Most cash application automation gets to about 70% touchless and stops. The clean payments match themselves; the messy remainder — the short payment with no explanation, the lump sum covering forty invoices, the remittance buried in a PDF — gets routed to a person, and the match rate plateaus. Getting from there to 90% and beyond is a different problem than getting to 70%, and it is the problem worth solving, because the last 20 points are where the cost lives. Here is how teams do it.
TL;DR
Cash application is matching incoming customer payments to the open invoices they settle. The touchless match rate is the share matched automatically with no human involvement, and it is the key measure of cash application efficiency. Most automation reaches roughly 70% touchless and plateaus, because the easy payments (one payment, one invoice, a clear reference) match on simple rules while the hard ones require reading and judgment that rule-based systems cannot do.
The last 20 to 30 points — the gap between a plateaued 70% and a strong 90%+ — are blocked by a specific set of exceptions: payments with missing or unreadable remittance information, lump-sum payments covering many invoices with no breakdown, short payments with no stated reason, deductions that need judgment to classify, payments referencing wrong or old invoice numbers, and remittances arriving in inconsistent formats (PDFs, emails, portals, spreadsheets). These are reading-and-reasoning problems, not matching-rule problems, which is why adding more rules does not get past the plateau.
Reaching 90%+ requires AI that reads remittances in any format, reasons about ambiguous payments the way a human would, and learns from each resolution so recurring exceptions stop recurring. The playbook: aggregate remittance data from all channels, apply AI that reads and interprets rather than just matches, handle the exception types deliberately rather than dumping them in a queue, let the system learn from human resolutions, and keep the underlying receivables data clean.
One caveat: the touchless rate is only meaningful alongside accuracy and the cost of the remainder. A high rate achieved by auto-matching aggressively and creating downstream errors is worse than a slightly lower rate with clean, auditable matches. For the broader AR context, see The Top AI Tools for Accounts Receivable Automation and Cash Application and Why Most Agentic AP Pilots Stall at 70% Touchless.
Why cash application automation plateaus at 70%
The pattern is consistent across finance teams: cash application automation deploys, quickly reaches around 70% touchless, and then stalls. Understanding why reveals how to get past it.
The first 70% is the easy 70%. These are the payments that fit clean rules: a single payment for a single invoice, with a clear invoice number in the remittance, in the exact amount due. A simple matching engine handles these because the match is unambiguous, the data points line up, and no judgment is required. Most cash application tools, and most homegrown automation, get here without much difficulty.
The plateau happens because the remaining payments are categorically different. They do not fail to match because the rules are not good enough; they fail because matching them requires reading unstructured information and exercising judgment, which rule-based systems cannot do. A remittance that arrives as a PDF has to be read. A lump-sum payment has to be reasoned about — which invoices does it cover, and in what amounts? A short payment has to be investigated — why is it short, and is the difference a deduction, a dispute, or an error? These are not matching problems with better rules as the answer; they are comprehension and reasoning problems.
This is why adding more matching rules does not break the plateau. Teams try, writing ever more elaborate rules to catch more cases, and hit diminishing returns fast, because the hard cases are not rule-shaped. The plateau is structural to rule-based matching, and getting past it requires a different capability: AI that can read and reason, not just match. This is the same dynamic that causes AP automation to stall at similar levels, analyzed in Why Most Agentic AP Pilots Stall at 70% Touchless.
The anatomy of the exceptions that block 90%+
To get past the plateau, you have to know exactly what is in the unmatched remainder. The last 20 to 30 points are made up of a recognizable set of exception types, each requiring reading or reasoning.
Missing or unreadable remittance information. The payment arrives but the information about what it pays does not, or it arrives in a form a matching engine cannot parse. The match requires finding or reading the remittance, often in a separate email, attachment, or portal.
Lump-sum payments covering many invoices. A customer pays one amount for forty invoices with no breakdown. Matching requires reasoning about which combination of open invoices the payment settles, sometimes with a discount or short payment mixed in.
Short payments with no stated reason. The payment is less than the invoice. Resolving it means determining why: a deduction, a dispute, a partial payment, or an error, each of which is handled differently.
Deductions needing classification. A deduction might be a valid trade promotion, an agreed allowance, a freight charge, or an unauthorized chargeback. Classifying it correctly requires judgment about the customer, the agreement, and the context.
Wrong or old invoice references. The payment references an invoice number that does not exist, is already paid, or is mistyped. Matching requires inferring the correct invoice from other clues.
Inconsistent remittance formats. Remittances arrive as PDFs, emailed spreadsheets, EDI files, bank-portal data, and scanned documents, each in a different layout from each customer. A rule built for one format fails on the next.
The common thread is that every one of these requires reading unstructured data or exercising judgment, which is precisely what rule-based matching cannot do and what AI reasoning can. The composition varies by business — a company selling into large retailers has more deductions; one with many small customers has more format variety — but the categories are consistent, and naming them is the first step to automating them.
