest companies grow where answering remains expensive and grow less where failed challengers pay response costs.
8. Difficulty moves. Marketing attention — and any misleading behaviour — shifts toward qualities that customers still cannot measure reliably.

Researchers working on auditing, enforcement, contract theory, market design, or the economics of AI can test these predictions.

Section 8Cite this work

For the formal results, cite the working papers. The economics paper is available on SSRN with a DOI. The institutions paper is hosted here while its repository posting is completed. For the definition and public explanation of post-opacity, cite this essay.

@techreport{gill2026economics,
  author      = {Gill, Amrit},
  title       = {The Economics of Machine Verification: Verification-Cost Shocks and the Extensive Margin of Monitoring},
  year        = {2026},
  month       = {August},
  type        = {SSRN Working Paper},
  number      = {7307578},
  doi         = {10.2139/ssrn.7307578},
  url         = {https://ssrn.com/abstract=7307578}
}

@techreport{gill2026institutions,
  author      = {Gill, Amrit},
  title       = {Machine Verification and the Institutions of Trust: Reputation, Certification, and Contract when Checking Becomes Cheap},
  year        = {2026},
  month       = {August},
  type        = {Working paper},
  url         = {https://post-opacity.com/papers/institutions-of-trust.pdf}
}

@misc{gill2026postopacity,
  author       = {Gill, Amrit},
  title        = {Post-Opacity: What Happens When Checking Becomes Cheap},
  year         = {2026},
  howpublished = {\url{https://post-opacity.com}}
}

Comments, counter-analysis, and measurement collaborations are welcome: amritbir1@gmail.com.