Back to IntelligenceDebtor Behaviour

    SIGNALS BRIEF: Adversary Back-Office AI Capability Assessment — Q2 2026

    Marcus EllertonMarcus Ellerton
    ·22 Jul 2026·8 min read

    CLASSIFICATION: UNRESTRICTED — CREDITOR DISTRIBUTION AUTHORISED
    PREPARED BY: M. ELLERTON, SIGNALS DESK
    REPORTING PERIOD: MAY 2026


    1. EXECUTIVE SUMMARY

    Your debtor's accounts payable department has acquired new capabilities. Not a new clerk. Not a new ERP module. A language model that drafts their dispute letters, their payment-plan counterproposals, and — we assess with high confidence — their excuses.

    This brief analyses newly published data from the Anthropic Economic Index (May 2026 snapshot), which maps observed AI usage against the task catalogues of 718 occupations across 121 countries. We have extracted the findings relevant to one question only: when your invoice goes unpaid in Frankfurt, Lyon, Manchester or Chicago, how likely is it that the correspondence stalling you was machine-drafted — and what should you do about it?

    Key judgments:

    • Finance-adjacent task categories are firmly inside the AI perimeter. Globally, tasks commonly done by Office & Administrative Support occupations account for approximately 7.9% of observed AI usage; Business & Financial Operations tasks for a further 5.8%. The clerk who processes your invoice and the analyst who disputes it both sit in heavily AI-touched territory.
    • Correspondence is the primary deliverable. "Email or message" outputs run above the global baseline in all four major creditor markets analysed. The battlefield of B2B collection — the letter, the reply, the artfully vague promise to pay — is increasingly automated terrain.
    • Adoption is uneven across jurisdictions. This asymmetry is actionable. Details in Section 3.

    Methodology note, because this desk does not launder statistics: the Index measures conversation content matched to occupational tasks. It tells us AI is used for tasks commonly done by, say, financial clerks. It does not tell us what percentage of financial clerks use AI, and it is a snapshot, not a trend line. Full methodology: anthropic.com/economic-index.


    2. THE CAPABILITY: WHAT THE ADVERSARY'S FINANCE DESK NOW AUTOMATES

    The Index classifies each AI interaction as augmentation (the human collaborates, iterates, stays in the loop) or automation (the task is delegated wholesale — drafted, done, sent). Globally the split is 51% augmentation to 49% automation. Nearly half of observed AI work usage is full delegation.

    Translate that to the AP department holding your money:

    Task family: financial correspondence. "Advise clients or respond to inquiries about financial matters via phone, email, or chat" ranks among the top eight most AI-matched work tasks in the United States, the United Kingdom, and France. This is the exact task family that produces the reply to your dunning letter.

    Task family: information retrieval. Searching databases and reference material is the single most AI-matched task in every market analysed. Operationally: when your debtor's clerk checks the contract terms, the delivery record, or the interest provisions of your jurisdiction before deciding how long they can stall you, they now do it in seconds, with citations.

    Output family: the email. Across the US, UK, Germany and France, "email or message" as an AI output runs consistently above the global baseline of 4.6% — reaching 5.9% of outputs in the United States.

    Field implications

    The stalling letter you received last Tuesday — the one that acknowledges the invoice "in principle," raises a previously unmentioned quality concern, and proposes a review call in three weeks — may have taken ninety seconds to produce. It is grammatically immaculate. It cites your own payment terms back at you. It was possibly generated between two other tasks by someone who has never read your contract.

    This changes three things:

    1. Volume asymmetry is dead. The old attritional advantage — a creditor who writes more letters than the debtor can answer — is gone. The debtor's reply capacity is now effectively infinite and effectively free.
    2. Polish is no longer a signal. A well-drafted, legally literate dispute letter used to indicate the debtor had escalated internally or engaged counsel. It now indicates the debtor has an internet connection. Do not recalibrate your risk assessment upward because the correspondence reads professionally. Recalibrate on behaviour: payment, part-payment, documentation — not prose.
    3. Generic pressure gets absorbed. Templated dunning sequences are precisely what a machine answers best. What machines answer poorly: a collector who telephones, who references the specific shipment, the specific individual who signed the delivery note, and the specific consequence arriving on a specific date. Specificity is the counter-capability.

    Intercol Signals Desk · Q2 2026

    Adversary Capability Map

    AI adoption inside debtor back-offices — four theatres.

    USA3.87GBR3.35FRA3.97DEU2.40

    Intel · France

    Rank #1 / 121

    3.97Usage Index

    Augmentation 58% · Automation 42%

    Work-usage share: 5.4%

    HIGHEST DELEGATION POSTURE — full-spectrum automated evasion.

    Every excuse now arrives grammatically flawless.

    Hover a country to inspect. Default view: highest-index theatre.

    2.00
    4.00

    Anthropic Usage Index, May 2026 — share of AI usage vs share of working-age population. 1.0 = proportional.

