Picture a founder running an online shoe brand doing about ₹50 lakh a month. There is a storefront, performance marketing, design and content, a developer, someone on CRM and retention, two people on customer service, and merchandising decisions the founder still makes personally.
Over the last year the tools changed. Much of what that team does by hand can now be assisted by software: campaign variants, product copy, image sets, support replies, reporting, routine automation.
So the founder has a real decision with money on both sides. Use AI to run the same business with fewer people, or keep the people and run a bigger business.
Most commentary treats this as a values question. It is an arithmetic question first, and the arithmetic flatters neither side.
What actually changed for a small team
The honest summary is that execution got cheaper and judgement did not.
A marketer can test far more creative in a week. A designer can generate product imagery that used to need a shoot. A developer can ship a checkout change in an afternoon. Reporting that took a day of spreadsheet work assembles continuously.
What has not changed is knowing which variant is worth testing, which checkout change is worth shipping, and which support reply should never be sent automatically.
Output volume is easy to mistake for progress. Forty ad variants is not an achievement if the brand only had budget to test six properly.
The two paths, with the numbers on the table
Take the shoe brand at ₹50 lakh a month, a team of seven at a blended fully loaded cost of ₹65,000 each, so ₹4.55 lakh of monthly payroll, and a contribution margin of 35 percent. These are illustrative assumptions, chosen to be plausible rather than precise, and you should substitute your own.
The savings path. The founder consolidates three roles and runs the business with four people plus AI tooling. Payroll falls by ₹1.95 lakh a month. Tooling and implementation cost something. Revenue is assumed flat.
The growth path. The founder keeps all seven, adds the same tooling, and redirects the freed capacity into acquisition, conversion and retention work the team never had time for.
| Savings path | Growth path | |
|---|---|---|
| Monthly payroll | ₹2.60 lakh | ₹4.55 lakh |
| AI tooling and implementation, monthly | ₹40,000 | ₹55,000 |
| Review and correction time | Higher, concentrated on fewer people | Spread across specialists |
| Monthly cost change | About ₹1.55 lakh saved | About ₹55,000 added |
| Revenue assumption | Flat | Must increase to justify itself |
| Certainty | High | Low |
Now the number that actually decides it. The savings path runs at about ₹3 lakh a month all in. The growth path runs at about ₹5.1 lakh. So the growth path has to generate ₹2.1 lakh a month of additional contribution just to draw level. At a 35 percent contribution margin that is about ₹6 lakh a month of additional revenue, or 12 percent growth.
That is the honest bar: a specific percentage the growth path either clears or does not.
The saving is real, immediate and certain, and dismissing it ignores the founder's actual risk. It is also capped and happens once, while a revenue improvement compounds into every month after.
Who actually gets the productivity gain
Here the research is more useful than the opinion, and it complicates the case for cutting experienced staff specifically.
In a study of 5,179 customer support agents, Brynjolfsson, Li and Raymond measured a 14 percent average increase in issues resolved per hour after a generative AI assistant was introduced. The distribution is the interesting part. Novice and lower skilled agents improved by about 34 percent. The most experienced agents improved very little.
From a staffing point of view, AI raised the floor substantially and the ceiling barely, narrowing the gap between a new hire and a veteran on well defined work. The implication that gets missed is the second one: if AI closes that gap on routine tasks, the remaining value of the experienced person sits in the work that is not routine, which is exactly what a savings-focused reorganisation tends to leave unstaffed.
The second study makes the point sharper. Dell'Acqua, McFowland, Mollick, Lifshitz-Assaf and Kellogg, in a 2023 Harvard Business School and BCG working paper titled Navigating the Jagged Technological Frontier, ran a field experiment with 758 consultants. On tasks within the AI's capability, the AI-assisted group completed 12.2 percent more tasks, 25.1 percent faster, with quality rated more than 40 percent higher. On a task deliberately chosen to sit outside that capability, the AI-assisted group was 19 percentage points less likely to reach the correct answer than the group working without it.
The failure mode is confident output that reads well and is wrong, accepted by someone without the background to catch it. A junior marketer will not query a plausible-sounding attribution analysis. An experienced one will.
The cost that never reaches the spreadsheet
Checking AI output is work, and it is usually unbudgeted.
