Your processes work. Well enough that you haven't changed them. But “well enough” still means exporting from one system, patching it in Excel, uploading it back again. It means exception queues that swallow whole afternoons.
Meanwhile, executives keep asking what finance is doing with AI. And when you sit down to propose something, the questions multiply. Where do you start? How do you scope it? What's the business case?
Here's the truth: Your processes aren't broken, they're the status quo.
The problem is, staying in that zone keeps you from acting, which leaves you feeling stuck. You need to start with a mindset shift because you can't treat AI like a traditional enterprise project.
How traditional and AI projects are different
| Traditional projects | AI projects |
|---|---|
| Months of planning and requirements gathering | Iterative, with rapid testing and refinement in prod |
| Change management across departments | Surgical interventions with minimal disruption |
| System overhauls that touch everything | Augmentative automation within existing systems |
| Big-bang go-lives after long development cycles | Pilot-and-expand, with results in weeks |
Example
A two-week proof of concept at a manufacturer
A manufacturing company processing over a million invoices annually achieved only 30% straight-through processing, despite investing in OCR and validation pipelines. About 10% of invoices were flagged for missing or improperly formatted data.
Rather than spending weeks on requirements and scoping, they ran a two-week experiment where an AI agent extracted and formatted the problematic data fields, then fed the results back into their existing pipeline.
The agent took two hours to build and two days to integrate, and every output was validated as accurate — turning a multi-month project into a two-week successful proof of concept.
This shift changes how you identify, scope, and win approval for AI automation projects. Instead, you can target a specific pain point and see results in weeks. You don't need to replace entire systems either, because automation slots into the processes you already run.
The 5-step framework: From problem to approval
The traditional enterprise playbook doesn't work for AI. You need a framework built for how AI actually delivers value: start small, iterate quickly, expand on proof.
Here's the process: Identify → Scope → Design → Justify → Approve.
Identify. Identify opportunities
Scope. Scope the solution
Design. Design business requirements
Justify. Justify with ROI
Approve. Win executive approval
This framework helps finance managers handle three realities:
- Executive expectations. Leaders want AI but don't know how to evaluate it.
- Risk management. Finance needs confidence in outcomes before committing resources.
- Implementation reality. IT needs clear technical requirements tied to business goals.
This playbook gives you the framework to identify, scope, and win approval for AI projects. To make it easier to act on, we've also created a companion toolkit with ready-to-use templates for each step. As you move through the framework, you'll see where a template fits, and at the end, you can download the full package to put everything into practice immediately.
Now it's time to get practical. Step 1 starts with spotting the bottlenecks where AI can make an immediate impact.
Step 1
Identify opportunities
Where are humans “swivel-chairing”?
The first step is looking at the processes you've accepted as “just the way things are” and asking: Where are we paying the highest price for manual work?
The best AI opportunities aren't the biggest or flashiest problems. They're the hidden bottlenecks. These are spots where people get stuck in swivel-chair mode: bouncing between systems, manually copying data into Excel workbooks, cleaning up messy CSV exports.
Look for these signs
- Data gymnastics. Exporting, cleaning, and re-uploading between systems.
- Pattern recognition at scale. Applying the same rules across hundreds of items.
- Exception handling. Processes that run smoothly 80% of the time but consume hours in the other 20%.
- Document-heavy workflows. Reading, interpreting, and acting on unstructured files.
| Current state | AI opportunity |
|---|---|
| Valid invoices flow through your ERP automatically | Valid invoices continue processing as usual |
| 30% hit exceptions and require manual research | Exceptions trigger AI research and analysis |
| Specialists spend 60–80% of their time resolving them | AI takes on the detective work and applies rules consistently |
| Each exception takes 15–45 minutes | Humans review AI recommendations instead of starting from scratch |
| Decisions vary by person, slowing the close | Only true exceptions escalate for manual handling |
| Payments are delayed and early-pay discounts slip away | Increased on-time payment and discount capture rates |
Why it matters
- Exception resolution is faster
- Outcomes are more consistent across the team
- Specialists focus on actual exceptions
- Discount capture improves, and close cycles speed up
This is a surgical intervention. You are not replacing your ERP or rewriting workflows. You are adding intelligence exactly where it pays off.
