Usually both, in order: automate the rule-shaped volume that is burning your team's hours, then hire for judgement, relationships and growth. Field studies show AI lifts output on routine work and hurts it outside its strengths, so where you draw that line matters more than the tools.
For most growing Indian SMBs the answer is both, in a specific order: first automate the rule-shaped, high-volume work that is eating your team's hours, then hire for the work that needs judgement, relationships and accountability. Done in that order, the person you hire next is often a different and more valuable one than the person you were about to hire.
The instinct when a team is drowning is another pair of hands. But list what the drowning consists of and a pattern usually appears: copying data between systems, first replies to enquiries, assembling the same documents, chasing the same follow-ups. That slice is where automation works best, and India's new labour codes make the true cost of a hire worth recalculating.
- more issues resolved per hour when 5,172 support agents got an AI assistant
- 15%
- more issues resolved per hour when 5,172 support agents got an AI assistant
- gain for less skilled and less experienced agents; the most experienced gained little
- 30%
- gain for less skilled and less experienced agents; the most experienced gained little
- worse results when consultants used GPT-4 on a problem outside its strengths
- 23%
- worse results when consultants used GPT-4 on a problem outside its strengths
- of Indian leaders intend to use AI agents to extend workforce capabilities within 12 to 18 months
- 93%
- of Indian leaders intend to use AI agents to extend workforce capabilities within 12 to 18 months
01
Start with where the hours actually go
Before you write a job description, run a two-week time audit. Ask everyone on the stretched team to note each recurring task, how often it happens and how long it takes. You are looking for work that is frequent, follows the same steps every time and has an output you can check.
Then sort each task into one of three shapes. The examples below are hypothetical, from a small trading business.
| Shape of work | Examples | Best answer |
|---|---|---|
| Rule-shaped and frequent | Typing WhatsApp orders into the accounting system, dispatch updates, payment reminders | Automate |
| Routine but needs a check | Drafting quotes from a price list, first replies to enquiries, monthly reports | AI drafts, a person approves |
| Judgement and trust | Negotiating with key accounts, handling complaints, deciding credit terms | Hire, or keep with people |
02
What the evidence says AI is good and bad at
Some of the clearest field evidence comes from customer support. A study published in the Quarterly Journal of Economics in 2025 followed 5,172 agents at a Fortune 500 firm selling business-process software to small and medium-sized businesses. With an AI assistant, issues resolved per hour rose 15% on average and 30% for less skilled and less experienced agents. Agents with two months' tenure and the tool performed as well as agents with more than six months and no tool, while the most experienced saw small gains in speed and small declines in quality.
The International Labour Organization's 2025 index finds that one in four workers worldwide is in an occupation with some generative AI exposure, with clerical work the most exposed. Because most occupations consist of tasks that need human input, it concludes that transformation of jobs is the most likely impact.
The limits are just as clear. In an experiment with 758 junior BCG consultants, around 90% improved their performance on creative product-innovation work with GPT-4. On a business problem-solving task designed so that GPT-4's first answer would be wrong, those using it did 23% worse than those working without it, and many accepted its erroneous output at face value.
It's likely that GPT-4's ability to generate persuasive content contributed to this result.
03
What a hire really costs in India now
Salary is only the start. If PF and ESI apply to your business, the employer contributes 12% of basic wages and dearness allowance, split between the employee's provident fund and pension, and 3.25% of wages to ESI for employees earning up to ₹21,000 a month. Then come leave, gratuity, recruitment, a laptop and phone, and the weeks a new person needs before they are fully productive.
The four labour codes that took effect on 21 November 2025, consolidating 29 older laws, add more to plan for. Appointment letters are now mandatory for all workers, and fixed-term employees qualify for gratuity after one year of continuous service instead of five. The detail depends on your state, headcount and pay structure, so check it with your accountant before you budget a role.
04
Run the arithmetic both ways
Take a hypothetical Kochi trading firm. Two coordinators spend a combined 70 hours a month typing WhatsApp orders into the accounting system and sending dispatch updates, and the owner is about to hire a third at ₹20,000 a month, of which ₹10,000 is basic wages. Salary alone is ₹2,40,000 a year. If PF and ESI apply, the employer's PF share adds ₹1,200 a month and ESI ₹650, about ₹2,62,200 a year before leave, gratuity, recruitment and training.
Now cost the alternative: a flow that captures those orders and sends the updates automatically. Say it costs ₹50,000 to build and ₹3,000 a month to run, and returns 50 of those 70 hours. Year one comes to ₹86,000. If those hours mean the third hire is not needed, the build cost is recovered in under three months. Your numbers will differ; the method is the point.
Our rule of thumb: if an automation pays for itself within a quarter, build it before you interview anyone. Then look again at the hire. The firm above may still need a third person, but probably someone who can sell or chase overdue payments, not another pair of hands for data entry.
05
When hiring is the right answer
In Microsoft's Work Trend Index findings for India, published in August 2025, 93% of leaders said they intend to use AI agents to extend workforce capabilities in the next 12 to 18 months, and 92% said their company is considering adding AI-specific roles. Automation changes what you hire for; it does not end the need to hire. The signs that the next step should be a person are usually clear:
- The bottleneck is judgement or trust: key accounts, sales, site supervision, handling exceptions.
- Growth is sustained, not a seasonal spike a temporary arrangement could absorb.
- The work changes every time, so there is no stable process to automate.
- An automation would need a full-time minder anyway, in which case you are hiring regardless.
- Customers expect a person, and losing that relationship would cost more than the time saved.
06
Bring the team with you
Automation goes wrong when staff see it as a stealth layoff. Say plainly what it is for, removing drudgery so people can do work the business needs more of, and involve the person who does the task today in designing the replacement. They know the exceptions. When you do hire, give the new person the tools from day one: in the support study, the least experienced agents gained the most, and newer agents became less likely to leave.
Run the automated flow in parallel with the manual one for a couple of weeks, name one person who owns it, and document how it works so it survives staff changes. Our guide to the no-code automation stack covers the build order, and a CRM your team actually uses is often where the returned hours show up first.
What to do before your next hire
Pause the job advert for a fortnight unless the role is clearly judgement-heavy, then use the next 90 days like this.
Weeks 1 to 2
Time audit
The stretched team logs recurring tasks, how often they happen and how long they take.
Weeks 3 to 4
Sort and cost
Sort tasks into automate, AI drafts with a person approving, or keep with people. Cost the top candidate both ways.
Weeks 5 to 8
Build and run in parallel
Automate one workflow end to end, keep the manual process running alongside, and fix what breaks.
Weeks 9 to 12
Measure, then decide the hire
Count the hours returned, then rewrite the job description around the work only a person can do.
- Start the audit this week with the people who feel most stretched.
- Ask your accountant for the full yearly cost of the role you were about to advertise.
- Decide in writing what the next hire will do that automation cannot.
For help costing the first workflow, our AI and automation team offers a free audit.
Sources
- Quarterly Journal of Economics: Generative AI at Work (2025)
- BCG: How people create and destroy value with GenAI (2023)
- ILO: Generative AI and jobs, refined global index (2025)
- Microsoft: Work Trend Index 2025, India findings
- PIB: Government makes the four Labour Codes effective (2025)
- EPFO: EPF scheme contributions
- ESIC: contribution rates
- ESIC: coverage and wage limit
