‍

‍The Real Problem: Decentralization Broke Accounting

Twenty years ago, ERPs were true enterprise systems. Orders, billing, fulfillment, cash—everything lived in one unified data model. The accounting was essentially done automatically.

‍

Then customer innovation won out over internal controls.

‍

Companies adopted Stripe for payments, Zuora for billing, Salesforce for contracts. The ERP became just the final destination—not the hub. Accounting teams found themselves stuck between a ledger that needed compiled information and third-party data sources speaking different languages.

‍

This decentralization created three critical problems:

• Volume explosion: Transaction counts jumped from thousands to millions
• Data fragmentation: Financial context spread across 10+ disconnected systems
• Manual compilation: Accountants became data bridges, manually reconciling disparate sources

‍

The result? A 25% year-over-year increase in audit deficiencies, with revenue accounting as the #1 failure area.

‍

Understanding Deterministic vs. Probabilistic Outcomes

Before choosing between automation and AI, you need to understand a fundamental distinction:

‍

‍

For accounting, this distinction is critical. When an auditor asks "how did you calculate revenue recognition for this customer?", "the AI said so" isn't an acceptable answer. You need deterministic automation with complete audit trails.

‍The Data Foundation Problem: Garbage In, Garbage Out (Amplified)

Here's the uncomfortable truth about AI in accounting: it amplifies your existing data problems rather than solving them.

The Broken Pyramid

Investment analysts describe finance systems as a pyramid:

‍

‍

The promise: clean data flows from bottom to top, enabling strategic insights.

‍

The reality: That pyramid is broken for most companies. FP&A cobbles together whatever they need for strategic analysis. Accounting cobbles together what they need for compliance. The middle layer—the data foundation—is like a Jenga piece pulled out. The whole structure is wobbly.

LLMs need clean, atomic, highly correlated data to train on. But if your monthly close consists of accountants gathering disparate data and translating it into high-level journal entries every 30-40 days, where exactly will you train an LLM? What context does it have?

‍

What AI Actually Can't Fix in Accounting

Let's be specific about where AI falls short for core accounting work:

1. It Can't Provide Auditability

When an auditor asks "how did you calculate this?", you need to trace every step back to source transactions. AI's probabilistic nature makes this impossible. Even if it gets the right answer, you can't prove why it's right.

2. It Can't Guarantee Consistency

Run the same AI query twice, you might get different results. This "stochastic" behavior—where randomness is built into responses—is fundamentally incompatible with the immutability requirements of accounting periods.

3. It Can't Replace Domain Expertise

The governance of accounting data requires human expertise: understanding organizational context, applying accounting policies, knowing what metrics mean for your business model. AI can assist, but it can't replace that governance.

4. It Creates a Dangerous Efficiency Illusion

If humans need to review and validate every AI output anyway, have you really saved time? Or have you just shifted from one tedious task (data entry) to another (validating AI outputs)?

‍

Where Automation Wins: The Deterministic Layer

For most accounting teams, deterministic automation solves the immediate, critical problems:

‍

Close Process Acceleration

With properly automated data pipelines, companies move from 30-day closes to daily closes. Not because AI is magically booking entries, but because deterministic relationships between transactions and outcomes are captured continuously rather than compiled monthly.

Where AI Actually Adds Value: The Strategic Layer

Once you've built the deterministic foundation, AI becomes incredibly valuable for the strategic layer:

Contract Analysis and Data Extraction

AI scanning contracts to extract performance obligations, terms, and clauses is legitimate. The key: this extraction feeds into deterministic automation that calculates the actual accounting treatment.

‍

‍Anomaly Detection

Once you have clean, atomic, time-series financial data, AI excels at finding patterns and outliers:

• Which customer segments have highest default risk?
• Where are we seeing unexpected revenue leakage?
• What product combinations generate best margins?

These probabilistic answers don't need audit-level precision—they're directional insights that help you ask better questions.

FP&A and Forecasting

Scenario planning, sensitivity analysis, predictive forecasting—natural fits for AI because you're explicitly dealing with probabilities and ranges.

‍

Critical Insight: All of these valuable AI applications require the deterministic automation layer underneath. Without it, the AI has nothing reliable to train on.

