The State of Play

AI-driven expense management and reconciliation is a proven capability with a persistent implementation problem. The technology works: enterprise platforms routinely deliver 70-80% reconciliation time savings, near-perfect matching accuracy, and documented six-figure annual ROI. Vendor ecosystems are mature, analyst recognition is broad, and GA products from SAP Concur, Brex, and Trintech compete on integration depth and agentic automation. The question is not whether the tooling delivers value—it does—but whether organizations can muster the change management, data quality, and governance discipline to realize that value. Nearly half of finance departments still operate without any automation, and a widening gap between adoption rates and measurable productivity gains reveals that technology maturity has far outpaced organizational readiness. For teams prepared to invest in rigorous implementation, this is a confident buy. For those expecting plug-and-play ROI or struggling with data quality and governance capacity, the evidence counsels caution or deferral.

Agentic AI is now in production deployment across enterprise and mid-market segments, with major SMB inflection beginning. SAP Concur's Joule Expense Automation Agent and Pre-Submit Audit Agent reached GA in June 2026, with integration into Microsoft 365 Copilot enabling expense tasks directly within Outlook/Teams. Intuit released automated AI expense categorization to all QuickBooks Online Small Business plans (US, June 25, 2026), marking inflection in SMB-level adoption. Ramp (50K+ customers, $1B 2025 revenue) deployed Smart OCR with auto-vendor correction and 90%+ auto-coding (3.5x more transactions processed, 3x faster close). Microsoft released native AI Expense Agent for Dynamics 365 Business Central (April 2026) signaling tier-1 ERP platform commitment to embedded expense automation. Deployment outcomes validate the ROI thesis: Genpact's agentic AP suite for a 3.5M invoice/year global company delivered 7%→65% touchless processing, 18-29→9-14 day cycle, and identified $350M in duplicate invoices. Ramp cases (April 2026) show ABB Optical Group reduced audit prep from 2 months to 1-2 days (85-90% error reduction, $2M+ savings); Brex maintains 35,000+ customers with 99% zero-touch processing and 99% OCR accuracy; Trintech processes 150M+ daily transactions at 99%+ auto-match. Deloitte Q2 2026 controller survey confirms reconciliation adoption has moved mainstream: 44% of controllers deployed AI in accounting operations (up from 7% in 2023), with reconciliation automation at 61% adoption rate and a median close cycle improvement of 3.2 days (28% reduction). August 2026 updates: Payhawk achieves 68.6% fully autonomous retrieval success rates with 12x speed improvement over manual, reducing vendor invoice collection from 23.8→12.6 days; Moss (Berlin spend-management platform) reaches €1B+ valuation with 5,000+ customers and €70M+ ARR processing 2M+ transactions/month via production AI agents; Emburse data shows 48% of eligible enterprises now have recurring AI spend (up from 37% three years ago), with agents/workflow apps representing 37.5% of expense AI spend. AllPoints Fibre Networks consolidated fragmented expense stack (SAP Concur, ZoneCapture, shared cards) to unified Payhawk platform, eliminated 30 NetSuite licenses, achieved 100% receipt capture, and saved 3 hours weekly on manual uploads. Japan market maturity (12-year SAP Concur dominance, 42.8% vendor share) demonstrates sustained enterprise acceptance across regions and company sizes. Regional expansion: Zaggle deployments in India show 1-2 day reimbursements and 40% manual reduction.

