{"course":{"slug":"ai-careers","title":"Working in AI","url":"https://addaly.com/learn/ai-careers","level":"beginner","track":"work"},"lesson":{"slug":"inside-a-data-team","title":"Who sits around you, and how work arrives","url":"https://addaly.com/learn/ai-careers/inside-a-data-team","position":6,"of":85,"minutes":9,"answer":"The organisation's shape and data maturity determine your daily work more than the job title does, and both can be diagnosed with one question in the interview.","keyIdea":"The organisation's shape and data maturity determine your daily work more than the job title does, and both can be diagnosed with one question in the interview."},"body":{"format":"markdown","text":"## The cast\n\nJob adverts describe a role. They rarely describe the room. Most data and AI work happens inside a group that looks roughly like this:\n\n- **A business stakeholder** who owns a number (revenue, churn, fraud losses, delivery times) and does not care how it is computed.\n- **A product manager**, translating between that person and the team.\n- **An analyst**, answering questions with SQL and charts.\n- **An analytics engineer**, turning raw tables into clean, documented, tested models everyone else builds on.\n- **A data engineer**, moving data from source systems into the warehouse and keeping the pipelines alive.\n- **A machine learning or AI engineer**, when there is a model in production.\n- **A platform or infrastructure person**, often shared across several teams.\n- **A domain expert** (an underwriter, a clinician, a logistics planner) whose knowledge decides whether any of it is correct.\n\nSmall companies collapse several of these into one job, which is usually yours. That is not a warning. Doing four of them badly for a year teaches more than doing one of them well, and it is the fastest route to knowing which one you want.\n\n## Three shapes, three lives\n\nHow the group is arranged changes your day more than the tools do.\n\n**Centralised.** One data team serves the whole company through a request queue. You see breadth and every part of the business, and you develop shallow domain knowledge because you are never anywhere long. Requests arrive as tickets. The team's constant problem is that it is a service desk with a backlog.\n\n**Embedded.** Analysts and engineers sit inside product or business teams and report there. You get deep domain knowledge, real influence, and you are close enough to the decision to change it. The risks are isolation from other data people, no one senior to review your work, and career exposure if the product is cancelled.\n\n**Hub and spoke.** Embedded people with a dotted line to a central team that owns standards, tooling and career development. Most organisations end up here after trying the other two. It is the best of the three for someone early, because you get a domain and a reviewer.\n\nYou can identify the shape in one interview question: *\"If I need a number by Friday, who do I ask, and who checks it?\"*\n\n## How work actually arrives\n\nAlmost never as a specification. It arrives as:\n\n- A Slack message: \"can you pull the numbers for the board deck\".\n- A ticket with a title and no body.\n- A meeting where three people describe three different problems.\n- An incident: a dashboard is wrong and somebody noticed in a leadership meeting.\n\nThe first, unteachable-sounding, entirely teachable skill of the job is converting a request into a question with a decision attached. The move is one sentence: *\"What will you do differently depending on the answer?\"* If nobody can answer that, the work is decoration and can wait. If they can, you have just learned what precision the answer needs. A decision to spend fifty thousand rupees needs a rougher number than a decision to change pricing for every customer.\n\n## The maturity ladder, and why it decides your year\n\nData functions sit on a ladder, and where a company sits determines what you will actually do:\n\n1. **Spreadsheets.** Numbers live in files people email. Your job is extraction and reconciliation.\n2. **A database or warehouse exists** but nothing is documented. Your job is archaeology and writing the same joins repeatedly.\n3. **A transformation layer** (dbt or its equivalent) with tested, documented models. Your job becomes analysis rather than plumbing.\n4. **A metrics layer**, so \"active customer\" means one thing everywhere. Arguments about definitions stop.\n5. **Models in production**, with monitoring and retraining.\n\nAn advert for an AI engineer at a company on rung two is an advert for a data engineer who will spend the first year at rung three. That is not a bad job. It is a different one from the one advertised, and taking it without knowing is how people end a year unable to explain what they learned.