Episode 3: AI Projects are Strategic Expenditure

The transcript below has been lightly edited for readability while preserving the conversational format of the episode. It builds on the original source article with a deeper discussion of AI investment governance and Strategic Expenditure.

00:00:00 Host 1
Usually when we talk about enterprise finance, there’s this baseline expectation of total precision.

00:00:06 Host 2
Oh, absolutely, total precision.

00:00:08 Host 1
Right. It is a lot like architecture. I mean, if you need a new factory, the blueprints show exactly where the steel goes, how much the concrete costs.

00:00:15 Host 2
And the CFO points to the capital budget and says, there it is.

00:00:18 Host 1
Exactly. Funded or not funded, it’s clean, it’s, you know, comforting.

00:00:23 Host 2
Highly visible investments categorized neatly into predictable little columns.

00:00:27 Host 1
But step into the world of artificial intelligence adoption. And suddenly that pristine financial blueprint is just covered in invisible ink. The spending landscape we are looking at today is murky, it is unpredictable, and man, it is moving incredibly fast.

00:00:43 Host 2
It is the absolute definition of diagnostic muddy waters for capital planning teams. And I should clarify right up front, today’s deep dive isn’t about the technical code of AI.

00:00:53 Host 1
Right, no algorithms today.

00:00:55 Host 2
Yeah, we aren’t analyzing neural networks. This is a finance leadership discussion. Focusing entirely on investment prioritization, enterprise capability, and shifting from those operational reflexes to strategic foresight.

00:01:08 Host 1
Okay, let’s unpack this. Treating AI spending as a routine operational expenditure, OpEx, is a massive strategic trap.

00:01:16 Host 2
A huge trap.

00:01:17 Host 1
And to understand why, we have to look at how this technology actually enters a business. Because AI rarely starts in the boardroom with a dramatically stamped budget approval.

00:01:27 Host 2
Almost never.

00:01:28 Host 1
So let’s use a real-world scenario drawn from our sources. Imagine your HR department decides they need help sorting through a massive backlog of job applications.

00:01:37 Host 2
A very common problem.

00:01:38 Host 1
Right. So they find an AI resume screener. It’s a simple software-as-a-service tool. They just put a $20-a-month subscription on a corporate credit card. Or maybe your procurement vendor just casually says, hey, we added an AI forecasting co-pilot to the software you already use for a tiny monthly fee.

00:01:54 Host 2
And because it sits inside those operating budgets, it looks incredibly small. It looks entirely practical at the point of approval. I mean, a department head signing off on a new procurement automation tool or an HR resume screener. That does not trigger a board-level review. It just enters through OpEx.

00:02:11 Host 1
But wait, let me play devil’s advocate here. If the initial spend is only 20 bucks a month, who really cares if it’s labeled OpEx? At the end of the day, money is money, right? Why does the label actually matter?

00:02:22 Host 2
What’s fascinating here is how the label dictates the governance.

00:02:25 Host 1
Okay, how so?

00:02:26 Host 2
Well, the issue is not whether AI is classified as OpEx or CapEx from an accounting standpoint. The mistake isn’t funding AI out of an operational budget.

00:02:36 Host 1
Then what is the mistake?

00:02:37 Host 2
The mistake is allowing strategically significant decisions to masquerade as routine procurement.

00:02:42 Host 1
Oh, wow.

00:02:43 Host 2
Yeah, when you label it routine OpEx, finance assesses the immediate cost without assessing the long-term impact. The decision passes through a mid-level manager, but it completely bypasses enterprise-level tests.

00:02:55 Host 1
Tests for like data security.

00:02:57 Host 2
Data security, duplication of effort, or even just long-term capability value.

00:03:01 Host 1
It sounds like signing up for what you think is a $10 streaming service, but six months later, you realize you’ve accidentally outsourced your entire home mortgage to them.

00:03:10 Host 2
That is painfully accurate because AI projects move shockingly fast from these small, isolated use cases to absolute business dependency.