Governance and ROI accountability are now the primary adoption barriers, having displaced technology readiness. KPMG June 2026 survey shows only 7% of financial leaders report established AI ROI; 24% face active investor pressure to prove value; 42% lack visibility into AI spending. This mirrors broader pattern: Gartner reports 50% of generative AI projects abandoned after PoC due to token economics cost surprises (Uber and Walmart imposed hard caps after burning budgets), poor data quality, and weak governance infrastructure. KPMG May 2026 research found 75% of companies use AI in finance, but only 42% can audit their AI decisions—a material assurance gap. Grant Thornton's June 2026 assessment notes that AI shifts audit risk from traditional manual errors to algorithmic concerns: model bias, data quality opacity, and incomplete training data. Responsibility redistributes from direct human execution to system oversight, requiring auditors to validate AI-generated outputs. This reframes the practice: technical capability (95%+ categorization accuracy) no longer determines adoption; governance maturity does. Quality failures have emerged: Thomson Reuters investigation (August 2026) documented AI-generated errors in corporate finance—GPTZero identified fake footnotes and fabricated claims in PwC financial reports; Deloitte Australia refunded AU$98K (20% of contract value) after AI errors were found; similar incidents at EY and KPMG forced report retractions—signaling that AI systems require rigorous verification gates before publishing and that categorization errors compound in complex reporting workflows. BearingPoint's Q2 2026 CFO survey reveals the execution barrier: 73% of CFOs describe AI adoption as minimal or basic, and only 9% report scaling as expected—despite widespread piloting. The structural gap persists: Rossum data shows 49% of finance departments still operate with zero automation, and only 27% of organizations use AI in spend management (59% none). CFO intent remains high (51% plan AI spend), but only 25% of pilots scale to production. Payhawk June 2026 research confirms the integration gap: only 14% of organizations have successfully consolidated spend controls, consolidation, and automation—meaning 86% operate with fragmented systems or no automation. Critical limitation: independent chartered accountant assessment confirms that automated categorization is reliable only on recurring transactions; it fails on edge cases, VAT logic, and expense-vs-fixed-asset distinctions, requiring ongoing governance oversight. Practitioner accountability friction has surfaced: when AI misclassifies expenses, clients hold bookkeepers liable (not software vendors), creating friction for service providers and revealing that permanent human review remains required despite automation. An emerging risk has surfaced: AI-generated receipt fraud. SAP Concur data shows 67% of CFOs perceive AI receipt fraud as likely in their organization, with 15-20% of expense reports already containing non-compliance. This signals that expense automation systems must now embed fraud detection and governance controls as core requirements, not bolt-on features. For organizations with strong data foundations, governance discipline, and change management capacity, documented ROI is clear: 3-9 month payback, 111% first-year ROI, and 60-80% touchless processing. For those lacking these prerequisites—particularly CFOs under budget scrutiny—the gap between adoption ambitions and realized impact widens.

• 2018: AppZen emerged as category leader with 650+ enterprise customers by October, including 25 Fortune 1000 firms and named deployments at major companies (Amazon, Citi, Airbus). Concurrent growth in OCR-powered receipt automation across travel-and-expense suites (Concur, Expensify). Market penetration remains limited—industry survey in November showed nearly 50% of companies still fully manual—indicating significant adoption barriers around integration, policy definition, and user adoption.
• 2019: Continued vendor expansion and adoption validation. AppZen expanded into invoice and contract audit; SAP Concur documented case studies showing 87% time savings in mobile receipt submission (PoHwer). Industry surveys show growing consensus: 80% of fraud examiners support AI/ML in T&E programs, 44% of capital markets firms have reconciliation AI live or in POC, 75% identify reconciliations as highest AI-ROI function. Adoption barriers persist: 73% of mid-market companies still fully manual, SMBs face cost and integration hurdles. Emerging specialists (Reconcilio, SmartStream Air) demonstrate ecosystem maturation in reconciliation-specific automation.
• 2020: Market consolidation and feature maturity. AppZen reached 1,650+ customers (one-third of Fortune 500) with sustained double-digit revenue growth and product expansion into revenue and billing. SAP Concur and competing vendors released GA AI features for receipt capture and fraud detection (ExpenseIt, Concur Detect). Industry research from Big 4 firms confirmed widespread adoption of AI in financial processes, though implementation barriers persisted—surveys indicated lack of off-the-shelf software, data quality challenges, and organizational change management obstacles remained significant hurdles to broader deployment outside enterprise segment.
• 2021: Continued enterprise adoption and pandemic recovery. AppZen maintained market leadership with presence across one-fourth of Fortune 500; CFO surveys showed renewed confidence in finance automation spending post-pandemic. SAP Concur stabilized revenues as business travel recovered; Oxford Economics survey of 500+ global business leaders confirmed spend automation ROI and economic recovery drivers. SMB market expanded with products like Expensify deepening accounting system integrations. Barriers to adoption remained persistent: organizational change management, system integration complexity, and cost-of-implementation still limited deployment outside enterprise and larger SMBs.
