Flagging is not fixing

Every week, the AISideProjects corpus turns dozens of podcast episodes, newsletters, and IdeaBrowser editions into a shortlist of buildable ideas, scored and ranked. In October 2026, the top of that ranking kept landing on the same shape of business. The product is an agent that finds money a business is already owed, then does the unglamorous work of collecting it: filing the claim, attaching the proof, following up until someone pays or refuses in writing.

That shape won twice in two weeks. The restaurant invoice agent ranked first in the October 26 shortlist at 22/25, and the support-bot reconciliation tool ranked first in the October 19 shortlist, also at 22/25. A third entry, detention-fee recovery for trucking fleets, scored lower, and the reasons it scored lower are as informative as the wins.

Three workflows, one leak

A Buyers Edge audit of 11,000 invoices across 400 restaurants found overcharges on 35% of invoices, worth about 1% of invoice dollars. One percent sounds small until it sits next to the median full-service restaurant margin of 2.8% before tax (National Restaurant Association, 2025, cited in the source). A leak equal to a third of the typical margin stays unfixed because checking each invoice by hand costs more than the error on that invoice. The proposed agent reads every supplier invoice against the contracted price, documents the difference, submits the claim, and follows up with the distributor. Its price is 25% of whatever it recovers, so the customer pays only when money comes back.

Trucking has the same leak in a different uniform. Fleets earn detention fees when a truck waits too long at a dock, and the American Transportation Research Institute has measured average detention rising by roughly 30 minutes per stop since 2014. Small carriers, the target here is fleets of 5 to 100 trucks, routinely write the fees off: filing needs timestamps, contract terms, and repeated follow-up, and dispatchers have freight to move. The source puts direct unpaid detention at $3.6 billion a year. The proposed agent pulls arrival and departure times from the fleet's own tracking data, reads the free time and hourly rate from the broker's contract, files the claim the next day, and chases on days 7, 14, 21, and 30 until the broker pays or refuses in writing.

Ecommerce support is the third leak, and it is newer. Gorgias put an autonomous support agent in front of more than 2,000 brands in under a year, resolving 35 to 60% of tickets at $0.90 to $1.00 per resolution. The mistakes are concrete: a wrong refund amount, a duplicate refund, the wrong order cancelled, a subscription reported as paused that never paused, tracking promised and never created. The Cloud Security Alliance figures cited in the source say 53% of organisations saw AI agents exceed their permissions in the past year, and 31% of agent-adopting enterprises had an agent change money or rewards based on policy it invented. Ravelin estimates $1.5 billion a year lost by North American e-tailers to refund and policy abuse.

Why incumbents stop at the dashboard

The restaurant case names the pattern directly. MarginEdge, xtraCHEF, Restaurant365, and Ottimate all capture line items; Craftable goes furthest, matching invoice lines to contracts at SKU level and flagging overcharges before payment. None of them opens a claim. Enterprise recovery firms such as PRGX run contingency audits at 10 to 40% for top-20 retailers and manufacturers and have no restaurant practice at the 5 to 100 location tier. In the source's own summary: everyone flags the overcharge, nobody chases the credit.

There are sensible reasons incumbents stop there. Flagging is software: clean margins, no adversarial counterparty, no per-claim labor. Chasing puts the vendor in a dispute with the customer's own suppliers, in the restaurant case with distributors the restaurant needs again next week. A dashboard also sells to the person who owns the budget today, while a recovery service has to prove dollars before it earns them. The gap is structural, and structural gaps are where a solo builder can enter: the incumbent is not failing at the last step, it is declining it.

The exception: when flagging is the product

The support-bot tool breaks the pattern on purpose, and the design choice matters. It is read-only. Every night it cross-checks what the support agent promised in Gorgias or Zendesk against the records in Shopify, Stripe, and Recharge, and it surfaces only the mismatches in an exception inbox for a human to fix. It never touches a money system.

That restraint is the wedge. A brand burned by an agent issuing a full refund against a no-refund policy will not hand a second agent write access to its refunds. Verification earns trust precisely because it cannot act. The Cognitive Revolution's governance conversation points the same way from the oversight side: the episode's GovAI guest notes that monitoring of agents is mostly after the fact, that synchronous monitoring, watching in real time with the ability to block, is still the open gap, and that agent identifiers remain underdeveloped. Detection is unfinished work in support tooling because nobody was verifying money movement at all; in the restaurant and trucking cases, detection was already solved and the missing step was collection.

So the rule is not "always close the loop." The rule is: find the step where value stops, and check whether that step is missing because it is hard and adversarial (collect it) or because trust has not been earned yet (verify first, act later).

Where the sources disagree

The three cases do not agree on how much autonomy is safe, and the shortlist scores show the disagreement. Detention recovery is the most aggressive design, an agent conducting a drawn-out collections argument with brokers, and it ranked Watch at 17/25 rather than Build. The source itself names the risk: brokers can blacklist carriers who push claims too hard, and the relationship damage lands on the customer, not the software. It also puts the window at roughly 12 to 18 months before a platform or factoring incumbent bundles detention recovery into an existing workflow.

The ceilings deserve the same skepticism. IdeaBrowser tags all three ideas at roughly $5M to $15M in annual revenue potential. These are solid, service-shaped businesses with claim volume tied to a customer's transaction volume. That is a feature for a small team and a hard limit for anyone expecting software multiples on effort. Contingency pricing, 25% of recovered credits or 20% of back-claims in the detention case, aligns the fee with proof, and it also means revenue arrives only after the chase succeeds.

What this means for a solo builder

• Pick a leak where the per-item value is small and the volume is large. If a human can profitably check one item by hand, the agent has no opening.
• The proof burden is the pitch. These products sell recovered dollars, a number the customer can reconcile against their own accounts, which is a shorter argument than selling insight.
• Design the follow-up cadence as a product feature, not an afterthought. Days 7, 14, 21, and 30, with templates per counterparty, is where the restaurant and trucking designs actually live.
• Earn write access in stages. Start read-only where trust is low, price on outcomes where the counterparty is a third party rather than the customer's own systems.
• Watch the bundling window. HappyRobot raised $150M in August 2026 for freight AI agents; when a funded platform adds your claim type as a checkbox, the standalone window closes.

What would change this conclusion

Two developments would break the pattern. First, if the AP and support platforms ship claim-filing and cross-system verification themselves, and their distribution makes the standalone wedge worthless, the correct move becomes building on top of them rather than beside them. Second, if early recovery products show that counterparties simply refuse agent-filed claims at scale, by policy rather than by friction, then flagging was never a halfway step and the dashboard vendors had it right. Neither has happened yet in the sources tracked here. Until one does, the money is in the chase.

All articles · Research home