00:03:19 Host 1
Let’s stick with that HR resume screener then. So six months in, usage expands. The team loves it.

00:03:26 Host 2
Naturally.

00:03:26 Host 1
And costs grow from $20 to maybe $5,000 a month because they buy more seats. Then data requirements emerge.

00:03:35 Host 2
This is where it gets sticky.

00:03:36 Host 1
Right. HR wants to integrate the AI with the internal employee database to track long-term retention metrics. Workflows change. The HR team literally forgets how to process a raw resume without the AI tool screening it first.

00:03:49 Host 2
And suddenly the core capability of your organization is locked into external platforms or external vendors. You’re beginning to build, buy, or rely on a massive capability.

00:03:58 Host 1
Without really realizing it.

00:03:59 Host 2
Exactly. By the time the investment actually looks large enough in dollars for formal capital planning or portfolio governance, the organization has already committed.

00:04:07 Host 1
They’re already in too deep.

00:04:08 Host 2
Yeah, committed to tools, operating assumptions, and data pathways that are incredibly difficult to unwind. You are making a strategic decision about the future of the company’s capabilities, but you are governing it like you’re buying office supplies.

00:04:23 Host 1
That is wild, governing the future of the company like you’re buying pens and paper. So if OpEx is blinding us and letting these massive strategic dependencies just sneak through the back door, the natural reaction for any finance leader listening to this is probably to just slam the door.

00:04:39 Host 2
Right, just shut it down.

00:04:40 Host 1
Just push all AI spending into CapEx, put it through the rigorous capital planning gauntlet, problem solved.

00:04:46 Host 2
You would think so, but as our sources clearly lay out, CapEx is historically ill-equipped to handle AI.

00:04:53 Host 1
Why is that?

00:04:54 Host 2
Traditional capital planning was built for physical, predictable investments. You know, the planned upgrade, an asset replacement, a new logistics center?

00:05:03 Host 1
The tangible things.

00:05:04 Host 2
Exactly. Those traditional investments require a stable scope, a predictable cost profile, a clear implementation pathway, and an upfront deterministic return on investment.

00:05:14 Host 1
Let’s clarify that for a second. Deterministic ROI model, meaning a financial model that demands you know exactly how many dollars a project will return to the business before you’ve even spent your first dime.

00:05:25 Host 2
Right. It expects the future to behave predictably based on the capital you deploy today.

00:05:31 Host 1
Which is pretty rigid.

00:05:32 Host 2
Very rigid. Even if a physical plant upgrade carries some construction risk, the investment is bounded enough that conventional evaluation works. You can estimate exactly how many widgets the new machine will produce. But AI breaks every single one of those rules. The scope is going to evolve as the technology itself matures, use cases expand, and entirely new capabilities become available mid-flight.

00:05:57 Host 1
Applying CapEx models to AI is like, it’s like trying to budget for a cross-country road trip by pre-paying for exactly 14.2 gallons of gas.

00:06:06 Host 2
Before you know if you’re driving a hybrid sedan or a massive RV.

00:06:09 Host 1
Yes. The vehicle literally changes while you’re driving it.

00:06:13 Host 2
That is exactly it. Costs shift constantly as usage grows. You integrate the tool more deeply, your data storage requirements increase, API token usage explodes, and suddenly your ongoing support costs look very different.

00:06:27 Host 1
I bet.

00:06:27 Host 2
Plus, the benefits are notoriously hard to pin down in a single fixed business case.

00:06:33 Host 1
Because it’s not just one department benefiting.

00:06:35 Host 2
Exactly. The value might not sit within a single function. It emerges through better productivity, higher decision quality, customer experience improvements, and enterprise learning. All that compounds over time across multiple departments.

00:06:48 Host 1
So annual planning cycles are basically the enemy of AI adoption.

00:06:52 Host 2
They really are.

00:06:53 Host 1
I mean, it makes zero sense to lock in funding based on a static priority list from January when the technology itself fundamentally changes by March.