• 2022-H1: Market bifurcation visible: enterprise adoption validated by 628% three-year ROI (IDC) and ecosystem integrations (Amadeus-AppZen 19-second audits), yet mainstream penetration remained limited (1-in-4 companies with any automation, 2-in-10 with AI fraud detection). Forrester analyst report highlighted structural barriers (data quality, skills gaps, decision transparency concerns). Broader AI adoption plateau (26% of organizations, O'Reilly) suggested expense-specific growth saturation outside enterprise. Reconciliation identified as highest-perceived ROI function but fragmentation in bank connectivity and accounting standards slowed rollout.
• 2022-H2: Expense fraud recognized as persistent problem with $31,000 annual average loss per company, yet AI adoption for fraud detection remained limited despite clear ROI case. Vendor product maturity continued with integration certifications (Trintech-SAP), though analyst consensus underscored structural adoption barriers (data stability, skills gaps, organizational hesitation about black-box compliance decisions) limiting growth outside Fortune 500 segment.
• 2023-H1: Enterprise deployments mature with documented ROI: Trintech Cadency V10.3 achieves SAP certification enabling 500-hour monthly reconciliation savings (Serco), Payhawk reduces manual reconciliation from 32 hours weekly to near-zero through OCR/ERP integration. Post-COVID spending recovery drives broader violation detection needs. Yet mainstream adoption remains limited: survey data shows 49.7% of workers still don't file expenses due to friction (79% cite time-consuming processes, 32% use paper receipts). Implementation risks highlighted include data accuracy gaps, regulatory compliance complexity, and transparency/explainability concerns. Adoption concentrates in large enterprises; SMBs/mid-market held back by integration complexity and organizational change barriers.
• 2023-H2: Vendor platform maturity advances with SAP Concur reporting 51,000+ customer base delivering $54K annual ROI and 47% reduction in missing receipts; Brex demonstrates AI-enabled spend controls and autonomous receipt generation. Market projections accelerate: global expense management market projected to grow from $8.53B to $15.79B by 2032. Finance team expectations shift: 95%+ categorization accuracy and real-time visibility become baseline rather than differentiator. Yet structural adoption barriers persist—almost half of workers still don't file expenses, and mainstream deployment concentration remains limited outside Fortune 500. Manual processes continue creating compliance gaps and operational friction.
• 2024-Q1: Vendor product evolution continues with AppZen launching Coach Insights for anomaly detection and manager visibility. Market adoption remains fragmented: survey of 200+ finance professionals shows only 2% increased adoption of expense management software despite rising corporate card spending, revealing persistent barriers around friction, time burden, and policy gaps. Implementation risks surface around cloud cost escalation and FinOps governance, flagging that AI deployment requires strategic financial alignment to avoid cost overruns.
• 2024-Q2: Mid-market deployments continue with United Engineers and Constructors achieving $260K savings over 2 years and sub-1-hour reconciliation via Navan; La Trobe University scales to 1,200+ users on Expensify; Payhawk customers (AIOPSGROUP, ATU) report 2-hour daily savings and €2M VAT recovery. CFO survey shows 51% investing in AI for financial processes, with Gartner predicting 80% enterprise GenAI adoption by 2026. Yet adoption barriers remain: survey of 200+ finance professionals shows only 2% increased adoption despite rising corporate card spending.
• 2024-Q3: Vendor ecosystem integrates deeper into cloud ERP platforms (SAP embeds Concur into GROW premium edition). Mid-market momentum confirmed: CRE Inc. achieves 80% time reduction via Concur; Merck and Chobani report 90% audit time cuts and 60% cost reduction via AI auditing. Finance AI adoption surges: 58% of finance functions deployed AI (Gartner, +21 points YoY), with 39% using anomaly/error detection; 51% of CFOs investing in spend AI (up from 15% in Aug 2023). However, knowledge gaps and implementation risks emerge: 58% of CFOs lack AI knowledge, 34% of travel managers unprepared for tools, 30% of GenAI projects at risk of abandonment due to poor data quality and unclear ROI. Mainstream adoption remains concentrated in enterprises and willing mid-market firms.
• 2024-Q4: Digital platform adoption accelerates: 64% of businesses transitioned to digital expense platforms; market valued at $13.9B (2025) expanding to $31.9B by 2032 at 12.6% CAGR. Mid-market deployments continue: HTS Group Ltd achieves 87% error reduction and 30 hours monthly savings with Capture Expense; broader survey shows 59% of enterprises leveraging expense software for cost control. Persistent barriers remain: 45% of companies still use manual reporting despite wanting automation, 49.7% of workers don't file expenses due to friction. CFO investment remains elevated but execution gaps persist between spending intentions and successful deployment outcomes.
