CFOs Are Pumping the Brakes on AI Spending—What Does This Mean for Engineering Roadmaps?

General
1

I just got out of our Q2 planning meeting, and the vibe was… different. Our CFO, who six months ago was asking “why aren’t we doing more with AI?”, just put a hard pause on two of our AI initiatives until we can show clearer ROI projections.

Turns out we’re not alone. I’ve been digging into this and found that only 14% of CFOs report seeing clear, measurable ROI from their AI investments so far. Even more telling: AI spending is shifting out of innovation and R&D budgets (where ROI requirements were loose) into operational technology budgets—meaning AI projects now face the same scrutiny as ERP implementations or headcount decisions.

The Budget Reality Check

Here’s what I’m seeing change in real-time:

Before (2024-2025): “Let’s experiment with AI across the org! Innovation budget approved.”

Now (2026): “Show me the business case. What’s the payback period? How does this compare to hiring two more engineers?”

The shift makes sense from a finance perspective. After 18+ months of investment, CFOs want to see results. But it’s creating some tough conversations for those of us trying to balance innovation with accountability.

What This Means for Product & Engineering Roadmaps

For my team, this means:

  1. Re-prioritizing mid-flight initiatives - We have 3 AI projects in progress. Two are getting paused, one is getting doubled down on because it has clear usage metrics.

  2. Better business case discipline - I’m now building financial models for AI features the same way I would for pricing changes or new product lines.

  3. Shorter proof-of-concept cycles - No more “let’s explore this for 6 months.” We need signal in weeks, not quarters.

  4. Cross-functional alignment on metrics - Product, engineering, and finance need to agree upfront on what success looks like.

The Framework I’m Using Now

When pitching continued AI investment, I’m forcing myself to answer:

  • What business metric moves? (Not “efficiency improves” but “support ticket resolution time drops 30%”)
  • What’s the alternative cost? (What if we hired people instead or bought an off-the-shelf solution?)
  • What’s the timeline to value? (Quarters, not years)
  • What’s the kill criteria? (If we don’t see X by Y date, we stop)

Questions for the Community

I’m curious how others are navigating this:

  • Are you seeing similar budget pressure around AI investments?
  • How are engineering leaders making the case for continued investment?
  • What metrics are actually moving the needle with your CFO?
  • Is anyone successfully protecting “explore” budget while also showing accountability?

The optimist in me thinks this is healthy pressure that will separate real AI value from hype. The realist in me worries we’re about to learn the wrong lessons and overcorrect just as the technology gets truly useful.

What’s your experience been?

2

This budget pressure is exactly what we needed, honestly.

I know that sounds counterintuitive, but hear me out. When AI spending lived in innovation budgets with loose ROI requirements, we were flying blind. We’d spin up projects because they sounded cool or a competitor announced something similar. The result? A graveyard of half-finished AI experiments and a lot of team cynicism about “initiative of the month.”

What Happened When We Got Scrutinized

Last year, our CFO put similar pressure on our AI investments. My initial reaction was defensive—“they don’t understand innovation!” But the scrutiny actually forced us to answer questions we should have been asking all along.

We had seven AI projects running. Under the new lens:

  • Three got killed immediately - couldn’t articulate business value beyond “sounds futuristic”
  • Two got combined - turned out they were solving adjacent problems
  • Two got doubled-down - clear metrics, clear value, clear timeline

The team morale actually improved after we cut the noise. Engineers want to work on things that matter, not science experiments that leadership doesn’t really care about.

The Three Questions Every AI Project Should Answer

Here’s what I now require before any AI initiative gets greenlit:

1. What manual process or business problem does this eliminate?

Not “makes things faster” (too vague) but “eliminates the 40 hours/week our support team spends categorizing tickets.” Specificity matters.

2. What happens if we do nothing?

Sometimes the answer is “we fall behind competitors.” Sometimes it’s “literally nothing changes.” The latter doesn’t get funded.

3. Can we measure success in the same way we measure our core business?

If your AI project improves user activation, measure activation rate. If it reduces costs, measure cost per transaction. Use existing business metrics, not invented AI metrics.

The Unexpected Benefit

What surprised me most: this actually accelerates the valuable AI projects.

When you’re not spreading budget and attention across seven mediocre bets, you can properly resource the two that matter. We went from having 5 engineers each split across multiple AI projects to having two focused teams of 8-10 engineers each. Velocity went way up.

@product_david your framework is spot on. I’d add one more question: “Who is the internal champion with budget authority who will be measured on this outcome?” If you can’t name that person, you don’t have organizational buy-in yet.

The era of AI for AI’s sake is over. Good riddance. Now we get to do the real work.

3

Coming from financial services, this budget scrutiny feels… normal? We’ve always had to justify technology investment with the same rigor as hiring decisions or compliance spend.

What’s interesting is watching the rest of the industry catch up to what regulated industries have been doing all along.

The CFO Language Translation

@product_david you mentioned building financial models for AI features now. That’s the key—engineering leaders need to speak CFO language, not ask CFOs to learn engineering language.

