Both are wrong.

I’ve spent the last few years working hands-on with AI tools across finance functions and M&A integrations, and the picture from inside the work is far more boring than the headlines and far more useful. A growing body of rigorous research from MIT, Stanford, Harvard, BCG, McKinsey, the OECD, and Gartner converges on the same conclusion: AI augments people, amplifies the quality of their work, and expands what organizations can achieve. In the near term, it does not replace the humans who actually make businesses run.

This isn’t wishful thinking. It’s what the evidence shows, and it’s what I’ve watched happen in real finance teams.

AI makes people better at their jobs

Starting with the study that anchors everything. Brynjolfsson (Stanford) and Li & Raymond (MIT) tracked 5,179 customer support agents across roughly three million chats, published in the Quarterly Journal of Economics.

The results were not subtle. Overall productivity rose 14% — more issues resolved per hour. But the gains weren’t evenly spread, and that’s the interesting part. Lower-skilled and newer workers improved by 35%, in both speed and quality. AI effectively compressed the experience curve: two months of AI-assisted work produced the competence of six months unassisted.

Crucially, workers weren’t blindly obeying the machine. They followed AI suggestions about 38% of the time, exercising professional judgment throughout. What the AI did was disseminate the best practices of top performers across the entire team, acting as a real-time coach. Customers were more polite. Retention went up.

The headline here isn’t “AI replaced the agents.” It’s “AI made the agents better — especially the ones who needed it most.”

AI amplifies expertise. It doesn’t substitute for it.

Here’s the critical caveat, and it’s where a lot of the breathless coverage falls apart: AI amplifies existing expertise. It does not manufacture it.

A 2025 Harvard Business School study (Bojinov, McFowland III, et al.) of IG Group employees found that AI lets workers with adjacent skills produce expert-quality output — but it hits a wall when the knowledge distance is too great. Three worker types tell the whole story:

• The domain expert (a web analyst) used AI as a force multiplier — already expert-level output, now delivered faster.

• The adjacent-skill worker (a marketing specialist) used AI to close the gap, producing output that nearly matched the experts.

• The distant-skill worker (a software developer working outside their lane) hit the wall — still lagging by 13% despite using the identical AI model.

The time savings were universal — conceptualization dropped 63% and writing tasks 75% across the board. The difference wasn’t effort or speed. It was output quality, and quality still tracks human expertise. The person matters profoundly.

Augmentation, not elimination

So how big is the change? Large, but lopsided toward reshaping rather than eliminating.

The BCG Henderson Institute estimates 50–55% of US jobs will be reshaped within three years — same or similar roles, but fundamentally new ways of working. By contrast, BCG puts outright job elimination at 10–15% over five years. Considerable, yes. But the vast majority of workers are being augmented, not displaced. And BCG’s modeling shows that when AI-driven productivity triggers more demand for end products, the need for human roles tends to grow, not shrink.

The productivity numbers are real money. LSE research pegs AI savings at roughly £14,000 per employee per year — about one full day a week. But notice the gap: trained employees saved 11 hours a week, untrained ones only 5. A 2x differential that comes down entirely to whether the organization bothered to teach people how to use the tools.

And it’s not making people miserable. BCG’s 2026 AI at Work report (11,749 employees across 14 markets) found that two in three regular AI users say it improved their job satisfaction.

The future workforce is humans with machines

McKinsey frames it cleanly: the future workforce isn’t humans or machines, it’s humans with machines. They size the long-term prize at up to $4.4 trillion in added productivity potential from corporate AI use cases alone.

But there’s a paradox worth sitting with. Over 80% of businesses using AI haven’t yet seen meaningful earnings gains. The reason is almost always the same: they’re stuck in “copilot mode,” running isolated tools bolted onto existing workflows, never unlocking systemic value. The value curve runs from copilot mode, to integrated workflows, to genuinely agentic AI, and most organizations have stalled at step one.

The OECD’s July 2025 review backs the underlying gains: productivity improvements of 5% to over 25% in customer support, software development, and consulting, with the largest gains going to less-experienced workers and the pattern holding across geographies.

The finance function is where the thesis is most testable

This is the part I care about most, because finance is where you can actually measure whether the augmentation thesis holds. Cycle time, accuracy, working capital, forecast quality, decision speed; all quantifiable, all on the dashboard.

The PE world is moving. Per the Accordion / Ramp AI in PE Index (April 2026): 59% of PE-backed firms now hold paid AI subscriptions, ahead of the broader market (41%) but behind venture-backed peers (77%). Manufacturing AI adoption across PE portfolios doubled from 27% to 52% in a single year; healthcare and life sciences rose from 23% to 48%. Yet fewer than a third of CFOs have implemented AI meaningfully, with “I don’t know where to begin” as the primary obstacle.