• 2025-Q1: Product maturity and ROI evidence advance. Vendor innovation continues: Expensify launches AI Travel platform with SmartScan; Emburse upgrades OCR to 95% accuracy with 39 expense categories. Research validates AI reconciliation benefits: peer-reviewed study shows error reduction to 0.4% and processing time halved. CFO investment accelerates sharply: 51% of finance leaders investing in AI (up from 15% in Aug 2023), with audit time reduction to 90%. However, implementation risks surface: Gartner projects 30% of GenAI projects abandoned by end-2025 due to data quality and unclear ROI; ACCA guidance highlights change management, data quality, and ethical concerns as persistent barriers. Mainstream deployment remains concentrated in enterprise and mid-market; SMB adoption constrained by cost and integration complexity.
• 2025-Q3: Mid-market deployments continue validating ROI. Ramp case studies document construction and retail wins: Construction One achieved 75% reconciliation time reduction (40→10 hours/month), Zola cut month-end close from 20-25 to 12-13 days. Payhawk deployments capture VAT recovery ROI (€2M). Market expansion accelerates: SaaS expense management market valued $7.5B (2025) to $23.6B (2033) at 15.4% CAGR; alternate analyst values it $3.06B→$5.62B by 2032. Adoption standardization evident: 66% of organizations report fully standardized T&E policies globally. However, compliance gaps persist: survey data shows 23% of employees find it acceptable to exaggerate claims and 39% of managers sign off non-compliant reports, indicating that technology adoption does not automatically resolve behavioral compliance challenges. Mainstream adoption concentrated in enterprises and willing mid-market firms with high transaction volumes.
• 2025-Q4: Vendor platform maturity advances with continued mid-market case study validation. Fyle case studies document Centric Infrastructure Group achieving 90% error reduction and sub-2-minute reconciliations via deep NetSuite integration. Continuum consulting report shows global investment advisory firm automating expense allocation with Alteryx and AI-driven parsing. Central Construction reduced monthly reconciliation from 40+ hours using Extend platform. Industry benchmarks solidify: 70-80% reconciliation time savings and 99% accuracy from automated matching reported across solutions. However, persistent adoption barriers remain visible: Rossum data shows 49% of finance departments still operate with zero automation—a sobering reminder that despite rising investment in AI for spend management, majority of finance teams remain manual. Market fundamentals stable with dual analyst perspectives on growth (SaaS segment $7.5B–$23.6B through 2033 at 15.4% CAGR). Mainstream adoption remains concentrated in enterprises and mid-market firms with high transaction volumes; SMBs continue facing cost and integration complexity barriers despite improved product accessibility.
• 2026-Jan: Fortune 500 adoption solidifies with GFOA data showing 78% deploy AI in finance and 92% adoption for expense processing (340% average ROI, $2.4M median savings). Accounting firms reach tipping point: AI adoption jumped from 9% (2024) to 41% (2025) with 77% planning increased investment. Vendor platforms mature: Brex achieves 99% OCR accuracy with 8-week deployments; Concur processes complex receipts under 30 seconds. However, ROI realization remains challenging: PwC survey shows 56% of CEOs see no measurable AI ROI, and Gartner forecasts only 12% of organizations see significant benefits despite $2.52T spending. Dext survey reveals implementation risks—50% of accountants report clients suffering financial losses from AI categorization errors. Mainstream adoption bifurcation persists: while Fortune 500 and mid-market firms deploy at scale, 49% of finance departments still operate with zero automation, indicating that technology maturity has outpaced organizational change capacity.