\n\nDiagnose it with a single question in the interview: **\"Where does the number on your main dashboard come from, end to end?\"** A confident four-sentence answer means rung three or higher. Hesitation, or \"it depends who you ask\", means rung one or two. Both are worth joining. Only one of them matches the advert."},"check":{"question":"In an interview, a candidate asks where the number on the company's main revenue dashboard comes from. The hiring manager says three teams maintain slightly different versions and finance reconciles them monthly in a spreadsheet. What does this most reliably tell the candidate?","options":[{"text":"The company is disorganised and the role should be avoided.","correct":false,"tempting":"Reads a maturity level as a character judgement; most organisations are here, and it is where the highest-impact early work exists."},{"text":"The team lacks a metrics layer and probably a tested transformation layer, so the first year will be plumbing and definition work regardless of what the advert called the role.","correct":true},{"text":"The role will involve building machine learning models on top of finance data.","correct":false,"tempting":"Assumes advertised work happens; a company reconciling revenue by hand is several rungs below production modelling."},{"text":"The company needs better dashboard software and the candidate should propose a tool in the interview.","correct":false,"tempting":"Treats a definition and ownership problem as a tooling problem, which is the most common wrong diagnosis in this field."}]},"prev":{"slug":"reading-a-job-advert","title":"Reading a job advert for what it actually is","url":"https://addaly.com/learn/ai-careers/reading-a-job-advert"},"next":{"slug":"degrees-and-doors","title":"Who actually needs a degree","url":"https://addaly.com/learn/ai-careers/degrees-and-doors"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Who sits around you, and how work arrives', in Working in AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-careers/inside-a-data-team (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Who sits around you, and how work arrives","courseTitle":"Working in AI","module":{"index":1,"slug":"the-terrain","title":"The terrain: who hires, for what, and where you fit"},"answer":null,"keyIdea":"The organisation’s shape and data maturity determine your daily work more than the job title does, and both can be diagnosed with one question in the interview.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-cast","runs":[{"text":"The cast"}]},{"t":"p","runs":[{"text":"Job adverts describe a role. They rarely describe the room. Most data and AI work happens inside a group that looks roughly like this:"}]},{"t":"ul","items":[[{"text":"A business stakeholder","b":true},{"text":" who owns a number (revenue, churn, fraud losses, delivery times) and does not care how it is computed."}],[{"text":"A product manager","b":true},{"text":", translating between that person and the team."}],[{"text":"An analyst","b":true},{"text":", answering questions with SQL and charts."}],[{"text":"An analytics engineer","b":true},{"text":", turning raw tables into clean, documented, tested models everyone else builds on."}],[{"text":"A data engineer","b":true},{"text":", moving data from source systems into the warehouse and keeping the pipelines alive."}],[{"text":"A machine learning or AI engineer","b":true},{"text":", when there is a model in production."}],[{"text":"A platform or infrastructure person","b":true},{"text":", often shared across several teams."}],[{"text":"A domain expert","b":true},{"text":" (an underwriter, a clinician, a logistics planner) whose knowledge decides whether any of it is correct."}]]},{"t":"p","runs":[{"text":"Small companies collapse several of these into one job, which is usually yours. That is not a warning. Doing four of them badly for a year teaches more than doing one of them well, and it is the fastest route to knowing which one you want."}]},{"t":"h2","id":"three-shapes-three-lives","runs":[{"text":"Three shapes, three lives"}]},{"t":"p","runs":[{"text":"How the group is arranged changes your day more than the tools do."}]},{"t":"p","runs":[{"text":"Centralised.","b":true},{"text":" One data team serves the whole company through a request queue. You see breadth and every part of the business, and you develop shallow domain knowledge because you are never anywhere long. Requests arrive as tickets. The team’s constant problem is that it is a service desk with a backlog."}]},{"t":"p","runs":[{"text":"Embedded.","b":true},{"text":" Analysts and engineers sit inside product or business teams and report there. You get deep domain knowledge, real influence, and you are close enough to the decision to change it. The risks are isolation from other data people, no one senior to review your work, and career exposure if the product is cancelled."