The Third Wave of the Spreadsheet

The Spreadsheet That Moves Work

Work does not move when technology changes. It moves when a cell in a cost model changes.

Every large migration of service work in the last thirty years has been driven by the same artifact: a cost-per-contact model with one variable highlighted. In the early 2000s that variable was wages. An English-speaking graduate in Bangalore or Gurgaon could handle the same call as an agent in Ohio or Newcastle, and the fully loaded cost difference was large enough that the rest of the slide barely mattered. Entire American and British mid-sized cities lost their largest employers over about a decade, and India built a services export engine on the other side of the same decade.

The scale of what got built is easy to forget. India's business process management industry reached roughly 54.6 billion dollars in export revenue in the 2024 to 2025 financial year by NASSCOM's estimate, employing on the order of 1.6 million people directly, with the wider technology sector employing several million more. Whole career ladders, whole suburbs, whole marriage-ad categories exist because of that spreadsheet.

I stood inside wave one. The company I later worked for, [24]7.ai, exists because of it. I ran floors at Freecharge where my teams handled 100,000+ tickets a month, which is what wave one looked like at full maturity: industrial, measured, proud of its handle times. If you want to understand what happens to Bangalore in wave three, you have to first be honest about what actually won it wave one. It was never only the price.

What Actually Won the First Wave

India did not win because it was cheap. Cheap got the pilot. Something else got the decade.

Plenty of places were cheap in 2003. What India built that its price competitors did not was the ability to industrialize trust: to take something as unstandardizable as a human conversation and wrap it in quality systems, escalation matrices, calibration sessions, accent training, workforce management, and six-sigma rigor until a bank in Charlotte would hand its most sensitive customer moments to a building it had never visited, full of people it had never met, and sleep fine. That is an astonishing organizational achievement, and it is routinely written out of the story by people who only ever saw the invoice.

I have lived the late, refined version of it. At Freecharge we took first response time from 8 hours to under 2 across a 100,000-ticket monthly volume, not by adding bodies but by rebuilding routing, macros, staffing curves, and escalation logic. Twice in my career I have cut a customer operation's cost roughly in half, and both times the saving came from systems redesign, not from making the queue walk faster. The skill underneath both stories is the same one that won wave one: turning service into an engineered system.

Hold that distinction, because everything in the rest of this piece depends on it. The queue, the repeatable eighty percent of contacts, was always the commodity layer, and commodities always move to the cheapest substrate eventually. The system-building skill wrapped around the queue was the differentiated layer. The tragedy of the current panic is that the industry is defending the commodity and neglecting the differentiator.

The Third Wave, by the Numbers

The polite version is in every industry report. The impolite version is on every entry-level hiring dashboard.

The public numbers first, with their sources attached, because this topic drowns in unattributed doom. NASSCOM's own leadership has called routine BPO roles the category at highest displacement risk from AI. PwC has estimated that AI can handle around 80 percent of routine customer queries, and tier-one support is mostly routine queries by definition. Industry analysts project on the order of a million outsourcing roles across India and the Philippines impacted by 2030; treat any such forecast as an estimate, but notice that nobody serious is forecasting zero. And the number that should end all complacency: India's top IT services firms added seventeen net employees across the first nine months of fiscal 2026, as reported across the trade press, in an industry that spent two decades adding thousands every quarter. The total job count still grew modestly per NASSCOM, which tells you composition is shifting, not collapsing. The bottom rung is where the fire is.

Klarna ran the public experiment for everyone. In February 2024 it announced its assistant was doing the work of 700 agents. By May 2025 its CEO was telling Bloomberg that the cost focus had gone too far, and the company was hiring humans back into a flexible support pool. Most commentary read that as a parable about empathy. I read it as a parable about instrumentation: they did not miss the people, they missed what the people knew. But notice what Klarna did not do. It did not rebuild the 700-seat floor. The work that came back was not the work that left, and that asymmetry, not the reversal itself, is the actual lesson.

What Ohio Can Teach Bangalore

The work that comes back is never the work that left.

Here is the part of wave one that India rarely studies, because winners study their victories and not their opponent's recovery. When service work left the American Midwest, the towns that waited for the call centers to come back waited forever. Nobody reopened the 2,000-seat floor. What eventually came back, where anything came back at all, was smaller, stranger, and more valuable per head: escalation desks for the angriest two percent, premium support sold as a product, specialist teams sitting next to engineering, roles that existed precisely because the offshore floor could not do them. The displaced agents who recovered fastest were not the ones who found another queue. They were the ones who moved adjacent: into the exceptions, into the relationships, into the systems that managed the very work that had left.