• 2026-Feb: Persistent ROI gap becomes focus of critical reassessment. NBER survey of 6,000 CEOs/CFOs reveals 69% AI adoption rate but 80%+ report zero measurable productivity impact over three years. Vendor product maturity continues: Trintech reconciliation platform achieves 99%+ auto-match rates on 150M+ daily transactions; accounting firms accelerate adoption with 92% of practitioners using AI tools. Independent practitioner assessments surface capability limitations: AI achieves 87% accuracy on routine categorization but fails on context-dependent edge cases and compliance-sensitive scenarios. Finance leaders cite critical concerns: explainability gaps, hallucination risks, data quality dependencies, and regulatory auditability gaps requiring robust governance. Forrester study commissioned by Navan validates deployment ROI where achieved (80% submission time reduction, 24 min/report savings, 8 hours/week finance team savings), demonstrating that technical maturity remains strong for organizations with change management capability. Mainstream bifurcation persists: adoption gap between well-resourced organizations and majority of finance teams facing implementation barriers. Window signals industry inflection point: while vendor capabilities improve, executive confidence in ROI realization weakens, indicating market maturity plateau ahead without stronger change management and risk governance practices.
• 2026-Mar: SAP Concur launched Joule Expense Automation and Pre-Submit Audit agents at Concur Fusion 2026, with Joule Receipt Analysis Agent reaching GA and new integrations with Microsoft 365, Amex, and Visa signaling broad ecosystem maturity; virtual card platforms claim 85% faster reconciliations by pre-coding GL at transaction time. Mid-market ROI evidence continues to accumulate — Indian deployments demonstrate 8-10 days saved per month and 280% cumulative 3-year ROI — while Deloitte data (3,235 leaders) confirms the structural tension: 88% adopt AI but only 20% generate revenue from it, and Forrester forecasts 25% of planned 2026 AI spend will be deferred to 2027 as CFO scrutiny on ROI intensifies.
• 2026-Apr: SEC elevated reconciliation automation to regulated control area in FY2026 examination priorities; World Bank-IMF report confirms 47% global adoption rate for AI reconciliation with 23% YoY growth, signaling institutional convergence. SAP Concur's Joule Expense Automation and Pre-Submit Audit agents moved to production deployment at Fusion 2026 with Microsoft 365 Copilot integration live, while Ramp case studies document enterprise-scale ROI (ABB Optical Group: audit prep from 2 months to 1-2 days, 85-90% error reduction, $2M+ savings). Mid-market case studies accumulate: Meridian Advisory (40-person firm) achieved 4x ROI in year one, 74% → 96% policy compliance, 15 hrs/month saved; BDO Denmark (1,400 employees) realized 50-82% processing acceleration and 80% admin time reduction. Coupa CFO research reveals the execution-readiness gap: CFOs losing 26 hours/month to manual reconciliation, 85% view AI as central to strategy but 92% fear implementation due to data fragmentation; only 5% have instant access to unified spend data. Adoption barriers persist: industry survey shows only 27% of organizations using AI in spend management, 59% using none—a widening gap between technology maturity and organizational change capacity despite AI achieving 95%+ categorization accuracy in controlled conditions.
• 2026-May: Ecosystem maturity advances with major vendor product launches, financial validation, and adoption signals reaching new highs. Consero Global survey (102 PE/VC-backed CFOs) shows 97% of finance leaders using or testing AI — up from 76% — with 75% seeing positive ROI within 12 months; yet 27% still rely on spreadsheets for reconciliation vs. 13% fully automated close, signalling persistent execution gap. SAP Concur Complete (Amex GBT partnership) confirms enterprise-scale deployment outcomes: Uber saved 5,000 hours, KUKA achieved 7x faster entries, Stellantis reported €1M+ savings across 30+ countries; SAP Joule's four-agent modular architecture (Receipt Analysis, Automation, Pre-Submit Audit, Validation) reached GA. Ramp (50K+ customers, $1B 2025 revenue) ships Smart OCR with auto-vendor-correction and AI procurement agents delivering 16% cost savings and 46 hours/month reduction; Accounting Agent at 90%+ auto-coding with 3.5x more transactions coded and 3x faster close. Microsoft releases native AI Expense Agent for Dynamics 365 Business Central with auto-receipt capture, policy enforcement, and project ledger posting—signaling tier-1 ERP platform architectural commitment to AI-first expense automation. Navan demonstrates sustained market momentum: Q4 FY26 revenue of $178M (+35% YoY), record NPS 47% and CSAT 96%, free cash flow positive ahead of schedule (Morgan Stanley Top Pick upgrade). Accounting firm case study: Greenfield Advisory cut month-end close from 3 days to <1 day with AI reconciliation agent, saving 18 billable hours/month per team member. Critical implementation gap remains: TechTonic Shifts analysis documents endemic failure mode—enterprises automate documented processes rather than actual workflows, causing deployment failures. Structural adoption barrier persists: only 27% of organizations using AI in spend management, 59% none.
