A concise editorial brief on the data science bubble: the excel-work reality, with the trade-offs and questions that matter before your next move.
The notebook promised more than the job
A graduate imagines experiments and models, then spends the first week reconciling columns from three exports. The disappointment is real, but the task may reveal the organisation’s actual data maturity. The important question is whether someone owns the definition, quality and decision the spreadsheet supports, or whether the graduate is being used as a human patch. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
A spreadsheet can be a system boundary
A spreadsheet can carry a business rule that no database or dashboard has made explicit. Cleaning it may expose duplicate customers, changing definitions or an unowned process. That is valuable work when the team learns from it and improves the system. It is a dead end when every month repeats the same manual correction with no permission to fix the source. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
The first request from finance
The first request from finance may be a table needed for a close, not a model. Ask what decision follows, how the number is checked, and who can change the definition. Data professionals earn trust by knowing what a number means before optimising how elegantly it is calculated. A reliable answer can be more consequential than an elaborate algorithm. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
Why titles expand faster than work
Companies use data-science titles for experimentation, analytics, forecasting, machine learning, reporting and client delivery. The title alone cannot tell a candidate which work is present. Read verbs in the description and ask for a recent project from question to use. A role that cannot explain its users or decisions may be a reporting queue with fashionable branding. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
The analyst who finds the missing definition
Finding that two teams define “active user” differently is not glamorous, but it can prevent a bad decision. The analyst needs curiosity, diplomacy and enough technical understanding to trace the source. This is a career asset when it leads to shared definitions, documented lineage or a better product question. It is exhausting when leadership wants a number without accepting its caveats. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
Modelling before measurement
A model cannot rescue a target that is not measured consistently. Before feature engineering, inspect missingness, labels, time windows, leakage and the decision threshold. Explain what the data does not contain. A junior who learns this discipline is building real capability, even if the first portfolio screenshot is less impressive than a synthetic prediction demo. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
The counterargument for operational work
Operational work deserves respect. Businesses need reconciliations, controls, reporting and careful investigation. The counterargument is that not every spreadsheet role should be sold as a path to modelling. Candidates can accept a reporting-heavy start if there is a named transition, mentorship and access to the underlying systems. Otherwise, price it as analytics operations rather than imagined research. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
When dashboards become decoration
A dashboard with many charts may still fail if nobody knows which action it changes. Ask who reviews it, how often, and what happened after the last insight. Visual polish cannot compensate for stale definitions or absent ownership. The useful dashboard makes a decision easier while preserving enough context to prevent false confidence. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
The politics of a clean number
A clean number can be politically convenient and analytically wrong. Stakeholders may prefer a single answer even when the range, denominator or data gap matters. A data scientist’s professional value includes saying what is known, what is inferred and what would change the conclusion. This can be uncomfortable, especially for a junior, so escalation and review culture matter. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
A junior’s real learning curve
A junior’s learning curve often begins with SQL, files, access requests, documentation and stakeholder questions before sophisticated models. That does not make the role fraudulent. It becomes a problem when the person cannot observe or attempt the next layer after proving reliability. Ask how juniors move from preparing data to interpreting, testing and owning a decision. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
What production changes
Production adds monitoring, latency, cost, permissions, versioning, rollback and human use. A notebook can demonstrate reasoning, but it cannot prove operational readiness. Ask whether the role ships models, supports analysts, or hands work to another team. The answer should shape the skills and evidence you build rather than the title you repeat. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
The portfolio problem
A portfolio that shows a Kaggle-style score may not show a business decision, data limitation or failure mode. A stronger case study can use a small public dataset and explain the baseline, evaluation, trade-off and non-use decision. Do not imply access to private company outcomes. Good evidence is reproducible enough to inspect and modest enough to trust. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
Interviewing the data role
Interviewers should be asked for the last request the team handled, the data owner, the consumer, the review process and the ratio of recurring reporting to exploratory work. Also ask what a successful first quarter produces. Vague answers are not proof of a bad role, but they increase uncertainty. A candidate can request a conversation with a future peer where appropriate. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
A six-month evidence test
At six months, evidence might include a definition improved, a manual step removed, an analysis that changed a choice, or a model whose limits are understood. There is no universal count or timeline. Compare promised work with observed access and authority. If learning is real but the title is inflated, the experience may still be useful; record it honestly. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
When to move closer to engineering
Moving closer to data engineering, product analytics, research, risk or operations can be a good choice when it matches the problems you enjoy. Do not treat engineering as the only escape from spreadsheets. The better move is toward a durable decision surface, stronger data ownership and work you can explain without exaggerating its sophistication. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.
A broader definition of data science
Data science is not defined by the glamour of a model. It is the practice of using evidence, uncertainty and computation to improve a decision, while understanding the system that produces the evidence. Some days contain Excel. The career question is whether you are learning to make that system more truthful and useful, or merely making recurring ambiguity faster. The editorial sources provide broad context about work and occupational change; they do not say that every data-science title contains advanced modelling or that spreadsheet work is inferior. Indian teams often need careful data cleaning, reconciliation, reporting, experimentation support and stakeholder translation. The right question is whether the work builds transferable judgement and whether the role’s description matches the actual operating need. Do not invent model accuracy, business impact or market salary evidence. Request access boundaries, tooling, review expectations and examples of recent work before accepting a label. The practical test is to write down what would change your mind. Ask for the relevant document, example, owner, or decision date; do not substitute a confident anecdote for evidence. A move can be attractive and still be a poor fit for your cash obligations, energy, location, or learning needs. Conversely, a less glamorous option may be the sounder experiment if it preserves runway and gives you a way to observe the promised work before making an irreversible commitment. Keep a short record of what was promised, what actually happened, and which assumption remains untested. That record is useful in a conversation, a review, or a later search, but it is not proof of a market-wide rule.