{"course":{"slug":"ai-careers","title":"Working in AI","url":"https://addaly.com/learn/ai-careers","level":"beginner","track":"work"},"lesson":{"slug":"research-before-the-call","title":"Forty minutes of research, and how to check it","url":"https://addaly.com/learn/ai-careers/research-before-the-call","position":46,"of":85,"minutes":8,"answer":"Research is aimed at how the organisation makes money, where the role sits and what remains uncertain, and every fact a model gives you needs a source you actually open.","keyIdea":"Research is aimed at how the organisation makes money, where the role sits and what remains uncertain, and every fact a model gives you needs a source you actually open."},"body":{"format":"markdown","text":"## What you are actually trying to learn\n\nNot trivia. Three things:\n\n1. **How this organisation makes money**, because everything about the role follows from it.\n2. **Where the role sits**: whose problem you would be solving, and what number would move if you did it well.\n3. **What is uncertain**: the thing you would want to know before saying yes.\n\nForty minutes gets all three. Four hours gets you a fact about their 2019 funding round that impresses nobody.\n\n## The sequence\n\n**The product page and the pricing page (10 minutes).** What do they sell, to whom, and what does it cost? Pricing tells you more than the About page: per-seat pricing means many small customers and churn matters; six-figure enterprise deals mean few customers, long sales cycles and a data team that supports sales.\n\n**All their open roles (5 minutes).** This is the most underused source in a job search. Twelve engineering roles and no data roles tells you where you would sit. Three roles mentioning Snowflake and one mentioning Databricks tells you a migration is happening. Hiring a Head of Data at the same time as your role means your manager may not exist yet.\n\n**Their engineering blog and GitHub organisation (10 minutes).** The stack, the scale, and the problems they thought worth writing about. If there is nothing, that is also information.\n\n**Money and trajectory (5 minutes).** Listed company: the annual report says the segments and the risks in their own words. Private: Crunchbase's free tier gives the funding stage and date. A company that raised in 2022 and not since is in a different mood from one that raised last quarter.\n\n**People (5 minutes).** Your interviewers' backgrounds on LinkedIn: what they did before, how long they have been there. Long tenures across a team is a good sign; everyone arriving in the last eight months means something changed.\n\n**Reviews (5 minutes), read carefully.** Glassdoor, AmbitionBox, Indeed. The selection is biased towards the delighted and the furious, so ignore the ratings and read only for repeated specifics. Four separate people mentioning the same manager or the same reorganisation is signal. Twenty people saying \"great culture\" is not.\n\n## Using a model without inheriting its errors\n\nA model is good at compressing what you found and bad at knowing what is true about a private company in your city. It will produce a fluent paragraph about their funding, their customers and their technology stack, some of which will be invented, and none of which is labelled.\n\nTwo habits fix this:\n\n- **Paste the source and ask about it**, rather than asking from memory. \"Here is their careers page and their pricing page: what do these imply about the shape of the engineering team?\" is a task it does well.\n- **Demand sources for anything it asserts.** *\"For each claim, give the URL. Mark anything you cannot source.\"* Then open two of them. When a search tool is available, this converts most of the guessing into checkable links; when it is not, the honest output is a very short list.\n\nNever repeat an unverified number in an interview. \"I saw you raised a Series B last year\" is a small, humiliating error when the round was a Series A and it was 2023.\n\n## What the research is for\n\nThree questions you could not have asked otherwise. That is the whole return.\n\n- \"Your pricing suggests most revenue comes from a small number of large accounts: does the data team mostly serve sales and customer success, then?\"\n- \"You are hiring a Head of Data alongside this role. Would I be reporting to that person once they start?\"\n- \"Your engineering blog described moving off Redshift in 2024. Is that finished, or would I be joining mid-migration?\"\n\nEach shows you did the work, and, more usefully, each answer changes whether you want the job.