}]},{"t":"p","runs":[{"text":"Hub and spoke.","b":true},{"text":" Embedded people with a dotted line to a central team that owns standards, tooling and career development. Most organisations end up here after trying the other two. It is the best of the three for someone early, because you get a domain and a reviewer."}]},{"t":"p","runs":[{"text":"You can identify the shape in one interview question: "},{"text":"”If I need a number by Friday, who do I ask, and who checks it?”","i":true}]},{"t":"h2","id":"how-work-actually-arrives","runs":[{"text":"How work actually arrives"}]},{"t":"p","runs":[{"text":"Almost never as a specification. It arrives as:"}]},{"t":"ul","items":[[{"text":"A Slack message: “can you pull the numbers for the board deck”."}],[{"text":"A ticket with a title and no body."}],[{"text":"A meeting where three people describe three different problems."}],[{"text":"An incident: a dashboard is wrong and somebody noticed in a leadership meeting."}]]},{"t":"p","runs":[{"text":"The first, unteachable-sounding, entirely teachable skill of the job is converting a request into a question with a decision attached. The move is one sentence: "},{"text":"”What will you do differently depending on the answer?”","i":true},{"text":" If nobody can answer that, the work is decoration and can wait. If they can, you have just learned what precision the answer needs. A decision to spend fifty thousand rupees needs a rougher number than a decision to change pricing for every customer."}]},{"t":"h2","id":"the-maturity-ladder-and-why-it-decides-your-year","runs":[{"text":"The maturity ladder, and why it decides your year"}]},{"t":"p","runs":[{"text":"Data functions sit on a ladder, and where a company sits determines what you will actually do:"}]},{"t":"ol","start":1,"items":[[{"text":"Spreadsheets.","b":true},{"text":" Numbers live in files people email. Your job is extraction and reconciliation."}],[{"text":"A database or warehouse exists","b":true},{"text":" but nothing is documented. Your job is archaeology and writing the same joins repeatedly."}],[{"text":"A transformation layer","b":true},{"text":" (dbt or its equivalent) with tested, documented models. Your job becomes analysis rather than plumbing."}],[{"text":"A metrics layer","b":true},{"text":", so “active customer” means one thing everywhere. Arguments about definitions stop."}],[{"text":"Models in production","b":true},{"text":", with monitoring and retraining."}]]},{"t":"p","runs":[{"text":"An advert for an AI engineer at a company on rung two is an advert for a data engineer who will spend the first year at rung three. That is not a bad job. It is a different one from the one advertised, and taking it without knowing is how people end a year unable to explain what they learned."}]},{"t":"p","runs":[{"text":"Diagnose it with a single question in the interview: "},{"text":"”Where does the number on your main dashboard come from, end to end?”","b":true},{"text":" A confident four-sentence answer means rung three or higher. Hesitation, or “it depends who you ask”, means rung one or two. Both are worth joining. Only one of them matches the advert."}]}],"check":{"question":[{"text":"In an interview, a candidate asks where the number on the company’s main revenue dashboard comes from. The hiring manager says three teams maintain slightly different versions and finance reconciles them monthly in a spreadsheet. What does this most reliably tell the candidate?"}],"options":[{"text":[{"text":"The company is disorganised and the role should be avoided."}],"correct":false,"tempting":[{"text":"Reads a maturity level as a character judgement; most organisations are here, and it is where the highest-impact early work exists."}]},{"text":[{"text":"The team lacks a metrics layer and probably a tested transformation layer, so the first year will be plumbing and definition work regardless of what the advert called the role."}],"correct":true,"tempting":null},{"text":[{"text":"The role will involve building machine learning models on top of finance data."}],"correct":false,"tempting":[{"text":"Assumes advertised work happens; a company reconciling revenue by hand is several rungs below production modelling."}]},{"text":[{"text":"The company needs better dashboard software and the candidate should propose a tool in the interview."}],"correct":false,"tempting":[{"text":"Treats a definition and ownership problem as a tooling problem, which is the most common wrong diagnosis in this field."}]}],"order":[0,3,1,2],"why":null},"prev":{"slug":"reading-a-job-advert","title":"Reading a job advert for what it actually is","minutes":7},"next":{"slug":"degrees-and-doors","title":"Who actually needs a degree","minutes":6}}}