Opportunity assessment framework
Use this scoring framework to evaluate which opportunities are worth tackling first.
Scoring
- 40+
- High priority
- 30–39
- Medium priority
- Below 30
- Consider other opportunities first
Scoring these opportunities gives you a starting point and a way to defend why you're tackling them first.
Toolkit
If you want to score your own processes, the toolkit includes an assessment worksheet to guide you.
Once you know what to fix, the next step is deciding how much to take on.
Step 2
Scope the solution
Start surgical, expand iteratively.
The biggest mistake in AI projects is trying to take on too much scope too soon. It feels logical, right? If a process is inefficient, why not fix the whole thing? But this approach usually kills momentum.
The smarter path is surgical. Start with one part of the process, prove the approach, then expand from there. Each phase builds confidence while teaching you about the process and the systems involved, and ultimately lets you notch the wins needed to continue investing.
Phased example
Invoice processing
Phase 1Invoice validation
- Current state
- AP processors check every invoice flagged for having incomplete data or bad formatting, usually reaching out to suppliers to ask them to fix the issue and re-submit the invoice.
- AI opportunity
- Agents are able to re-process those flagged invoices to validate that there is, in fact, an issue — and can either fix the issue and notify the supplier, or automatically ask the supplier to fix the issue and re-submit.
- Why start here
- You have a discrete list of requirements that suppliers must follow, making it a great use case for AI with a fast implementation.
Phase 2Basic exception handling
- Current state
- Specialists take 15 minutes to manually handle each case with partial or incomplete references or minor discrepancies.
- AI opportunity
- Agents intelligently match against master data (matching even with partial or truncated match keys), and are able to reason about common exceptions, such as unexpected currency or UOM.
- Why expand here
- Builds on the Phase 1 foundation and further reduces the number of invoices requiring manual handling.
Phase 3Advanced exception handling
- Current state
- The most labor-intensive invoice exceptions are often those that require input from internal stakeholders — whether that requires a site manager to double check a delivery and update a GR or a marketing manager to confirm their outside consultant delivered against the SOW.
- AI opportunity
- Agents can reason about problems and communicate with stakeholders effectively, gathering the information necessary to approve or reject a given invoice.
- Why it matters
- This expands coverage to the cases that really drive up cycle time by automating basic (but time-consuming) internal communications.
Why surgical works
- Each phase delivers results in weeks, not months
- Small failures become learnings instead of disasters
- You can pause or accelerate based on ROI and stakeholder confidence
- Budget is flexible, but prove success before asking for more
Scoping small is what gets you to measurable results. By tackling the most contained slice first, you create momentum and set yourself up for confident expansion.
Toolkit
You'll find a one-page scope template in the toolkit to help you define clear boundaries and success criteria.
Once the scope is clear, the challenge is to translate business rules into a language that both AI and IT can understand.
Step 3
Design business requirements
Unclear business requirements are a recipe for disaster, whether you're proposing an AI solution or not. And when it comes to process automation, unclear requirements usually stem from an incomplete understanding of the business process itself.
In order to design a solution, you need to dig deep into the current state through process discovery. For invoice processing, the business requirements would include:
- Invoice requirements. What do you tell suppliers they must include on their invoices? What formatting rules must they follow?
- Policies. Are there special arrangements for certain categories of suppliers? For specific suppliers? Are there different rules for different countries?
- Exceptions. What are the most common exceptions? Which ones take the longest to resolve? How are they resolved today?
Your job is to translate the core business logic (that won't change once the process is automated) into an enumerated list of business requirements.
You don't necessarily have to cover every possible scenario. Rather, you should focus on the highest impact opportunities, which are typically the cases your AP processors spend the most time on today.
Example
Invoice processing business requirements
- Current state
- AP specialists rely on judgment and experience. When a price mismatch appears, they check the PO, look for price change notifications, apply a 5% tolerance, and escalate if the situation feels uncertain.
- AI-enhanced requirements
- The same logic, but written explicitly.