\n\n## Two limits, one practical and one about manners\n\nResearch does not win interviews. It prevents the specific loss where you cannot say why you want this job rather than any job, and it stops you accepting something you would have refused with better information. That is worth forty minutes and not worth four hours.\n\nAnd there is a line on people. Reading someone's professional background is normal preparation. Reading their personal social media, mentioning their marathon time, or naming their children is not research, it is surveillance, and it lands exactly as badly as you would expect. Stay on the professional side and stop there."},"check":{"question":"A candidate is preparing for an interview at a private company. Which source most reliably reveals the shape of the team they would join?","options":[{"text":"The company's full list of currently open roles, read together.","correct":true},{"text":"A model's summary of the company, its stack and its recent history.","correct":false,"tempting":"Produces fluent, unlabelled claims about a private company where public information is thin, which is exactly the condition under which fabrication is most likely."},{"text":"The overall star rating on Glassdoor or AmbitionBox.","correct":false,"tempting":"Aggregates a sample skewed towards the delighted and the furious, so the number moves with mood rather than with structure."},{"text":"The company's About page and mission statement.","correct":false,"tempting":"Describes how the company wishes to be seen, and is written to be true of almost any organisation in the sector."}]},"prev":{"slug":"what-you-paste-into-a-chatbot","title":"Your CV, their take-home, and the terms","url":"https://addaly.com/learn/ai-careers/what-you-paste-into-a-chatbot"},"next":{"slug":"the-ai-screened-interview","title":"Recorded and machine-scored interviews","url":"https://addaly.com/learn/ai-careers/the-ai-screened-interview"},"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) 'Forty minutes of research, and how to check it', in Working in AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-careers/research-before-the-call (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":"Forty minutes of research, and how to check it","courseTitle":"Working in AI","module":{"index":5,"slug":"searching-with-ai","title":"Using AI in your own search"},"answer":null,"keyIdea":"Research is aimed at how the organisation makes money, where the role sits and what remains uncertain, and every fact a model gives you needs a source you actually open.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"what-you-are-actually-trying-to-learn","runs":[{"text":"What you are actually trying to learn"}]},{"t":"p","runs":[{"text":"Not trivia. Three things:"}]},{"t":"ol","start":1,"items":[[{"text":"How this organisation makes money","b":true},{"text":", because everything about the role follows from it."}],[{"text":"Where the role sits","b":true},{"text":": whose problem you would be solving, and what number would move if you did it well."}],[{"text":"What is uncertain","b":true},{"text":": the thing you would want to know before saying yes."}]]},{"t":"p","runs":[{"text":"Forty minutes gets all three. Four hours gets you a fact about their 2019 funding round that impresses nobody."}]},{"t":"h2","id":"the-sequence","runs":[{"text":"The sequence"}]},{"t":"p","runs":[{"text":"The product page and the pricing page (10 minutes).","b":true},{"text":" What do they sell, to whom, and what does it cost? Pricing tells you more than the About page: per-seat pricing means many small customers and churn matters; six-figure enterprise deals mean few customers, long sales cycles and a data team that supports sales."}]},{"t":"p","runs":[{"text":"All their open roles (5 minutes).","b":true},{"text":" This is the most underused source in a job search. Twelve engineering roles and no data roles tells you where you would sit. Three roles mentioning Snowflake and one mentioning Databricks tells you a migration is happening. Hiring a Head of Data at the same time as your role means your manager may not exist yet."}]},{"t":"p","runs":[{"text":"Their engineering blog and GitHub organisation (10 minutes).","b":true},{"text":" The stack, the scale, and the problems they thought worth writing about. If there is nothing, that is also information."}]},{"t":"p","runs":[{"text":"Money and trajectory (5 minutes).","b":true},{"text":" Listed company: the annual report says the segments and the risks in their own words. Private: Crunchbase’s free tier gives the funding stage and date. A company that raised in 2022 and not since is in a different mood from one that raised last quarter."