00:07:00 Host 2
And this raises an important question about how our traditional governance models actively distort decision-making.

00:07:06 Host 1
Okay, tell me more about that.

00:07:07 Host 2
When you force traditional CapEx models onto AI, you get bad outcomes. Projects get delayed because the total compounding value isn’t fully visible upfront to satisfy that deterministic ROI requirement we talked about.

00:07:21 Host 1
Right, it can’t prove its worth on day one.

00:07:23 Host 2
Exactly. Or they get approved too narrowly, like a committee only approves the very first HR use case and ignores how that data pipeline could eventually help operations.

00:07:33 Host 1
They miss the big picture.

00:07:34 Host 2
Or they get rejected entirely because they don’t fit a rigid model designed for a slower, more isolated world. Deterministic assumptions actively penalize investments where the value grows and evolves through continuous learning.

00:07:47 Host 1
Okay, so if OpEx is blinding us to the risk and CapEx is suffocating the innovation before it even starts, where does that leave finance teams? Are they just stuck in this loop forever?

00:07:58 Host 2
Well, this dangerous middle ground is exactly what emerges in most organizations. It’s disjointed, fragmented AI adoption spread across the enterprise.

00:08:07 Host 1
Which sounds chaotic.

00:08:08 Host 2
It is. AI activity is everywhere right now, but coordinated AI investment is almost non-existent. Without a centralized, coordinated governance lens, AI projects just multiply across the business without ever forming a coherent enterprise capability.

00:08:23 Host 1
Here’s where it gets really interesting. The sources paint a vivid picture of this. HR has their resume screener on a corporate card. Over in sales, they are using a totally different vendor’s predictive analytics tool. Operations is trialing an inventory co-pilot they bought out of their quarterly surplus. They are all doing their own thing, paying different vendors and using different data standards.

00:08:45 Host 2
And each of those initiatives might be useful locally to that specific team, but the overall pattern becomes incredibly fragmented.

00:08:52 Host 1
You get a long tail of disconnected AI activity.

00:08:55 Host 2
Exactly. Siloed pilots. Duplicated spend. Disconnected tooling. This is the definition of shadow AI. Teams are procuring and utilizing AI tools completely off the radar of IT and enterprise finance.

00:09:08 Host 1
It’s like 10 different departments independently trying to invent the wheel, but one is buying train tracks, one is buying jet fuel, and one is buying bicycles.

00:09:15 Host 2
And none of it connects.

00:09:16 Host 1
None of it gets the whole company from point A to point B.

00:09:19 Host 2
And the tragedy is what the business loses in the process. When teams solve similar problems separately, the organization loses the opportunity to build shared data foundations.

00:09:30 Host 1
Yeah, that’s a huge miss.

00:09:32 Host 2
They miss out on reusable platforms. The success stays local. Say the sales team learned something great about prompting their AI tool.

00:09:40 Host 1
But that lesson is never transferred to HR.

00:09:42 Host 2
Exactly. The organization is spending the money, bearing all the cost and friction of transformation, but actual enterprise capability stalls because there is no scale.

00:09:52 Host 1
Okay, so we’ve established that OpEx hides the danger, CapEx kills the innovation, and doing nothing leaves you with a fragmented mess of shadow AI.

00:10:00 Host 2
Pretty bleak summary, but yes.

00:10:02 Host 1
But the sources don’t just leave us with the problem, thankfully. They introduce a completely new operating model. Strategic expenditure, or StratEx. So it’s StratEx. The core formula laid out in the material is StratEx equals CapEx plus OpEx.

00:10:15 Host 2
It recognizes that modern strategic investment spans both accounting categories. Let’s look back at that HR resume screener that ballooned into an enterprise dependency. The subscription, the model usage, the cloud consumption, the vendor capability itself, that might sit cleanly in OpEx. But the data infrastructure required to run it securely, the internal integration to your employee database, the implementation programs, it sits in CapEx.

