The entire operation runs on a single Postgres instance that crunches 0.8 s per query against 11 GB of clickstream, donation and coupon-redemption logs. They rent a $4.50/month VPS, pipe raw CloudFront logs through grep + awk into a 30-line Python loader, and refresh a 17-row pivot table every morning at 07:14. Last quarter the sheet flagged that readers who open three newsletters inside 72 h are 4.3× more likely to whitelist the site. The staff now sell 3-mail bundles to local coffee shops for $0.09 CPM, undercutting regional radio by 92 % and still clearing 63 % margin.
The same scrapheap approach lifted a Polish third-league handball squad from 14th to 3rd place on a €9,700 seasonal budget. They clipped a $40 Kinect sensor above the gym entrance, fed 30 fps joint positions into R, and discovered that left-handed wings average 0.7 s less air-time when receiving passes on the right edge. Coaches drilled 19 set-plays exploiting the asymmetry; goals from that zone rose from 11 to 38, translating into four extra wins and €110 k prize money.
Action list:
1. Pick one pain point that costs you money today.
2. Collect whatever raw feed already exists-CSV exports, server logs, paper receipts, Google Takeout.
3. Clean it with open-source tools you already know: sed, Excel, or SQLite; stop at 80 % perfection.
4. Run a single query that answers the pain; if it runs longer than 90 s, slice the data set by date.
5. Change one variable next Monday, measure delta for two weeks, keep it only if gain > 3 %.
Repeat the loop every sprint. After four cycles you will have outperformed opponents who budget six figures for dashboards they never read.
Pin the One North-Star Metric that Pays the Rent
Track the cash left after ad spend; call it Net Revenue after Marketing (NRAM). A $2 CPM newsletter list, 4 % click-thru, $0.35 eCPM sponsorship, and 70 % open equals $9.80 NRAM per 1 000 sends-anything lower than $7 stalls your runway.
NRAM > $7 means you can pay a $2 500 monthly AWS bill and one $3 000 writer without touching seed money. Below that, kill segments: 30-day inactive readers, EU traffic that needs GDPR banners, and Android 6 users whose fill rate caps at 40 %.
- Compute NRAM nightly with this 12-line Python script pulling from Mailchimp, Stripe, and Facebook Ads APIs.
- Slack bot shouts if NRAM dips 5 % two days straight.
- Freeze all non-brand campaigns until NRAM recovers.
Case: indie SaaS trimmed five under-$0.10 ARPU geos, re-allocated $1 200 to U.S. retargeting; NRAM jumped from $6.40 to $11.90 in 18 days, extending runway by 4.3 months.
Ignore LTV models until NRAM stays above $10 for six consecutive weeks; vanity metrics like pageviews or app installs only clutter the dashboard and burn hours you do not have.
- Build a Looker Studio tile showing yesterday’s NRAM in 72 pt red.
- Share view-only access with every contractor; no one books design hours if the tile glows crimson.
- Pay a 20 % quarterly bonus tied to NRAM growth, not MRR.
One metric, one color, one Slack channel-that is the entire growth program for a scrappy crew operating out of a WeWork closet.
Scrape, Clean, Store for Free: The No-Code Stack Anyone Can Clone
Spin up a Google Colab notebook, pip-install requests-cache and pandas, then aim BeautifulSoup at a target site’s XML sitemap; 30 lines of code pull 12 000 product pages in 18 min, dumping JSON to a free GitHub repo every 1000 records so you never lose work. Point the same script at a public API-say, the Bundesliga’s open endpoint-and add ratelimit (1 call per 2 s) plus exponential backoff; the cached run finishes in 6 h, costs zero credits, and yields 4 seasons of 9 300 shots with xG, xA, freeze-frame coordinates. Strip accents with unidecode, drop rows missing lat/lon, cast dates to ISO-8601, and compress the 210 MB dataframe to 38 MB parquet; GitHub LFS stores it for free under the 1 GB quota. Schedule a weekly refresh with GitHub Actions: a 90-second workflow clones the repo, runs the notebook, pushes the updated file, and opens an issue summarising new rows (median 312 per week) so you track drift without touching a server.
| Tool | Quota | Typical throughput | Exit format |
|---|---|---|---|
| requests-cache + SQLite | ∞ local hits | 1 400 pages/min | JSON |
| GitHub LFS | 1 GB/mo | - | parquet |
| Google Colab | 12 h session | 2.3 GB RAM | notebook |
| GitHub Actions | 2 000 min/mo | 1 min/run | commit diff |
Need a GUI? Import the file into a free Glitch project, click Remix, drop in a 1-click SQLite-to-REST template, and you have an OData endpoint serving 60 000 rows per second; front it with a 30-line Retool page that lets non-coders filter by date, export CSV, or fire a POST webhook back to the notebook for re-scraping. The whole pipeline clones in under 5 min, survives on zero spend, and scales until the repo hits 100 GB-at which point mirror to Backblaze B2 (free tier 10 GB) and update the Action secret; no rewrite needed.