}]},{"t":"p","runs":[{"text":"People (5 minutes).","b":true},{"text":" Your interviewers’ backgrounds on LinkedIn: what they did before, how long they have been there. Long tenures across a team is a good sign; everyone arriving in the last eight months means something changed."}]},{"t":"p","runs":[{"text":"Reviews (5 minutes), read carefully.","b":true},{"text":" Glassdoor, AmbitionBox, Indeed. The selection is biased towards the delighted and the furious, so ignore the ratings and read only for repeated specifics. Four separate people mentioning the same manager or the same reorganisation is signal. Twenty people saying “great culture” is not."}]},{"t":"h2","id":"using-a-model-without-inheriting-its-errors","runs":[{"text":"Using a model without inheriting its errors"}]},{"t":"p","runs":[{"text":"A model is good at compressing what you found and bad at knowing what is true about a private company in your city. It will produce a fluent paragraph about their funding, their customers and their technology stack, some of which will be invented, and none of which is labelled."}]},{"t":"p","runs":[{"text":"Two habits fix this:"}]},{"t":"ul","items":[[{"text":"Paste the source and ask about it","b":true},{"text":", rather than asking from memory. “Here is their careers page and their pricing page: what do these imply about the shape of the engineering team?” is a task it does well."}],[{"text":"Demand sources for anything it asserts.","b":true},{"text":" "},{"text":"”For each claim, give the URL. Mark anything you cannot source.”","i":true},{"text":" Then open two of them. When a search tool is available, this converts most of the guessing into checkable links; when it is not, the honest output is a very short list."}]]},{"t":"p","runs":[{"text":"Never repeat an unverified number in an interview. “I saw you raised a Series B last year” is a small, humiliating error when the round was a Series A and it was 2023."}]},{"t":"h2","id":"what-the-research-is-for","runs":[{"text":"What the research is for"}]},{"t":"p","runs":[{"text":"Three questions you could not have asked otherwise. That is the whole return."}]},{"t":"ul","items":[[{"text":"“Your pricing suggests most revenue comes from a small number of large accounts: does the data team mostly serve sales and customer success, then?”"}],[{"text":"“You are hiring a Head of Data alongside this role. Would I be reporting to that person once they start?”"}],[{"text":"“Your engineering blog described moving off Redshift in 2024. Is that finished, or would I be joining mid-migration?”"}]]},{"t":"p","runs":[{"text":"Each shows you did the work, and, more usefully, each answer changes whether you want the job."}]},{"t":"h2","id":"two-limits-one-practical-and-one-about-manners","runs":[{"text":"Two limits, one practical and one about manners"}]},{"t":"p","runs":[{"text":"Research does not win interviews. It prevents the specific loss where you cannot say why you want this job rather than any job, and it stops you accepting something you would have refused with better information. That is worth forty minutes and not worth four hours."}]},{"t":"p","runs":[{"text":"And there is a line on people. Reading someone’s professional background is normal preparation. Reading their personal social media, mentioning their marathon time, or naming their children is not research, it is surveillance, and it lands exactly as badly as you would expect. Stay on the professional side and stop there."}]}],"check":{"question":[{"text":"A candidate is preparing for an interview at a private company. Which source most reliably reveals the shape of the team they would join?"}],"options":[{"text":[{"text":"The company’s full list of currently open roles, read together."}],"correct":true,"tempting":null},{"text":[{"text":"A model’s summary of the company, its stack and its recent history."}],"correct":false,"tempting":[{"text":"Produces fluent, unlabelled claims about a private company where public information is thin, which is exactly the condition under which fabrication is most likely."}]},{"text":[{"text":"The overall star rating on Glassdoor or AmbitionBox."}],"correct":false,"tempting":[{"text":"Aggregates a sample skewed towards the delighted and the furious, so the number moves with mood rather than with structure."}]},{"text":[{"text":"The company’s About page and mission statement."}],"correct":false,"tempting":[{"text":"Describes how the company wishes to be seen, and is written to be true of almost any organisation in the sector."}]}],"order":[3,0,1,2],"why":null},"prev":{"slug":"what-you-paste-into-a-chatbot","title":"Your CV, their take-home, and the terms","minutes":8},"next":{"slug":"the-ai-screened-interview","title":"Recorded and machine-scored interviews","minutes":9}}}