00:10:43 Host 1
Traditionally, those are separated by accounting classifications, which completely fractures the governance. The OpEx part gets approved operationally by the HR director. The CapEx part goes through IT project controls. So the strategic impact is just split across multiple functions, budgets and planning cycles.

00:11:00 Host 2
And that separation might perfectly satisfy accounting rules, but it provides absolutely zero investment governance.

00:11:06 Host 1
None at all.

00:11:07 Host 2
StratEx treats the business decision as one unified choice. You stop asking, is this CapEx or OpEx? And you start asking, is this initiative strategically significant enough to require structured evaluation, prioritization, and ongoing review?

00:11:21 Host 1
So what does this all mean for you, the listener? If I’m a CFO or an FP&A director tuning in right now, what am I telling procurement or IT to flag? I can’t have my executive team review every $20 SaaS tool that an intern downloads.

00:11:37 Host 2
No, of course not. You establish specific strategic thresholds. It isn’t just about the initial dollar amount.

00:11:43 Host 1
Okay, what is it about?

00:11:45 Host 2
A StratEx trigger might be activated when an AI tool requests connection to tier-one enterprise data, like customer records or employee databases.

00:11:53 Host 1
Oh, that makes a lot of sense.

00:11:54 Host 2
Or it might trigger when a tool alters a core workflow that impacts compliance, or even when user adoption crosses a certain percentage of a department. Once a tool crosses any of those thresholds, it shifts from operational procurement into the StratEx portfolio.

00:12:09 Host 1
So let’s walk that HR tool through a StratEx review then. HR buys the $20 tool. Procurement sees it’s cheap, but notices it requires API access to the secure employee database. Boom.

00:12:21 Host 2
The trigger is pulled.

00:12:22 Host 1
It is no longer just a routine expense. It’s now categorized as a StratEx initiative.

00:12:27 Host 2
Exactly. And now you need a different operating model to govern it. First is visibility. You add this HR tool to your enterprise portfolio.

00:12:34 Host 1
Keeping it all in one place.

00:12:36 Host 2
Perhaps leveraging tools like Stratex Online to maintain portfolio visibility across both CapEx and OpEx. You can now see the HR tool alongside the sales analytics tool and the operations co-pilot in one unified dashboard.

00:12:50 Host 1
You assess the projects at the portfolio level, not just as individual requests in a vacuum.

00:12:55 Host 2
Yes, and you might see that HR and sales are both paying different vendors for essentially the same natural language processing capability, and you can consolidate that spend.

00:13:03 Host 1
Oh, nice.

00:13:03 Host 2
And crucially, this changes the cadence of governance. Annual evaluation is far too slow for an AI portfolio.

00:13:10 Host 1
Yeah, we established that January to March is a lifetime in AI.

00:13:13 Host 2
Right, you need continuous review and dynamic prioritization. So in a quarterly StratEx review, finance looks at the HR tool. The initial metrics are great, the tool is learning, but to scale it safely across the whole company, IT realizes they need $50,000 in CapEx to build a secure proprietary data pipeline.

00:13:32 Host 1
Under the old model, HR would have to wait until the next annual planning cycle in January to request that CapEx, basically stalling the project for nine months and killing all the momentum.

00:13:42 Host 2
Exactly. But under StratEx, because you review continuously and assess the unified strategic value, you can dynamically reallocate funding from a lower-priority project right then and there. Wow. The evaluation model expands beyond just financial return to include operational impact, shared capability, and strategic alignment. You govern the capability, not just the accounting bucket.

00:14:06 Host 1
Implementing continuous portfolio-level StratEx reviews. I mean, for some finance executives, that sounds like a lot of heavy lifting that might actually slow down agility.

00:14:15 Host 2
It’s a common fear.

00:14:16 Host 1
But the sources point out that evaluating AI requires a completely new definition of risk to ensure the business doesn’t paralyze itself while trying to govern.

00:14:24 Host 2
Most traditional investment frameworks, especially in finance, are obsessed with evaluating the risk of action.