• 2026-Jun: Adoption crossed a statistical mainstream threshold: Deloitte's 2026 controller survey documented AI reconciliation adoption at 61% with a median 3.2-day (28%) close cycle reduction — up from 7% in 2023 — signalling the shift from early-adopter to majority deployment. Vendor GA activity accelerated simultaneously: Workday launched the Sana Travel Agent for T&E as a GA product embedded in its platform, Sim.ai released agentic SAP Concur templates (expense classifier, auditor, reconciler, policy enforcer) across 70 APIs, and Payhawk's Summer '26 release added native SAP S/4HANA integration — though Payhawk's own research found only 14% of organizations have successfully consolidated spend controls, reconciliation, and automation, quantifying the structural execution gap. The governance gap now defines the next constraint: KPMG found 75% of companies use AI in finance but only 42% can audit their AI decisions, while Grant Thornton's June assessment confirmed audit risk has migrated from manual-error detection to algorithmic validation (bias, opacity, incomplete training data), reshaping what finance teams must govern rather than what they must deploy.
• 2026-Jul: Vendor GA acceleration and enterprise adoption inflection confirmed mid-to-late July. Competitive category maturity signalled by Ramp's independent validation (34% reduction in manual receipt review via fine-tuned LLM, 79% infrastructure cost advantage) and Navan's enterprise customer wins (5 major deployments: Evotec, Cummins, Enbridge with CAD $2M projected annual savings, ISOPLUS, Viessmann), replacing SAP Concur in enterprise reviews — while Intuit's rollout of AI expense categorisation to all QuickBooks Online SMB plans marks mainstream SMB inflection. Multi-vendor GA progression: SAP Concur Expense Automation and Pre-Submit Audit agents (with Microsoft 365 Copilot integration), SAP Cash Management Agent (70% reconciliation time savings per internal research, Q1 2026 GA), and BlackLine's Verity Prepare multi-agent reconciliation system (92% manual time reduction at Delaware North, emphasizing audit-ready governance). Market research confirms adoption inflection: expense management market $8.18B (2025) to $23.03B (2035) at 10.9% CAGR, with manual reconciliation dropped to <18% (mid-market 2026, from 42% in 2020). CFO ROI pressure is now the primary constraint: KPMG survey (2,145 C-suite) found only 7% have established AI ROI and 42% lack visibility into AI spend. A critical execution gap persists: while receipt capture automation is solved (99%+ accuracy), independent practitioner analysis identifies a structural bottleneck—coding rules and approver logic retention across organizational turnover. Expense categorization reliably handles recurring transactions but fails on VAT logic, edge cases, and expense-vs-fixed-asset distinctions requiring ongoing governance. Microsoft's Intelligent Risk Engine risk-routes 1M annual expense reports and Masraff launched a 7-agent autonomous expense platform; spend-management-agent market sizing projects $8.6B (2025) to $56.4B (2034). Two independent CFO surveys (KPMG, ERA Group) reinforced the ROI-visibility gap, finding roughly 78% cannot connect AI spending to business outcomes and about half have scaled back deployments pending measurement capability. Late-July evidence quantified the manual-process baseline and reconciliation-specific returns more precisely: GBTA Foundation data put manual expense-report cost at $58/report with a 19% error rate and $500K in annual rework, while an aggregated industry synthesis (Gartner, McKinsey, Deloitte, BlackLine, HighRadius) confirmed reconciliation automation delivers 40-60% close-cycle reduction, 70-90% fewer GL errors, and 95%+ anomaly-detection accuracy.
• 2026-Aug: Named deployments validate mid-market ROI (Payhawk's AllPoints Fibre Networks case eliminated 30 NetSuite licenses and 3 hrs/week manual uploads) alongside continued vendor scale (Moss reached €1B+ valuation, 5,000+ customers, €70M+ ARR; Payhawk hit 68.6% autonomous retrieval and cut vendor invoice collection 46%). Emburse's AI Index shows 48% of enterprises with recurring AI expense spend, but accountability friction persists: Thomson Reuters documented fabricated citations in Big 4 reports and unrecovered ROI, while practitioner commentary confirms liability for AI miscategorization still falls on human bookkeepers, not software.