For those who do begin, the finance-function results are concrete and already documented, not projected:

• 30% faster close cycles: month-end close reduced by up to a third

• 15% DSO reduction: directly supporting cash flow

• 35% better forecast accuracy: at one PE-backed industrial company, within a single quarter

• 8% working capital reduction: alongside halved reporting time, also within one quarter

EY’s read is that the highest-value finance organizations pair process standardization with AI-driven automation, elevating teams from transactional processing to strategic analysis. PwC’s 2026 PE framework makes the exit case directly: AI-enabled insight helps middle-market portfolio companies grow revenue, expand margins, and look more attractive to buyers. These are EBITDA-improving, exit-value-enhancing outcomes.

Gartner puts the most ambitious number on it. Their Finance Technology Bullseye report, examining 60+ technologies, projects up to 10 margin points by 2029 for CFOs who deploy AI strategically. But it’s conditional. The CFOs who get there manage technology as a portfolio rather than chasing isolated pilots, and they embed AI into core processes rather than layering it on top. Three-quarters of CFOs are raising tech budgets in 2026; nearly half by 10% or more.

The honest nuance: where AI does displace

I’d be a poor advocate for evidence if I only cited the comfortable parts.

The WEF estimates 85 million jobs displaced by AI — but 97 million new roles emerging, for a net gain of roughly 12 million globally. Stanford’s Digital Economy Lab (November 2025) found a 6% employment decline for workers aged 22–25 in the most AI-exposed occupations since late 2022. Anthropic’s February 2026 analysis flags clerical and cognitively structured roles as facing the highest near-term exposure.

Displacement is real, and it’s concentrated. But the displacement literature tends to miss the demand-expansion effect BCG identifies: when augmented workers produce more and better output, markets grow and demand for human labor can increase. The canonical example is the ATM. It didn’t eliminate bank tellers — it cut tellers per branch while improving branch unit economics enough that the number of branches grew. Net teller employment held.

The decisive variable is strategic clarity. Organizations that genuinely reinvent how they work — letting humans and AI collaborate — create far more value than those simply bolting AI onto yesterday’s workflows.

The real barrier isn’t the AI

If you take one practical thing from this, take this: the barrier to realizing AI gains usually isn’t the AI. It’s the state of your data infrastructure.

Accordion’s research identifies three blockers that show up again and again in PE portfolio companies: fragmented ERP systems, inconsistent data definitions, and weak data governance. These are particularly acute in buyout-backed companies, which tend to carry older systems and more governance layers than their venture-backed counterparts. EY’s prescription is to close system gaps, standardize processes, and eliminate complexity first.

AI amplifies the quality of your data. It cannot rescue bad data.

What this means for business leaders

Five takeaways I’d put in front of any CEO or board:

1. Invest in training. Trained employees save 11 hours a week versus 5 for the untrained — yet 68% of employees received no AI training in the past 12 months. That’s value sitting unrealized on the table.

2. Treat AI as a quality lever, not just a speed lever. Novice workers using AI produce better output, not just faster output. That’s the CFO and CEO case.

3. Go after finance-function ROI. 30% faster close, 15% DSO improvement, 35% better forecast accuracy, 8% working capital reduction. The gains are sitting right there.

4. Understand the margin prize is conditional. Gartner’s 10-point, 2029 projection requires portfolio management and process integration. CFOs chasing isolated use cases will fall short.

5. Fix the data foundation first. Standardize before you scale. AI amplifies data quality; it can’t manufacture it.

The bottom line

The AI augmentation thesis isn’t optimism bias. It’s the weight of evidence from the most rigorous economic research of the last three years.

From where I sit, an operationally focused CFO with significant M&A integration experience, the finance function is where that thesis is most testable, because it’s where we can measure cycle time, accuracy, working capital, forecast quality, and decision speed. The results are already in.

When we train our people to use these tools and standardize our enterprise data, we make our people more capable, our controls stronger, and our organizations more valuable. That’s not AI replacing people. That’s AI making people, and the businesses they run, worth more.

I walked through the full evidence base in a companion presentation here. Happy to discuss how the augmentation thesis applies to your finance function or M&A integration work.

The natural question is what changes when AI is introduced into that model.

The sequencing does not change. In acquisition-led businesses, speed to financial clarity remains the priority. The objective is still to understand performance quickly, validate it against the investment thesis, and establish a reporting capability that can be deployed repeatedly as the business grows. What AI changes is the pace and scalability of execution.

Activities that previously required weeks of manual effort, such as chart of account mapping, transaction classification, variance analysis, reconciliations, and data validation, can now be significantly accelerated. Management reporting, including first-draft lender and board packs, can be produced faster and with greater consistency. This fundamentally changes the capacity of the finance function, as small teams can deliver high-quality, decision-grade reporting at a pace that supports rapid integration and ongoing acquisition activity.

In practice, this means reducing the time between acquisition and financial visibility. The faster you can establish reliable reporting, the faster you can assess whether the business is performing in line with the thesis and take corrective action where required.

It also strengthens repeatability. AI makes it easier to standardize reporting structures, onboard new entities, and maintain consistency across an increasingly complex group. The model built in Part 1 becomes easier to deploy again and again, without a linear increase in cost or effort.

