The Same Mistake in Finance
The Klarna story might seem distant from your finance function. It's not. The same pattern is showing up in The Hackett Group's research on finance AI adoption.
Organizations are planning AI for deterministic processes — cost accounting (22% planning, only 4% piloting), general accounting and close (26% planning, <4% piloting), and transaction processing (18% planning, 4% piloting). These are rules-based processes where traditional automation delivers better results.
Meanwhile, the processes where AI actually excels — FP&A, forecasting, anomaly detection — show the highest success rates.
The CFO/Controller perception gap explains why. From the C-suite, AI looks transformative: sophisticated board commentary, automated dashboards, scenario modeling. But as one Financial Controller told researchers:
"Strategic commentary may look automated, but many operational finance staff still export spreadsheets, fix data manually, and stitch together numbers before AI tools can even begin their analysis."
CFOs are seeing the presentation layer. Controllers are seeing the engine room. And the engine room is still manual.
What Winners Do Differently
While most organizations struggle, a small percentage are achieving breakthrough results. The difference isn't better technology — it's understanding where AI actually works.
Allianz transformed its FP&A function by deploying AI for data aggregation, natural language queries, and anomaly detection. Result: 60% reduction in manual workload, with analysts redirecting time toward strategic analysis.
Aviva deployed 80+ AI models across its claims operation — but specifically for pattern recognition tasks like liability assessment and routing. Result: 23 days cut from assessment time, 65% fewer customer complaints, and Net Promoter Scores that increased seven-fold.
Lufthansa used AI to consolidate procurement data from 20+ ERP systems, applying pattern recognition to identify spend anomalies and price volatility. The same approach now powers their Scope 3 carbon reporting.
The pattern is consistent: AI succeeds when deployed to processes involving pattern recognition, unstructured data synthesis, and anomaly detection. It fails — or adds unnecessary complexity — when applied to deterministic, rules-based processes that need 100% accuracy and audit trails.
The Insight
"Use GenAI where it's about language, not math." - Boston Consulting Group
Many finance workflows involve deterministic questions with only one correct answer: verifying sales, reconciling accounts, ensuring regulatory compliance. GenAI isn't built for those. Traditional automation — Python scripting, RPA, SQL-based queries — delivers superior results for precision work.
AI transforms processes involving pattern recognition across large datasets, synthesis of unstructured information, and generating insights from complex patterns. Budget variance analysis, forecasting, document processing, spend analytics — these are AI's sweet spot.
The 5% of organizations achieving tangible AI ROI aren't smarter or better funded. They're deploying AI where it actually works — and using traditional tools where those work better.
This Week's Action
Close the perception gap. Ask: "For each AI tool we've deployed, how much manual work happens before the AI can do its job?"
If the answer involves exporting spreadsheets, fixing data, or stitching together numbers — you've found where the real transformation needs to happen. And it might not be AI.