Run Cohort Churn Tests on 100 Users with Google Sheets in 30 Minutes
Import your 100 rows into Col A: user_id, Col B: signup_date, Col C: last_active_date. Add Col D cohort with =EOMONTH(B2,0), Col E days_active with =C2-B2. Pivot: rows = cohort, values = AVERAGE(E). Any cohort averaging <14 days flags churn risk; paint it red. https://librea.one/articles/lakers-get-good-news-on-top-112-million-trade-target.html
Build a survival matrix: F1:AX1 fill 0-29. In F2: =COUNTIFS($D:$D,$D2,$E:$E,>=&F$1)/COUNTIF($D:$D,$D2). Drag. Conditional-format <0.7 in orange. You now see the exact day each monthly group drops under 70 % stickiness without SQL or BI seats.
Save the sheet as template, append fresh exports nightly with Apps Script in 12 lines; the whole loop keeps under the free 500 k cell quota and runs in 43 s on a 1-vCPU Chromebook. Share view-only links to stakeholders; they comment inside cells, cutting Slack noise by half.
Automate Slack Alerts When KPIs Drift 5% Without Writing Code
Connect Google Sheets to Slack via Zapier: trigger fires when a cell with =ABS((B2-C2)/C2) hits 0.05; route webhook to #kpi-alerts; cost 0¢ under Zapier’s 100-task monthly quota.
Sheets formula =IF(ABS((Current-Target)/Target)>0.05,1,0) toggles a helper cell; Zapier watches that cell; Slack message lands in 30 s, tagging @channel if drift >10 %.
Google Analytics 4 exports daily active users to Sheets through the free add-on; same Zap checks column; alert fires when yesterday’s DAUs drop 5 % versus median of last 14 days.
Stripe’s Sigma query select sum(amount) from charges where created > current_date - 1 pushes to Sheets via Coupler; Zap threshold set to 5 % below trailing 30-day average; alert includes exact revenue gap and customer count.
Mixpanel’s Cohorts API streams retention day-7 to Sheets; conditional formatting paints cell red at 5 % decline; Zap grabs RGB value #ff0000 as trigger; Slack posts cohort size, vertical bar emoji, and link to the live board.
Airbnb’s open-source StreamAlert fork on GitHub needs zero deploy: paste the shareable Sheets URL, set env SLACK_WEBHOOK, choose 5 %; runs on Glitch free tier, sleeps after 5 min idle, wakes on cron every 15 min.
One indie studio saved 11 h monthly after replacing manual hourly checks with this no-code stack; churn alert caught a 6 % dip within 12 h, letting them push a patch that recovered $3.8 k ARR.
Keep the helper cell outside print area; hide row 1 so editors don’t overwrite formula; lock sheet with Show warning only; Zapier task history keeps 90-day log for audits.
Swap Costly BI Tools for Metabase on a $5 VPS: Step-by-Step Script
Rent a 1 vCPU / 1 GB RAM VPS at Hetzner CX11 for €3.49, add a floating IPv4, and run curl -fsSL https://get.metabase.com | bash -s v0.47.9; the jar lands in /opt/metabase and starts in 9 s.
Create a 2 GB swapfile so the JVM never hits OOM: fallocate -l 2G /swap && chmod 600 /swap && mkswap /swap && swapon /swap && echo '/swap none swap sw 0 0' >> /etc/fstab.
Systemd unit below keeps the service alive after 512 MB heap; save as /etc/systemd/system/metabase.service, systemctl enable --now metabase and it answers on port 3000 in 12 s.
nginx reverse proxy with free LetsEncrypt TLS trims 40 % of outbound bytes; one cert covers bi.yourdomain.test and renews itself, dropping average response to 68 ms.
Snapshot the 10 GB disk right after install; restoring via Hetzner console takes 38 s, cheaper than any managed Redshift or Power BI Pro seat at $10 per user each month.
Feed CSV dumps from S3: set MB_DB_TYPE=postgres, point MB_DB_CONNECTION_URI to a $5 managed Postgres with 100 GB storage, and Metabase caches queries in 4 s on 20 M rows.
One CX11 handles 35 concurrent users, 280 questions per hour, peaks at 84 % CPU; upgrade to CX31 (€7.49) when daily unique logins exceed 120.
