# Three questions this dataset can answer These come from the deal-level table `raw.startup_funding`, which has 3,044 deals, 2,459 distinct startups, 115 cities and 57 round types. The year-by-year query results above back the figures below. ## 1. Is Indian startup funding shrinking, or just consolidating into fewer, bigger deals? - **Deal count falls steadily:** 993 deals in 2016, 687 in 2017, 310 in 2018 and 111 in 2019. - **Median disclosed ticket rises:** about $1.0M in 2016, $2.25M in 2017, $4.0M in 2018 and $12.0M in 2019. - **Totals don't follow the deal count:** disclosed totals were about $8.6B in 2015, $10.4B in 2017, $5.1B in 2018 and $9.7B in 2019. Capital per deal is clearly climbing even as activity thins. - **What to check next:** whether a few mega-rounds drive the 2019 total, and whether the drop in deal count is real or reflects less complete reporting in later years. This serves the "accelerating or slowing?" decision for timing a raise. ## 2. Where does the money go: which cities and verticals attract capital, and does that concentration change over time? - **Geography:** the 115 cities can be ranked by deal count, total dollars and median ticket. That tells you where to open an office or source deals. - **Verticals:** the same ranking applies to sectors, but the raw `Industry Vertical` field needs cleaning first. It has 822 distinct values, and the distinct count by year is 724 in 2015 versus 19 in 2016. That looks like a change in how the field was recorded, not a real change in sectors, so any trend needs a normalised taxonomy. ## 3. What does a "typical" raise look like, and who writes the cheques? - **Benchmarks:** the 57 `InvestmentnType` values (seed, angel, Series A/B/C, private equity and so on) can be crossed with vertical and city to give median and percentile ticket sizes. Founders can use these to set a raise size. - **Investors:** every deal has an `Investors Name` value, so you can build a map of which investors repeatedly back which stage, sector and city. Those entries are free text with multiple investors per row, so they need splitting and name cleaning first. ## Data-quality caveats - **Amounts:** only 2,066 of 3,044 deals (about 68%) have a numeric `Amount in USD`, so every dollar figure covers disclosed deals only. For example, 2015 has 646 disclosed amounts out of 935 deals. - **2016 total:** it came back as `nan`, so it is unusable until the offending amount values are found and cleaned. - **Dates:** 7 rows have unparseable dates and one parses to a bogus year (0015). A few other rows have odd year values. I haven't yet run the city, vertical or round-type breakdowns. I can run them next, after cleaning the amount and vertical fields.