Why the Best AI Implementations Start with Discovery

Finance leaders are under pressure to “do something with AI.” The instinct is to grab a promising use case, spin up a prototype, and show momentum. But in many cases, the result is a tool that never gets used, or solves a problem that wasn’t really worth solving.

According to the latest McKinsey report, while adoption is accelerating, very few companies report a material financial impact from AI. In my work with finance teams, I’ve seen why: discovery is often skipped, rushed, or superficial.

In this issue, we’ll explore what discovery really looks like in practice — and how finance leaders can use it to filter out noise, build smarter, and avoid the trap of chasing the wrong ideas.

📢 Upcoming Events for Finance Leaders – Save the Dates!

What separates finance leaders who talk about AI from those who actually use it?

Community, context, and consistency.

Last Friday, I hosted our latest Fractional CFO Corner inside the AI Finance Club—over 100 participants joined to discuss real-world AI use cases, tools, and lessons learned.

If you’ve been meaning to start but don’t know where to jump in, this is your next move:

📅 Free webinar: How to Become an AI CFO
🧠 Hosted by Nicolas Boucher
🗓️ June 27 | 11 AM ET / 5 PM CET

Seats are limited—grab yours now!

Discovery Over Delivery: The Missing Phase in Most AI Projects

According to the latest McKinsey report, nearly 80% of companies now use generative AI in at least one function. Yet roughly the same percentage report no material earnings impact from their efforts.

We’re in what McKinsey calls the “gen AI paradox”: widespread adoption, minimal value. And from what I've seen in real projects, one of the reasons is that companies often skip the discovery phase.

They get excited, they build something, and only then realize: this isn’t solving a meaningful problem, or it’s not technically feasible, or the team isn’t ready to use it.

This is especially critical in finance, where expectations for AI-driven efficiency are high, but processes are complex, data is sensitive, and regulatory tolerance for mistakes is low.

Why “Obvious” Use Cases Often Fail

Let me share a story from a recent engagement.

I was recently giving a presentation on “What AI is” to a professional services firm when a powerful use case surfaced. We were discussing AI for compliance checks, and coincidentally, the company had just been hit with a $500,000 regulatory fine for missing a key parameter in a client report. A documentation oversight had slipped through the cracks. The incident was fresh, painful, and very real. Naturally, the team’s response was immediate: “Let’s fix this with AI.”

Everyone agreed — this seemed like a great problem to solve.

We scoped it out, explored how AI could help, and outlined a prototype.

But fast forward four weeks, that fine turned out to be a one-off event in over ten years. The process we hoped to automate turned out to be far more complex than it first appeared. And the potential AI solution wasn't as simple or low-lift as we thought. Suddenly, the use case didn’t seem so strategic.

This story isn’t unique. It's exactly how so many AI projects start — with a spike of enthusiasm and urgency that fades once you realize the business case is thin or the solution is hard to scale.

The Cost of Skipping Discovery

When teams skip discovery, they often mistake urgency for scale and novelty for impact.

Skipping structured discovery leads to:

• Solving rare edge cases instead of recurring pain points

• Burning resources on one-off prototypes

• Building tools that no one uses after demo day

This is one reason so many vertical (function-specific) use cases stay stuck in pilot mode.

What Strategic Discovery Actually Looks Like

Discovery isn’t about slowing things down. It’s about making sure you build the right things for the right reasons.

And that starts with asking the right questions — not about AI, but about the business.

Step 1: Identify Pain Points

The real starting point for discovery is not use cases. It’s pain.
What’s slowing the business down?
Where is the team wasting time, making errors, or constantly firefighting?
What are the recurring frustrations in compliance, reporting, forecasting, or data reconciliation?

Until those pain points are clearly articulated, any AI conversation is just guesswork. Once surfaced, these issues can be mapped to specific AI capabilities, such as automation, summarization, anomaly detection, prediction, and so on.

Step 2: Prioritize and Frame

Once multiple opportunities are on the table, the next step is prioritization.
Which problems are the most frequent, the most expensive, or the most error-prone?
What would the impact be if we solved them?
This is where high-level filtering begins — to narrow in on problems that are not just painful, but also addressable.

Step 3: Validate Use Cases

Only after mapping business needs to AI capabilities do we validate individual use cases. That includes:

• Business validation: Is this a recurring and costly problem?