Research by Stanford and BetterUp Labs, published in Harvard Business Review, surveyed 1,150 US desk workers. Around 40 percent said they had received low quality AI generated work in the previous month, and dealing with each instance took an average of one hour and 56 minutes. The researchers estimated the cost at about 186 dollars per employee per month, though that figure is built from self reported time and salary, so treat it as an indication rather than a measurement.
The time figure is the one worth carrying across. Two hours per incident, landing on whoever has to fix it. In a consolidated team that person is usually the founder.
This connects to the most sobering finding in the area. MIT's NANDA initiative reported in 2025 that roughly 95 percent of organisations saw no measurable profit and loss impact from their generative AI investments. That report is preliminary rather than peer reviewed, its success bar is subjective, and published descriptions of its sample are inconsistent, so it is not settled evidence. It remains the clearest signal that individual productivity gains have been easier to achieve than company level profit gains, and both paths above have to survive it.
What leaves when the specialist leaves
Suppose the shoe brand takes the savings path and the senior performance marketer is one of the roles consolidated.
The tools remain. What goes is the knowledge that a particular cohort buys twice in ninety days, that a discount above a certain level attracts returns, and why the brand stopped selling a profitable looking SKU that generated complaints. None of that is in the ad account, and none of it is written down.
Workforce reduction attributed to AI is real and worth sizing correctly. Challenger, Gray and Christmas recorded 116,175 US job cut announcements citing AI through August 2026, out of 529,914 announced cuts in total, so roughly a fifth. Significant, and a minority of the total, which is a useful corrective to both the panic and the dismissal.
If you do consolidate, the mitigations are unglamorous and they work: write down the decision history before the person leaves, name one accountable owner per function even when one person owns several, and keep the reporting in shared systems rather than in someone's spreadsheet. The related failure, where nobody owns a number at all, is covered in why your ecommerce numbers never match.
Where growth would actually come from
If the growth path has to clear 12 percent, it is fair to ask where that comes from. Vague claims about efficiency do not get there. Specific lifts might.
| Area | What AI assists | What the human owns | What you measure |
|---|---|---|---|
| Acquisition | Creative variants, audience research, competitor scanning | Which tests are worth budget | Cost per acquisition |
| Conversion | Product copy, image sets, page variants | Merchandising and pricing judgement | Conversion rate |
| Retention | Segmentation, journey drafts, personalisation | Offer economics and brand voice | Repeat purchase rate |
| Merchandising | Demand patterns, availability analysis | What to stock and discount | Contribution margin |
| Operations | Exception flagging, routine automation | Policy and escalation decisions | Order accuracy, cost per order |
A one point conversion improvement, a three point lift in repeat purchase rate and a modest reduction in acquisition cost together clear 12 percent comfortably. Any one of them alone probably does not.
That is the real test of the growth path. It needs several improvements landing across the funnel, which needs people with enough context to find them. Deciding which of those to automate first has its own method, covered in what you should automate first, and the question of how much autonomy to grant is in what to automate and what stays human.
A ninety day way to decide
Reorganising first and learning afterwards is the expensive order. Reverse it.
Month one, measure. Pick three or four activities that consume real time and record how long they take and what they produce. Most founders are guessing at this baseline.
Month two, introduce AI into those activities with the existing team and track two things: how much faster the work gets, and how often output needs correcting. That correction rate predicts whether a smaller team would cope.
Month three, check whether capacity converted into anything commercial. More output is not the test. A moved metric is.
Only then decide the structure, with evidence instead of a forecast. A useful reference point for what the tooling and implementation side realistically costs is in what AI automation costs in India.
Where this leaves the decision
The savings path is defensible when a function genuinely has surplus capacity, when the business needs cash certainty, or when the work being consolidated is routine enough that the research above predicts AI will cover it well.
The growth path is defensible when the business has identifiable demand it is not capturing, and people with enough domain knowledge to turn extra capacity into revenue rather than extra output.
What the evidence does not support is treating the two as interchangeable. One produces a known, one time, capped improvement. The other produces an uncertain improvement that compounds, and requires the capability to execute it. Those are different financial instruments, and a founder should price them differently.
Run the break-even number for your own business before the structure changes. If the growth you would need is plausible with the team you have, that is the stronger investment. If it is not, the saving is the honest answer, and it should be taken deliberately rather than described as a transformation.




