Artificial intelligence is no longer a futuristic concept. It is a critical part of how finance functions operate today. Yet most finance organizations we talk to struggle to use AI effectively because they don’t have a strategy — they have a tool list.
The difference matters. A tool list asks “what should we buy?” A strategy asks “how should AI sit inside our operation?” The second question is the one that actually produces leverage.
Here’s the model we use when we help finance organizations design their AI approach.
Why finance needs an AI strategy, not an AI vendor
AI can improve efficiency, reduce costs, and unlock new capabilities. Without a clear plan, AI projects either fail outright or deliver isolated wins that don’t compound.
A real strategy does four things:
- Aligns AI with actual business goals instead of running technology experiments
- Identifies the right type of AI for each type of work
- Manages risk and compliance — especially around financial data
- Plans for integration and scaling so AI fits how the business actually operates
Without that thinking, you end up with five AI tools that don’t talk to each other and a team that’s stopped using most of them.
Three layers of AI agents
The framework that works best in finance has three layers. Each one plays a different role, and they compound when designed together.
Layer 1: Doer Agents
Doer Agents handle routine, repetitive, rule-based tasks. Speed and accuracy on volume work. In finance, that’s things like invoice data extraction, transaction coding, automated reconciliations, report generation, and standardized document processing.
Tools like Hyperbots for AP automation and Blackline for reconciliations live here. The measurable outcomes are throughput, error rates, and cost per transaction.
Doer Agents free up your team’s capacity — they don’t change what your team does with that capacity. That’s what the next two layers are for.
Layer 2: Human-in-the-Loop (HITL)
The middle layer is where AI and human judgment combine. The AI does a first pass; a human reviews, approves, or overrides before action is taken.
In finance, HITL shows up in flows like anomaly detection in transactions (AI flags, controller reviews), draft narrative generation (AI drafts the commentary, CFO edits), and exception triage (AI surfaces, human decides).
HITL is the layer where AI adoption actually sticks in finance, because it preserves accountability. The human remains responsible for the decision. The AI remains useful because it handles the first 80% of the work.
Layer 3: Intelligence Tools
The top layer is where AI stops executing tasks and starts supporting decisions. This is analytical, synthesizing work: forecasting, scenario modeling, variance analysis, board narrative generation, audit preparation summaries.
Tools like Claude, ChatGPT, and purpose-built FP&A tools like Snowfire.ai live here. The outcome isn’t a processed transaction — it’s a sharper decision.
Intelligence tools are where the intelligence layer earns its name. They don’t do the work; they make the humans doing the work meaningfully better.
How the three layers work together
The mistake most teams make is treating AI as a one-layer decision. “Let’s buy Hyperbots” (Doer). Or “Let’s start using ChatGPT for reports” (Intelligence). Both can be right individually. Neither builds a strategy.
A real AI strategy in finance uses all three layers and designs them to pass work between them cleanly. Doer Agents process transactions. HITL flags the ones that need judgment. Intelligence tools help humans make that judgment faster and back the final decision with analysis.
When that flow works, the finance function gets faster and sharper at the same time. When any one layer is missing, the other two break down.
Where to start
If you haven’t started, pick one finance process and do a three-layer audit:
- What parts of this process are repetitive and rule-based? That’s Doer Agent work.
- What parts need judgment but currently happen too slowly? That’s a HITL opportunity.
- What parts are pure analysis and synthesis? That’s the Intelligence layer.
Then pick one tool for each. You’ll almost certainly end up with more than one — and that’s how a real AI strategy is supposed to look.
Liv Data helps finance organizations design and implement three-layer AI strategies that actually stick. Book a strategy session to map where AI belongs in your finance function.