Margin or Growth: Where Should AI Productivity Gains Go First?

By Victor Hoang · Updated Oct 10, 2026

Sooner or later every leadership team hits the same moment. A team adopts AI, hours get freed up, and someone asks the obvious question: so what do we do with that time? The default answer is to take it out of the budget. It's measurable, the board understands it, and it shows up in next quarter's numbers.

Our view is that the default is often the weaker move. Cost cuts are real and sometimes right. But companies that redeploy freed capacity into growth work tend to compound, because the gains fund the next round of capability instead of ending at the P&L line. That only works if you decide the split deliberately, before the capacity appears. Otherwise it quietly evaporates into slack, and nobody gets either outcome.

Key Facts

• In a study of 5,179 customer support agents, access to an AI assistant raised productivity by 14% on average, and by 34% for novice and low-skilled workers.
• In Danish labor data, researchers found precise null effects on earnings and recorded hours from chatbot adoption, ruling out effects larger than 2% two years after ChatGPT's launch. Employers reorganized tasks around AI instead.
• BCG's 2025 study found that future-built companies, about 5% of those surveyed, show 1.7x the revenue growth and 1.6x the EBIT margin of laggards. Per the full report, they reinvest the proceeds of early AI success in new capabilities, tools and innovations.
• In IBM's survey of 2,000 CEOs, only 25% of AI initiatives had delivered expected ROI and only 16% had scaled enterprise-wide.
• Klarna's CEO told Bloomberg that AI customer service was cheaper than human staff but lower in quality, and the company began hiring human agents again.

Why Cost Cutting Is the Default

Margin is the easy choice for good reasons. Savings are countable, so a CFO can defend them. They're also controllable: you can cut a contractor budget or freeze a requisition this month. Growth is slower, riskier and depends on customers, not just on your own decisions.

There's a deeper reason too. The planning cycle rewards whatever can be written into a budget line. If the business case for an AI tool said "saves 6 hours per person per week," the natural follow-up is to ask which cost line shrinks. We looked at how that measurement problem plays out in how CFOs should measure AI investment returns.

None of this makes cost reduction wrong. If you're in a price-compressed market, carrying real overcapacity, or short of runway, banking savings first is sound. The argument here is narrower: as a reflex, it often leaves value on the table.

What the Evidence Says About Where Gains Actually Land

Three findings are worth holding in mind before you plan.

Freed time is not the same as freed cost. Take the best-known field study, in which 5,179 support agents resolved 14% more issues per hour with an AI assistant. That's capacity. Whether it becomes a smaller team, more tickets handled, better service or a faster ramp for new hires is a management decision the study doesn't make for you. Notably, the biggest gains went to novices (34%), which points at onboarding speed as one use of the gain, not only headcount.

Gains often don't show up in the aggregates by themselves. The Danish study of chatbot adoption found no measurable change in earnings or recorded hours, even though workers reported benefits. The authors describe employers reorganizing tasks, adding work such as content generation and AI oversight. Put plainly: when nobody directs the freed capacity, it gets absorbed by new tasks and by slack, and the P&L never sees it.

The companies pulling ahead reinvest. BCG's 2025 study of 1,250 firms describes a small group of "future-built" companies with 1.7x the revenue growth and 1.6x the EBIT margin of the 60% it calls laggards. In the full report, BCG says these companies are pulling further ahead as they reinvest the proceeds of earlier success in new capabilities, tools and innovations. It also says their targets go beyond efficiency. This is a correlation from a survey, not proof that reinvestment causes the lead, and leaders may simply be better-run firms. But it matches the compounding logic: gains that fund the next capability beat gains that end in a lower cost line.

One more data point cuts the other way, and it's a useful warning. Klarna pushed AI into customer service, then its CEO told Bloomberg that the AI service was cheaper but lower in quality. Coverage of the reversal says the company is hiring human support staff again. Cutting to the savings number, without a view on what the capacity was protecting, can reverse itself.

The Case for Growth First, and Its Limits

Why would redeployment tend to compound? Three mechanisms.

1. Revenue scales without a matching cost line. A rep who spends less time on admin and more time with buyers adds pipeline. The freed hours are the same either way, but only one use grows the top line.
2. Capability builds on itself. Teams that use saved time to build the next workflow, the next dataset or the next product feature get a second round of gains. A cost cut gives you one round.
3. Morale and adoption. People adopt tools faster when the gain means better work, not a smaller team. We covered the fear side of this in how leaders respond to AI job anxiety.

The limits are just as real. Growth work can fail. Redeployed people need somewhere useful to go, and not every function has demand to absorb new output. A support team that resolves 14% more tickets per hour doesn't help if ticket volume is flat and you have no better use for the time. Where there's no growth outlet, the honest choice is margin, or attrition without backfill.

A Decision Table: Margin, Growth, or Both

Rather than argue it in the abstract, sort each function by two questions: is there unmet demand for this team's output, and is the market punishing your cost structure?

The split numbers are illustrative starting points, not benchmarks. What matters is that someone wrote down a number before the tool went live.

How to Set the Split Deliberately

The failure mode isn't picking the wrong answer. It's never picking, so the capacity gets absorbed by meetings, rework and new tasks. A few moves fix that.

Decide per function, before rollout. Set the target split in the business case: how many hours or dollars of the expected gain go to cost, how many go to a named growth project. Our executive decision framework for AI workforce strategy is a good companion for structuring that choice.

Name the growth work. "Spend the time on higher-value work" is not a plan. Write down the accounts, the experiments or the roadmap items the capacity is meant to fund, and who owns them.

Track both sides. Measure the hours freed and where they went. If the freed time isn't visible in either a lower cost line or a higher output line within two quarters, the gain has leaked.

Protect the people who make redeployment work. Reallocating people is only possible if they have the skills for the new work. The cost of delay is covered in the hidden cost of delaying AI upskilling, and the role shifts are mapped in which roles AI eliminates and creates.

Tell the board the split. Directors who hear "we expect X in savings" will ask where it is next quarter. Directors who hear "60% to savings, 40% to this named growth bet, reviewed in two quarters" can hold you to a real plan. The framing in talking to your board about AI workforce investment applies directly.

What to Do Next

1. List your three biggest AI deployments and write down the hours or cost each is expected to free. If you can't, that's the first finding.
2. Run the decision table on each function and set a target split with an owner.
3. Name one growth project per function that the freed capacity will fund, and set a review date two quarters out.
4. Set a floor for margin so growth spending can't quietly eat every saving. A rule such as "at least a third of measured gains reach the P&L" keeps finance on board.
5. Revisit the split every half year. Markets change, and so does the right answer. A company under pressure now may be ready to reinvest in a year.

For smaller teams, the same logic shows up as leverage: see how small teams generate big revenue with AI for how the growth side looks when headcount stays flat.

Learn More

• How CFOs Should Measure AI Investment Returns
• The Executive Decision Framework for AI Workforce Strategy
• The Hidden Cost of Delaying AI Upskilling
• How to Talk to Your Board About AI Workforce Investment
• AI Roles: What's Eliminated and What's Created
• Upskill vs. Hire: The AI-Native ROI Case