    INTERCOL · SIGNALS DESK

    anthropic.com/economic-index

    3. THEATRE ASSESSMENTS: FOUR CREDITOR MARKETS

    One definition before the theatre reports. The Anthropic Usage Index is a country's share of observed AI usage divided by its share of working-age population. A score of 1.0 means usage exactly proportional to population; 2.0 means double what population alone would predict. All figures: May 2026.

    🇺🇸 UNITED STATES — Index 3.87, rank #12 of 121

    Assessment: HIGHEST DELEGATION POSTURE.
    The US shows the most automation-leaning profile of the four theatres: roughly a 50/50 augmentation-to-automation split, against 55/45 in the European markets. Business & Financial Operations tasks run at 6.3% of US usage — above the global baseline — and "advice or recommendation" outputs run markedly hot (15.1% vs 10.7% baseline). Assessment: US debtor correspondence is the most likely of any theatre to be fully machine-generated with minimal human review. Counter-measure: force the human into the loop. Requests that require a signature, a call, or a decision by a named officer strip away the automation layer.

    🇫🇷 FRANCE — Index 3.97, rank #10 of 121

    Assessment: HIGH ADOPTION, COLLABORATIVE DOCTRINE.
    France ranks highest of the four overall — top ten globally — but runs a 54/46 augmentation split: the human stays in the loop. Financial-inquiry tasks appear in the top-eight matched tasks here too. Assessment: expect AI-assisted rather than AI-generated correspondence — faster, sharper, but with a human owner who can be reached, persuaded, and held to what was written.

    🇬🇧 UNITED KINGDOM — Index 3.35, rank #16 of 121

    Assessment: HIGH ADOPTION, CORRESPONDENCE-FORWARD.
    The UK profile runs "email or message" outputs at 5.8% (baseline 4.6%) and financial-inquiry tasks in the top matched set. Augmentation-leaning at 55/45. Assessment: the UK debtor's finance desk is well-equipped to conduct long, polite, beautifully formatted delay campaigns. The countermeasure is the same one this desk has recommended since before the machines arrived: shorten the correspondence phase. Escalate on schedule, not on tone.

    🇩🇪 GERMANY — Index 2.40, rank #26 of 121

    Assessment: LAGGING ADOPTION — ASYMMETRY FAVOURS THE PREPARED CREDITOR.
    The contrarian finding. Germany — reflexively assumed to run the most automated back office in Europe — trails France, the US and the UK substantially, at #26 globally with an index of 2.40. Work-related usage sits at 38% of German AI activity, below the 43% global norm. Assessment: the German AP department is, on this evidence, more manual than its reputation suggests. Two implications. First: German processing and dispute cycles still run at human speed — build that latency into your escalation timeline rather than misreading it as evasion. Second: a creditor operating with AI-assisted case preparation holds a genuine tempo advantage in this theatre. Use it.


    4. RECOMMENDED CREDITOR POSTURE

    1. Assess behaviour, not prose. Payment actions, not letter quality, drive risk scoring. The prose is free now.
    2. Weaponise specificity. Named individuals, exact shipment references, dated consequences. Machines deflect the generic; they cannot deflect the particular.
    3. Compress the letter-writing phase. Every additional round of correspondence now costs the debtor nothing. Escalate on the calendar, not on the reply.
    4. Force human checkpoints. Signatures, calls, named-officer decisions — anything a language model cannot supply on the debtor's behalf.
    5. Mind the German exception. Slower ≠ stalling. Calibrate per theatre.

    Or delegate the entire problem to a desk that reads these signals professionally.

    Contact Us — Free Review


    Source data: Anthropic Economic Index, May 2026 snapshot (anthropic.com/economic-index). Figures describe observed AI usage matched to occupational task catalogues; they measure tasks, not people, and constitute a snapshot, not a trend. This desk reports what is observed. Interpretation, as ever, is a matter of tradecraft.

    Marcus Ellerton

    Written by

    Marcus Ellerton

    Director, Market Intelligence

    Marcus leads Intercol's market intelligence function, tracking corporate debt exposure, insolvency trends, and payment behaviour patterns across European and North American markets. Before joining Intercol, he spent twelve years in credit risk analysis at two of London's largest institutional lenders, where he built early-warning models for corporate distress that were adopted across their commercial lending divisions. He created The Turbulence Report™ series — Intercol's research programme that maps the gap between what companies say in annual reports and what their balance sheets actually show. His work has covered cases from Carillion to Volkswagen, using only officially filed data to identify the patterns that precede payment failure. Marcus holds an MSc in Financial Risk Management from ICMA Centre, Henley Business School. He writes about industry risk, corporate debt analysis, and the signals that credit departments miss.

    AI in debt collectionAnthropic Economic IndexB2B debt recoverydebtor behaviouraccounts payable AIintelligence briefing
    Share
    ATTENTION PLEASE

    Ready to recover your receivables?

    Would all passengers holding outstanding invoices please proceed to the assessment desk. A specialist will review your receivables, confirm what's recoverable, and present your recovery options — at no cost and no obligation.

    Response in 24 hours
    💷Cost: £0
    40+ jurisdictions
    Contact

    This is the final call for outstanding receivables.

    NEW — Get a $/£/€500 fee credit on your first caseClaim now →