• Workflow mapping: Where would AI fit, and what would trigger its use?

• Technical feasibility: Do we have the right data and systems to support this?

• Change readiness: Will the team adopt it? Are there compliance concerns?

If this framework feels familiar, it should — it builds directly on the four pillars of AI readiness I introduced in a previous issue: Process, People, Data, and Governance. Discovery is how we put those pillars into practice.
(If you missed that edition, here’s the link: Is Your Organization AI-Ready?)

Step 4: Sequence the Roadmap

Once you’ve identified and validated multiple strong opportunities, the final step is sequencing.
Not everything needs to be built at once.
Discovery helps you define what to start with, what to plan for later, and what needs more foundational work before automation is viable.

In short, strategic discovery isn’t about validating a single flashy idea. It’s about taking a structured look at the business, identifying where AI can help most, and ensuring every step is aligned with readiness and long-term impact.

A CFO’s Role in Discovery

You don’t need to write code or architect models. But you do need to lead discovery with financial clarity and strategic filters.

Here’s how to do it well:

Be the Translator. Finance is full of processes that look repetitive but aren’t structured. Your job is to help identify which tasks are genuinely automation-ready, not just frustrating.

Define a Use Case Threshold. Set minimum criteria: frequency, effort, error risk, business value, and feasibility.

Create a Discovery Space. Establish a recurring area for your team to propose, evaluate, and refine AI ideas. Short workshops, structured templates, and a shared dashboard can go a long way toward turning chaos into clarity.

The most successful AI initiatives in finance don’t start with building — they start with understanding.

They ask:

• Is this a meaningful problem?

• Can we solve it reliably with AI?

• Will our team adopt and trust the solution?

If you get discovery right, you don’t just reduce waste — you raise your odds of actually getting the impact that McKinsey and others say is missing.

3 AI Use Cases That Looked Promising, But Didn’t Survive Discovery

In nearly every finance team I work with, AI brainstorming surfaces a handful of exciting ideas. But once we apply structured discovery, many of them get re-scoped or dropped entirely.

Here are three real-world examples of use cases that looked great on the surface but didn’t hold up under closer evaluation:

1. Report Automation Where Human Judgment Still Wins

Teams were eager to automate monthly board and investor reports with AI-generated summaries. But they ran into a wall:

• The reports needed context and nuance

• Narrative emphasis shifted each month based on strategic priorities

• Final versions required political judgment, not just data

Lesson: If the end result needs careful framing or tone, AI can assist, but it shouldn’t lead.

2. Forecasting Without a Foundation

A leadership team wanted AI to improve revenue forecasting. But once we dug into the data, we found:

• Inconsistent historical records

• Differing revenue recognition methods across regions

• Manual adjustments that obscured patterns

Lesson: AI can’t create clarity out of chaos. If the past is messy, the future will be too, no matter how smart the model is.

3. Workflow Automation No One Actually Wanted

One team built a bot to flag unusual expense claims — smart in theory. But in practice:

• Staff were hesitant to engage with it

• Managers didn’t want to create friction with employees

• The resolution process was still manual and slow

Lesson: The biggest bottleneck wasn’t technical, it was behavioral. If people aren’t willing to adopt the tool, even the best-designed AI will go unused.

Closing Thoughts

The most impactful finance AI projects I’ve seen didn’t start with a brilliant idea — they started with a disciplined process.
That process begins by surfacing real business pain points, mapping them to AI capabilities, and filtering out what isn’t worth building.

As a finance leader, your job isn’t to say yes to every use case — it’s to create the conditions where AI can actually deliver value.

Discovery isn’t a delay. It’s your best accelerator.

If you’re planning your next AI initiative, start with one question:
“What problem are we really solving — and is it worth solving with AI?”

Need help structuring your discovery process? Book a consultation!

We Want Your Feedback!

This newsletter is for you, and we want to make it as valuable as possible. Please reply to this email with your questions, comments, or topics you'd like to see covered in future issues. Your input shapes our content!

Want to dive deeper into balanced AI adoption for your finance team? Or do you want to hire an AI-powered CFO? Book a consultation! 

Did you find this newsletter helpful? Forward it to a colleague who might benefit!

Until next Tuesday, keep balancing!

Anna Tiomina 
AI-Powered CFO