00:14:29 Host 1
Right. What happens if the project fails? What if implementation takes longer? What if costs spiral out of control or nobody actually uses the new software?

00:14:38 Host 2
Those questions still matter, obviously. AI projects carry massive implementation risks. Models hallucinate. Internal data isn’t clean enough to train on.

00:14:47 Host 1
Change management is always harder than people think.

00:14:50 Host 2
Vendor decisions can create dependencies that are like concrete. They absolutely matter. But AI introduces a massive, often completely unmeasured risk into the equation. The risk of inaction.

00:15:02 Host 1
The risk of doing nothing.

00:15:03 Host 2
What happens if the organization waits too long?

00:15:06 Host 1
But let me play devil’s advocate again here, because I hear this all the time from business leaders. Isn’t it safer to just let the competitors make the expensive mistakes first? Yeah, let them bleed cash, figuring out the messy integrations, the data pipelines, the user training. And then we just swoop in and buy the polished out-of-the-box AI tool a year later when the market matures.

00:15:29 Host 2
It’s a really common thought process, but it is deeply flawed in the context of AI.

00:15:33 Host 1
Why?

00:15:34 Host 2
The sources dismantle this by explaining the concept of operational lag. You can certainly buy a polished software interface a year later. What you cannot instantly buy is the enterprise learning. You cannot instantly buy the integrated data foundation that your competitor has been building and refining for 12 months. You cannot buy the culture of adaptation.

00:15:55 Host 1
Because AI models train on company-specific data. While you were waiting for the safe investment, your competitor used messy, early AI to figure out how to lower their cost base.

00:16:07 Host 2
Exactly. Their model is getting smarter specifically about their business. They shifted customer expectations. Your internal capability gaps widened. They learned, tested, and scaled.

00:16:17 Host 1
Wow. So waiting isn’t actually safe at all.

00:16:19 Host 2
In a slower investment environment, waiting could genuinely be treated as caution. In an AI-enabled environment, waiting is an active decision and it carries a heavy, heavy cost. The risk of inaction shows up as competitive disadvantage and widening capability gaps. It doesn’t appear immediately in a traditional financial model, but it shapes the future of the business just as materially as a failed multi-million-dollar capital project.

00:16:43 Host 1
So if we connect this to the bigger picture, your risk evaluation model must look at both sides of the ledger. Finance teams must weigh what could go wrong if we proceed equally against what will be lost if we do not.

00:16:56 Host 2
Precisely. You do not have to abandon financial discipline to move faster, but you must recognize that delay and fragmented experimentation have a very real price tag.

00:17:06 Host 1
Yeah, it requires moving from isolated approval decisions to a structured portfolio-level discipline.

00:17:12 Host 2
Exactly.

00:17:13 Host 1
Well, this has been incredibly eye-opening. To recap, AI is bypassing the board and entering the business as stealthy, routine OpEx. But because it scales so fast and builds rapid dependency, it completely breaks the rigid, static rules of traditional CapEx planning. And if left alone, it causes fragmented, wasteful shadow AI across different departments. Ultimately, it demands a new mindset, strategic expenditure, or StratEx, that actively governs the unified business decision dynamically while carefully calculating the dangerous cost of falling behind.

00:17:45 Host 2
It forces us to govern the strategic significance of the spend rather than just classifying the transaction.

00:17:51 Host 1
So as we wrap up, I want to leave you with a specific question to mull over. Think about your own organization right now. If you asked your department heads to list every small AI subscription, every cloud token usage or embedded AI vendor tool they are currently relying on, how long would that list be?

00:18:09 Host 2
Probably longer than you think.

00:18:10 Host 1
And more importantly, could your business still function tomorrow if all those routine tools suddenly vanished?

00:18:16 Host 2
That is the true measure of strategic dependency.

00:18:18 Host 1
Thank you for joining us on this deep dive into the true cost and strategic weight of AI. Take these insights straight into your next capital planning meeting.