Total monthly burn: VPS €3.49 + Postgres €5 + domain €0.80 = €9.29, saving $1 140 yearly against Tableau Cloud’s 12-user tier while keeping full SQL access and nightly PDF reports mailed by a 9-line Python cron.
Convince Stakeholders With 3-Chart Stories Built From CSV Exports
Export last quarter’s CSV, drop it into RAWGraphs, pick Line + Area, set date as X and cumulative revenue as Y; screenshot the spike on week 12, paste into slide 1 with the title 12-day campaign added 38 %. Stakeholders read numbers, not paragraphs.
Slide 2: same sheet, filter rows where deal_source = referral, group by region, feed the pivot into DataWrapper, choose stacked bar; the 9 % bar for Nordics crushes the 1 % for South. Add red border; finance directors notice color before legends.
Slide 3: divide gross margin by paid_hours, scatter plot in Google Sheets, limit to SKUs with >200 orders; the dot floating alone at 62 % margin and 17 h demand gets a circle and the label Discontinue. One image kills the endless which SKU to drop debate.
Keep each graphic inside 640 px width; 12-point Calibri on axes; no gridlines darker than #D0D0D0; compress PNG to < 90 kB. A 1.2 MB deck mails faster, loads on phones, and avoids IT complaints.
Send the three slides as PDF, name it Q3_decision_pack, attach the raw CSV. Executives forward the PDF; analysts still want the numbers. One mail covers both tribes.
Re-use the template next month: paste new CSV, refresh pivot, charts auto-update. Total prep time drops from 4 h to 18 min, freeing analysts to chase the next 38 % spike instead of polishing slides.
FAQ:
We only track goals and assists. What’s the cheapest way to add one extra data point that actually moves the needle for a semi-pro soccer team?
Start logging the location on the pitch where each pass is received. A $40 tablet, a free app like Futebol Analytics, and one volunteer who tags the video during the match is enough. After five games you’ll see which zones your wingers lose the ball in most often. Shift them five metres deeper and you’ll cut the opponent’s counter-attack frequency by roughly 12 % without buying new players or new kit.
Our handball club has no analyst, just two coaches who already work full-time jobs elsewhere. How many hours per week does one of them need to invest to get a clear picture of what to fix in defence?
Block 90 minutes every Monday night. Use the first 30 min to upload the match video to Kinovea (open source). Tag only three events: blocked shot, offensive foul drawn, failed slide. The software auto-counts. Spend the next hour sorting clips by frequency. The item that shows up most often is the leak; drill it in the next practice. After four weeks you’ll have a 30-clip playlist that replaces a 20-page scouting report nobody reads.
We’re a high-school volleyball program with zero tech budget. Can we predict which freshman will spike hardest two years from now without force plates or jump mats?
Measure wingspan minus standing reach with a tape measure; divide by height. The ratio correlates 0.64 with future spike speed (tested on 112 players in Brazil). Kids in the top quartile (ratio ≥ 0.42) add ~7 km/h to their spike by senior year. Film the approach twice a year and keep the clips in a shared Google Drive folder. You’ll spot late bloomers early and save scholarship money for the ones who actually project upwards.
Community basketball teams here play in tiny gyms with no camera stands. How do we collect at least one useful team stat live without a second pair of eyes?
Give the scorekeeper a $12 foot pedal (USB) and pair it with the SimpleScout phone app. Tap once for paint touch, tap twice for paint touch that leads directly to a score. After the game export the csv, divide second-tap total by first-tap total. If the ratio is under 0.35 your guards are entering the lane but not creating looks; run a 15-minute passing drill after practice until the number climbs above 0.45. That single ratio predicts offensive rating better than field-goal % in most U-18 leagues.
We run a low-budget cycling team. Power meters are out of the question. What’s a proxy for threshold that we can test in a parking lot?
Find a quiet 1 km loop. Warm up, then ride the loop at the fastest pace that still lets you breathe through the nose only. Record lap time and heart-rate at the end of lap 3. Repeat every two weeks. When lap time improves but HR stays the same, threshold is rising; when both climb, fatigue is building. The nose-breathing rule keeps intensity anchored at ~80 % of FTP (validated against lab tests on 30 amateur riders). All you need is a stopwatch and a $25 chest strap.
We track everything—sprints, passes, pressures—but still lose; what are we measuring wrong?
Stop counting events and start measuring timing. Low-budget outfits beat richer ones by turning raw numbers into when stories. Example: instead of recording 78 pressures per match, log the seconds after possession is lost before the first pressure arrives. If that number creeps under two seconds for three straight games, the other side starts coughing up the ball in dangerous spots; your expected goals rise without signing anyone. Strip every other metric for a month and watch whether faster pressure correlates with points. If it does, keep it; if not, bin it. One timing stat beats ten volume stats every Saturday.