Run that forward. When the repeatable eighty percent of contacts moves to models, what remains on the human side of the line is not a smaller version of the old queue. It is a different occupation wearing the old department's name. The twenty percent the model hands back is by construction the hardest, angriest, most contractual, most ambiguous work in the company, and it arrives with the customer already annoyed that a bot tried first. I wrote about that failure mode separately as the Handoff Cliff. Staffing that work with yesterday's tier-one playbook, tier-one training, and tier-one pay is the single most common mistake I see in AI-era operations planning, and it is the exact mistake the Midwest made in reverse when it assumed returning work would look like departed work.

The Money, the Mess, or the Machine

Three directions out of the queue. Every surviving role I am staffing sits on one of them.

Strip the euphemisms out of every AI-era workforce deck I have seen or written, and the roles that survive sort into three families. I use this as a career compass with my own teams, so it is written in the second person on purpose.

The money. Move toward revenue ownership: renewals, expansion, adoption that leads to expansion, the commercial outcomes of the account. AI deflects questions; it does not own a number, and no CFO alive will let it own one soon. A person who can stand in front of a renewal and be accountable for it has left the queue permanently. This is the classic support-to-success jump, and it is still the widest door, though it now demands commercial fluency and not just empathy. My hiring playbook for these roles is published here.

The mess. Move toward the exceptions: the escalations, the regulated conversations, the data-loss incidents, the contractually sensitive moments, the twenty percent the model hands back. This work gets harder and more valuable as automation grows, because everything easy has been strained out of it. The mess is also where judgment gets witnessed, and witnessed judgment is what gets people promoted. The error most companies make is treating exception handling as tier one with extra steps. It is closer to incident command, and should be hired, trained, and paid like it.

The machine. Move toward the system itself: the person who reads what ten thousand customers are telling the AI every day, finds where it fails, redesigns the flows, owns the escalation logic, writes the playbooks the agents and the models both run on. Every deployment I have led needed more of these people than we had. In my org designs this person is worth several of yesterday's team leads, because they multiply everyone else, human and machine alike. This is the direction most open to queue veterans, because nobody knows the work's failure modes like the person who handled it for five years.

The compass fits in one line: move toward the money, the mess, or the machine, because those are the three things the spreadsheet cannot put in a cell.

For Leaders: Staffing the Post-Queue Organization

If you run one of these organizations, the kindest thing you can do is stop pretending, and the smartest is to re-invest where the spreadsheet cannot see.

Both times I cut an operation's cost in half, the headline number came from systems, and the lasting value came from what we did with the people the systems freed. The honest sequence that has worked for me, laid out at length in Building the Autonomous Customer Organization, compresses to this:

None of this is charity. The twenty percent queue decides your renewals, the transcript stream decides your roadmap, and the machine roles decide whether the automation compounds or decays. Underfund the three of them and the spreadsheet's savings quietly leak back out through churn, which no one ever traces back to the slide that caused it.

The Rung That Burns

A personal note on ladders, from someone whose ladder started on a shop floor.

My career did not start in a SaaS tool. It started on Indian retail floors, at pharmacy counters and apparel stores, watching what makes a person come back to a shop and what quietly loses them forever. The service queue was my industry's bottom rung the way the shop floor was mine: not glamorous, but a place where someone with no network and no brand-name degree could enter, be measured honestly on effort, and climb. That is what the entry numbers this year are really telling us. Seventeen net hires in nine months is not a statistic about efficiency. It is a ladder with its bottom rung on fire, in the one industry that taught two generations of Indian families to believe in ladders.

So I will not write the comfort piece, because the queue is leaving and my own hands are on the work that moves it. And I refuse to write the panic piece, because I watch what actually happens inside AI-first operations every day, and the humans who move toward the money, the mess, or the machine are not surviving the shift. They are being promoted by it. India did not win the first wave by being cheap. It won by industrializing trust at a scale nobody believed possible. Trust still does not fit in a cell of the spreadsheet. The queue was never the asset. The people who could build a system you would trust with your angriest customer at midnight: they were the asset, and they still are.