The playbook: how teams reach 90%+ touchless
Getting past the plateau is a deliberate process, not a matter of buying a tool and hoping. Five moves, in roughly this order, are how teams reach and sustain 90%+ touchless match rates.
1. Aggregate remittance data from every channel
Much of cash application difficulty is that the payment and its remittance arrive separately and through different channels: the payment via bank feed, the remittance via email, portal, or attachment. The first move is to bring all remittance sources together so the matching process has the information it needs in one place. Without this, even good AI is working with incomplete data, and the touchless rate is capped by missing remittance information.
2. Apply AI that reads and interprets, not just matches
This is the core move. Replace or augment rule-based matching with AI that reads remittances in any format, extracting the relevant data from PDFs, spreadsheets, scans, and emails, and that reasons about ambiguous cases rather than routing them to a queue. This is what addresses the comprehension-and-judgment exceptions that constitute the plateau. The distinction matters: an OCR tool that reads a remittance but cannot reason about a short payment only solves part of the problem, while AI that both reads and reasons addresses the full exception set.
3. Handle each exception type deliberately
Rather than treating the unmatched remainder as one undifferentiated queue, handle the exception types deliberately: configure how lump sums are reasoned about, how short payments are investigated, how deductions are classified. Naming and addressing each category, as in the anatomy above, is what systematically converts the remainder to touchless, rather than leaving it as a generic exception pile a human works through.
4. Let the system learn from human resolutions
When a human does resolve an exception, that resolution is valuable data. A system that learns from it, applying the same reasoning to future similar cases, converts each human resolution into future automation, so the exception queue shrinks over time rather than staying flat. This is the difference between a static automation that holds at a fixed rate and one whose touchless rate climbs as it learns your customers’ patterns. It is also what makes 90%+ sustainable rather than a one-time configuration achievement.
5. Keep the underlying receivables data clean
AI cash application reasons against your open receivables, so if that data is stale, duplicated, or inconsistent, even good AI matches poorly. Keeping the receivables ledger clean and current, through good reconciliation discipline, is the foundation that lets the matching AI perform. This connects cash application to the broader reconciliation work covered in The Best AI Reconciliation Software for Mid-Market Finance Teams and the matching discipline in Best Software for Automated Bank Statement Matching.
Why architecture matters: reaching 90%+ the right way
There is a wrong way to hit a high touchless rate, and it is worth naming, because the headline number can hide it. A system can be tuned to auto-match aggressively, forcing matches on ambiguous payments to push the rate up, which creates downstream errors: misapplied cash, wrong invoices closed, and reconciliation problems that surface later and cost more to fix than the manual match would have. A 92% touchless rate achieved this way is worse than an honest 85%.
This is why the touchless rate has to be evaluated alongside accuracy and auditability. The right way to reach 90%+ is with AI whose matches are correct and explainable, where you can see why each payment was matched as it was, and reconstruct the decision if an auditor or a customer questions it. Because cash application feeds revenue recognition, the matches are not just operational — they have accounting and audit consequences — so a fast but opaque or error-prone high match rate is a liability. (On why a confidence score is not the same as an audit trail, see When Confidence Scores Lie.)
This is where deterministic, reasoning-based AI fits the problem well. Kognitos approaches cash application as agentic automation that reads messy remittances and reasons about the exceptions — short payments, lump sums, deductions — in plain language, applying each resolution to future matching and logging every decision with the reasoning behind it. Because it is deterministic, the same payment data produces the same match every time, and because the reasoning is explicit, every match is explainable and auditable. That combination — reading and reasoning to get past the plateau, plus determinism and auditability so the high match rate is trustworthy — is what reaching 90%+ the right way requires. The point is not a tool; it is that the way you reach a high touchless rate matters as much as the number, because cash application feeds the financial statements. (More on the architecture: What is Neurosymbolic AI? and What is English as Code?)
Book a working session with a Kognitos solutions engineer → Try Kognitos free →
The payoff: why the last 20 points are worth it
Closing the gap from a plateaued 70% to a strong 90%+ delivers value out of proportion to the 20-point difference, because the last 20 points are the expensive part. They are the payments consuming the most human time per item, the ones causing unapplied cash that inflates DSO, and the ones generating the customer friction of collections chasing already-paid invoices. Automating them frees the most time, clears the most artificial DSO, and removes the most friction.
The DSO connection is direct: payments that sit unapplied because they are hard to match keep receivables artificially high, and clearing them through better cash application lowers DSO, which frees cash — roughly $2.7 million per day of DSO for a billion-dollar-revenue company. So the touchless rate is not an operational vanity metric; it ties directly to cash and working capital, which is why getting past the plateau is worth the deliberate effort. The full DSO connection is in How to Reduce DSO with AI: A 2026 Playbook, and the related efficiency measure in Accounts Receivable Turnover: How to Calculate It and Improve It with AI.