AI solution should:
- Access purchase order data
- Search for vendor price change notifications
- Calculate variance percentages
- Apply the 5% tolerance rule
- Escalate to a human if confidence is below the threshold
| Component | Purpose | Example |
|---|---|---|
| Clear objective | Define exactly what the AI solution should do | “Review invoices for compliance with payment terms” |
| Relevant context | Provide policies, thresholds, and rules | “Net-30 terms with 2% early-pay discount” |
| Output format | Specify how the result must be delivered | “Compliance status” |
| Constraints | Set limits to prevent overreach | “Do not approve payments over $10,000 without escalation” |
| Examples | Show the AI what “good” looks like | “Example of a compliant invoice: [details]” |
Why this matters
- Makes expectations explicit instead of relying on tribal knowledge
- Creates system-level requirements IT can implement
- Ensures consistency across teams and across time
- Builds confidence with executives who demand auditability
Designing business requirements for AI isn't about writing every rule. It's about teaching the system how to think, where to stop, and how to stay inside your business guardrails.
Toolkit
The toolkit includes a requirements doc you can adapt to capture objectives, context, and system details for your project.
Once the rules are clear, the next question is whether the project pays off.
Step 4
Justify with ROI
Even with impeccable design, any AI project will stall without a business case. Executives need numbers they can trust, not just a vision of efficiency.
The key is to frame ROI in finance's own language: time, accuracy, speed, and employee capacity.
Finance-specific metrics
- Time savings (hours per week recovered)
- Error reduction (percentage improvement in accuracy)
- Processing speed (cycle time improvements)
- Employee satisfaction (less mundane work, more strategic focus)
| Processor hours spent per week | All-in hourly rate | Annual cost | |
|---|---|---|---|
| Current | 160 | $25 | $208,000 |
| Future | 32 | $25 | $41,600 |
| Annual savings | $166,400 | ||
| Implementation cost | $30,000 | ||
| Payback period | 3 months | ||
Why this matters
- ROI builds executive confidence and secures funding
- Clear numbers protect you from pushback and skepticism
- Framing ROI around hours and dollars ties the project to finance's core mission
ROI is your proof point that small, surgical projects deserve executive backing and future expansion.
Toolkit
An ROI calculator is part of the toolkit, built for finance teams to turn hours and savings into clear payback numbers.
With ROI in hand, it's time to make the pitch.
Step 5
Win executive approval
With ROI numbers in hand, you're ready for the final step: turning analysis into a compelling pitch.
Executives don't want jargon or technical detail. They want clarity, confidence, and a low-risk path to results.
| Element | What to cover |
|---|---|
| Problem | Quantify the current pain with data (e.g. hours, costs, errors) |
| Solution | Show the surgical AI intervention you're proposing |
| Proof | Highlight the pilot approach and realized results |
| Investment | State the cost and timeline clearly |
| Return | Share the savings, payback period, and long-term upside |
Why this works
- Positions AI as low-risk, not a big-bang gamble
- Anchors the project in measurable outcomes, not hype
- Creates an expandable path: prove, then scale
Approval comes when executives see a project that is specific, measurable, and safe to test.
Your job is to show them this isn't a transformation for transformation's sake. It's a pilot with upside.
You now have the whole framework. The final section will help you put it into practice with templates and immediate next steps.
Getting started
Reaching this point means you now have a complete playbook for turning frustration into progress. You don't need to overhaul your systems or wait for a massive transformation project. With this 5-step framework, you can start small, prove value quickly, and build momentum.
Think of this as your first step into a new way of working. Instead of living with manual bottlenecks and detective work, you now have a method to identify the opportunities, design requirements that IT can actually implement, scope them with confidence, justify the ROI, and win executive approval.
Example starting points
- Invoice processing and exception handling
- Sales order entry and processing
- Cash application and remittance matching
- Customer billing and collections
- Customer credit management
- Vendor requisition, sourcing, and onboarding
- GL monitoring and continuous audit
- Financial reconciliation
Immediate next steps
Download the complete template kit
Use the Opportunity Assessment to identify your first target
Apply the 5-step framework to a single process
Schedule a follow-up to track progress and build momentum
This is the moment to move from recognizing inefficiencies to building solutions. You don't have to change everything at once. You just need to start with one well-scoped project.
That's where the toolkit comes in. Inside, you'll find the complete set of templates for each step:
Opportunity Assessment Worksheet
Project Scope Template
Business Requirements Doc
ROI Calculator