‍

The Right Implementation Sequence

Based on real-world implementations with high-volume consumer transaction companies, here's what actually works:

‍

Phase 1: Fix the Data Foundation (Deterministic Automation)

1. Audit your current state: Map all systems where financial data originates
2. Define your data governance: Let accounting own the definitions of key metrics
3. Build deterministic pipelines: Connect source systems with proper relationship modeling
4. Implement continuous reconciliation: Don't wait 30 days to know if things balance
5. Ensure immutability: Build protections so prior periods can't change without documented adjustments

‍

Phase 2: Enable Strategic Analysis

With clean data flowing, you can now:

1. Create management reports that refresh daily instead of monthly
2. Give business partners access to financial data they can trust
3. Free up accountants from compilation to analysis

‍

Phase 3: Layer in AI for Insights

Only after phases 1 and 2 are you ready to productively use AI:

‍

1. Train models on your clean historical data
2. Use AI for pattern recognition and outlier detection
3. Apply AI to forecasting and scenario planning
4. Implement AI-assisted contract review feeding deterministic systems

‍

‍

‍

Why Sequence Matters: Companies that jump straight to AI without fixing the foundation inevitably hit a wall. The 95% pilot failure rate isn't because AI doesn't work—it's because companies try to use probabilistic solutions for deterministic problems.

‍

‍

Evaluating Vendors: What to Look For

The explosion of "AI-powered accounting" solutions makes vendor selection treacherous. Here's how to cut through the noise:

‍

The Human Element: Elevation, Not Replacement

One persistent fear: will automation and AI eliminate accounting jobs? The real-world evidence says no—but roles absolutely evolve.

‍

‍

The accounting profession's current crisis isn't about too many accountants. It's about too many accountants stuck in soul-crushing compilation work when they could be doing strategic analysis.

The Shift in Role

Controllers who spent 70% of their time on monthly close compilation now spend that time on:

• Margin analysis by product, customer, and channel
• Pricing strategy informed by actual contribution data
• Business partnership translating ops into financial outcomes
• Proactive identification of revenue leakage and opportunities

That's not just better for the business—it's why people chose accounting in the first place.

‍

Your Action Plan

If you're leading accounting or finance at a high-growth company, here's your roadmap:

Immediate Steps (This Week)

1. Audit your data chaos
2. Map every system where financial transactions originate. Be brutally honest about what's manual, what breaks, what changes unexpectedly.
3. Calculate cost of status quo
4. What's your actual monthly close time? Hours in reconciliations? Opportunity cost of your best people doing compilation?
5. Define your north star metrics
6. Let accounting own the governance of revenue, contribution margin, and other key measures. Document what these mean in your specific context.

‍

Short-Term (0-6 Months)

1. Prioritize revenue automation
2. If you haven't solved revenue-to-cash with high-volume transactions across multiple systems, that's priority #1. Biggest compliance risk and highest business value.
3. Build or buy the deterministic layer
4. Focus on automation providing complete audit trails, maintaining period immutability, and handling your specific billing platforms.
5. Resist AI pilot pressure
6. Unless you have clean, atomic financial data flowing continuously, AI pilots will fail. Get the foundation right first.

‍

Long-Term (6-18 Months)

1. Measure what matters
2. Track close time and reconciliation hours, but also business impact. Are your financial insights changing operational decisions?
3. Layer in strategic AI
4. Once your deterministic foundation is solid, experiment with AI for pattern recognition, forecasting, and analysis in low-risk applications.
5. Invest in your team
6. The automation dividend isn't headcount reduction—it's elevation. Help your team develop strategic skills to use their freed-up time.

‍

‍

The Bottom Line

Automation and AI are both transforming accounting—but in fundamentally different ways, solving different problems, and requiring different sequences.

‍

Automation handles the deterministic problems accounting is built on: precise calculations, audit trails, period immutability, reconciliations.

‍

AI handles the probabilistic problems that extract insight from that foundation: pattern recognition, forecasting, anomaly detection, scenario planning.

‍

Trying to use AI without automation is like building the top floor without the bottom floors. It looks impressive in demos, but it doesn't work in production.

‍

*This guide draws from conversations with Jason Berwanger (CEO, HubiFi), Isaac Heller (CEO, Truli), and Em Daigle (Founder, Automates community), along with insights from real-world implementations across high-growth SaaS and consumer subscription businesses.

‍

‍