Here’s what’s worked for me when making the case for AI investment to finance:

Instead of: “This will improve model accuracy by 15%”
Say: “This reduces fraud losses by an estimated $2.3M annually, with 6-month payback”

Instead of: “We need to invest in our ML infrastructure”
Say: “Current manual review process costs $180/hour in analyst time. Automation brings that to $12/hour with higher accuracy”

Instead of: “Our competitors are doing this”
Say: “Customer churn is 8% higher in segments where we lack AI-powered features. That’s $4.1M in lost revenue.”

The Metrics Finance Actually Cares About

In my experience, CFOs respond to three categories:

  1. Revenue impact - Does this help us sell more, retain customers, or enable premium pricing?
  2. Cost reduction - Does this eliminate headcount need, reduce infrastructure costs, or prevent losses?
  3. Risk mitigation - Does this prevent regulatory fines, reduce fraud, or improve compliance?

Everything else is interesting but not compelling.

The One Exception: Strategic Positioning

There IS budget for “we need to build this capability or we’ll be obsolete in 3 years” arguments. But you can’t make that case for everything. Pick your one or two strategic bets carefully.

@vp_eng_keisha you mentioned focusing resources on two teams instead of spreading thin—we did exactly the same thing. Went from 11 AI “workstreams” to 3 properly staffed initiatives. Made all the difference.

The hardest part for engineering leaders is accepting that we can’t do everything. CFO scrutiny is just forcing that prioritization earlier.

4

This moment reminds me of 2018 when cloud migration was getting the same scrutiny. “Why are we spending millions to move to AWS when our data centers work fine?”

The CFOs who demanded ROI proof for cloud weren’t wrong—they separated the companies who migrated strategically from those who just followed hype. Same thing is happening now with AI.

This Is Actually a Strategic Opportunity

What I’m seeing: This budget pressure is forcing the conversation from point solutions to enterprise-wide AI strategy.

When AI lived in innovation budgets, every team could spin up their own AI project. We ended up with:

  • 23 different ML models across the org (with overlap)
  • 4 different AI vendors with redundant contracts
  • No shared infrastructure or learnings
  • Massive technical debt

CFO scrutiny is killing that chaos. Now we’re asking: “What are the 3-5 AI capabilities that matter for our entire business?” and building those as platform investments.

The Portfolio Approach

@product_david you mentioned re-prioritizing mid-flight initiatives. Here’s the framework I’m using for AI portfolio management:

Tier 1: Core operations (60% of budget)
Clear ROI, proven technology, directly supports existing business model. These are table stakes—customer service automation, fraud detection, personalization. Defend these with hard metrics.

Tier 2: Strategic enablers (30% of budget)
Building capabilities that enable new products or business models. Harder ROI to prove upfront, but directionally aligned with where the business is going. Defend these with OKRs and quarterly checkpoints.

Tier 3: Exploration (10% of budget)
Genuine R&D where we don’t know the answer yet. Accept that most will fail. Defend the existence of this bucket, not individual projects within it.

The Warning Nobody Wants to Hear

Here’s my concern: Don’t let perfect ROI measurement kill innovation entirely.

I’ve seen this movie before. In the late 2000s, companies that cut all R&D during the recession came out weaker. The ones that protected strategic bets (even without proven ROI) came out stronger.

The difference between pivoting and quitting is whether you’re learning. If your AI experiments are generating organizational learning about where the technology works and doesn’t work, that has value even if individual projects fail.

@eng_director_luis is right that we need to speak CFO language. But CTOs also need to help CFOs understand that some valuable investments don’t have provable ROI upfront. That’s why it’s called innovation, not optimization.

The companies that thread this needle—accountability for most spending, protection for genuine exploration—are going to win the next 5 years.

5

Reading this thread as someone who lived through a startup failure partially because we ignored unit economics… y’all are describing my nightmares :sweat_smile:

When “Move Fast” Meets “Show Me The Money”

My last company (RIP) raised a Series A on the promise of AI-powered design tools. We spent 18 months building cool ML features that designers loved in demos. When it came time to raise Series B, investors asked the same questions your CFOs are asking:

  • What’s the CAC?
  • What’s the retention?
  • What’s the path to profitability?

We… didn’t have great answers. We had engagement metrics and NPS scores and “users love this!” But we couldn’t draw a straight line from our AI investment to revenue.

We failed. Not because the product was bad, but because we optimized for innovation theater instead of business fundamentals.

The Design Systems Parallel

Now I lead design systems at an agency, and I’m facing a mini version of this same conversation. “Why should we invest in a design system when designers can just… design?”

The turning point was when I stopped talking about consistency and craft (designer language) and started talking about time-to-market and rework costs (business language).

  • Before design system: 40 hours to design + build a new feature
  • After design system: 12 hours (reusable components)
  • ROI: Clear reduction in design and eng time

@eng_director_luis your “CFO language translation” examples are :fire: - that’s exactly the shift I had to make.

My Question for This Group

How do you measure AI impact on user experience quality?

Like, our AI features might make the product more delightful or reduce user frustration, but that’s hard to put a dollar value on until it shows up in retention metrics (which lag by months).

@product_david I love that you’re forcing the question “what if we hired people instead?” That was the calculation that killed my startup—it turned out hiring one really good designer was more cost-effective than our AI personalization feature. Hard truth.

I guess my takeaway: I’m team “healthy pressure.” But also team “don’t let the pendulum swing too far the other way.” Some of the best product experiences are the result of someone caring about craft even when the immediate ROI isn’t obvious.