The constraints, however, do not change. AI does not create accountability. It cannot enforce close timelines, define ownership of numbers, or resolve ambiguity in how performance is measured, and it cannot replace the governance and discipline required to produce reliable outputs. If those foundations are weak, AI will scale inconsistency rather than clarity.

The role of the CFO remains unchanged. It is to ensure the business has timely, reliable financial insight, with performance measured clearly against the drivers of value in the deal thesis. AI is a force multiplier, but only when the underlying operating model is sound. Adopting AI for its own sake achieves nothing. The advantage is in compressing the time to financial clarity and making that capability scalable across every acquisition.

Lenders, investors, and boards are still underwriting the same thing: how quickly you can produce numbers they trust, how consistently you can repeat that process, and how clearly those numbers demonstrate performance against the deal thesis.

Make Acquisitions Great Again by The Tech-Savvy CFO

Speed to Financial Clarity: Building Lender-Ready Reporting in 10 Weeks

In acquisition-led businesses, value is created or lost against the deal thesis, often within the first six to twelve months. Reliable financial reporting early is what lets you understand the business and track performance against that thesis with confidence, and it has to be repeatable at scale as further acquisitions follow…

Read more

4 months ago · one137th

As Group CFO of a private equity-backed defense industrial business, I stepped into an organization that had recently completed its third acquisition and was preparing for another. Operations spanned three states, three accounting systems, two manufacturing execution systems, and several distinct finance processes. The immediate requirement was not system transformation. It was speed to financial clarity.

The lender expectations were already in place: quarterly reporting, covenant compliance, budgets, forecasts, headcount reporting, and operational KPIs. Mezzanine financing had been raised before my arrival, but none of the required reporting had been fully established. The accounting team was capable and committed, though the finance function had not yet been built to support group consolidations, intercompany eliminations, structured forecasting, or a disciplined, synchronized month-end close.

Before you can implement any system, you need numbers you can stand behind. So the first six weeks went into building that foundation by hand. I started by defining a single corporate chart of accounts to act as the group’s shared structure, then mapped each entity’s source ledger to it line by line. From there the work was deliberate and unglamorous. Payroll structures had to be aligned, intercompany activity identified and eliminated, and every balance tied back to the prior tax and statutory numbers so there was continuity a lender could trust. The result was a full group consolidation model in Excel, transparent enough that anyone could follow the logic from source ledger to consolidated result.

That model did two things at once. It gave immediate visibility into group performance, and it became the baseline for FP&A systems that would allow us to measure performance against the deal thesis. Just as important, it was a deep dive into the weeds, so was an excellent opportunity to understand each business unit and the people.

In parallel, we ran a structured vendor search for an FP&A platform, and I brought a recommendation to the Board. We selected Planful at the first quarterly board meeting on the strength of their implementation credibility and cost effectiveness for a group approaching 300 people. The Planful implementation team were genuinely excellent. They had clearly done this many times in acquisitive, mid-market environments, they understood the constraints of a lean finance team, and they shaped the stand-up around our capabilities rather than forcing us onto theirs. Because we had already proven the consolidation manually, the build was fast. We carried a model we trusted into a tool that automated it.

Within ten weeks we had professional and repeatable reporting in place. Group consolidation, budgeting, forecasting, and management reporting were fully operational and owned by a finance team of four.

Just as importantly, we established cadence. The business had previously relied on ad hoc reporting and an annual, tax-driven close. Within four months we implemented a monthly close by Day 25, improving to Day 15 after the first full quarter-end cycle. That was the inflection point. For the first time the business could measure performance against its deal thesis with consistency and confidence.

Covenant reporting was delivered in full and on schedule, and the same framework supported the integration of a subsequent acquisition without loss of visibility or control. The model proved both effective and repeatable. By prioritizing speed to financial clarity over business unit systems replacement, we deferred an MES upgrade for more than 18 months, including through another acquisition, without compromising reporting quality. Capital and management focus stayed aligned to value creation.

What gets underwritten is how quickly you can prove the business is tracking to the investment thesis. The systems and how you do that are incidental, but doing it well and quickly matters a lot.

In Part 2: how this approach evolves with AI, and what actually changes.

It’s a strange thing, a dimensionless number that determines the strength of the electromagnetic force. Change it even slightly and you change the size of atoms, the chemistry of matter, and ultimately the shape of the universe itself.

Finance borrowed the same symbol, α, for something different: the part of return that isn’t simply explained by market exposure. Alpha is the edge, the return created by process, judgment, information, and design rather than just riding the tide.

I liked the overlap.

Small numbers often determine large structures. Tiny shifts in assumptions compound into very different outcomes. In physics, understanding the constants matters because the entire system rests on them. In investing and consulting, the same idea applies: understand the parameters that actually matter, build around reality rather than stories, and avoid designing systems that fail under stress.

That’s the thinking behind one137th.