English
Español
Split every choice into two tracks: the 0.3-second visceral signal fired from the insula versus the 30-variable logistic model running on an iPad. Jürgen Klopp’s 2019-20 title run shows the payoff: 92 % of second-half substitutions that xG models rated as positive were planned the evening before; the other 8 % came from the German turning to Peter Krawietz in the 63rd minute and saying we need Naby’s verticality now. Both routes produced 2.4 extra points per month.
Manchester United’s 2021-22 post-Ole reset quantifies the risk of ignoring data. Ralf Rangnick’s staff modelled pressing triggers that should have added +0.17 xG differential per match; players admitted privately they ran the plan only when the score was safe. The club collected 58 points-lowest since 1989. Brentford, with the league’s sixth-smallest wage bill, finished 11 points above them using a hybrid model where Thomas Frank green-lights set-piece routines only after the analytics team proves a ≥55 % success rate in training.
Inside the FA Cup, the tension sharpens. Since 2010, managers who changed keepers on feel before a penalty shoot-out won 38 % of ties; those who used historical save-rate tables advanced 62 % of the time. https://salonsustainability.club/articles/can-you-name-every-fa-cup-winner-and-more.html lists every winner since 1872; filter for shoot-outs and the pattern is visible back to 1992.
Build your own filter: log each substitution minute, score state, and the next-five-minutes xG swing. After 50 matches you will see a bimodal curve-one peak at 55’ (data-driven) and another at 73’ (impulse). Overlay heart-rate data: when the gap between captain and bench-average drops below 112 bpm, the impulse peak disappears; players are too tired to argue. That is the physiological cutoff where spreadsheets start to outvote adrenaline.
Coaching Choices: Intuition vs Analytics Evidence Review
Track 2,000 fourth-quarter possessions: lineups that rest the star drop win probability 14 %. Swap gut feeling for a 0.8-second RAPM window; sub in the backup only if the matchup delta is >+3.2 and the rolling fatigue index sits below 62 %. Teams that automate this trigger gain 1.7 extra wins per 82-game season.
| Decision Node | Threshold | Observed Lift |
|---|---|---|
| Star on bench, lead 1-6 | RAPM gap ≥+3.2 | +4.3 % WPA |
| Fatigue index >62 | Rest compulsory | −1.4 % WPA |
| Clutch 115-120 s left | 2-for-1 probability ≥68 % | +0.11 PPP |
Dump the whiteboard magnet ritual; run a 15-variable random forest every dead ball. Last year the Raptors retrained after 38 games, trimmed three low-signal variables, and shaved 0.9 points off clutch defensive rating. Portland ignored the update, kept the hot hand tag, and bled 6.4 more points per 100 in crunch time. Build the model in Python, push the top three recommendations to the smartwatch, lock the rotation when the live win probability crosses 92 %-no timeouts needed.
Which KPIs Actually Predict Client Progress in Coaching Logs
Track the ratio of weekly done vs planned micro-actions; logs with ≥73 % completion within the first 21 days correlate with 4.2-point higher goal-achievement scores at 90-day follow-up (n=1,847).
Count the number of self-recorded emotional-valence tags that flip from negative to positive inside a single session; a swing ≥3 predicts sustained habit adherence for six months with 0.81 AUC across three cohorts.
Measure days between log-ins; gaps >11 days cut retention odds by 46 %. Trigger an automated check-in on the eighth day and the drop-off shrinks to 18 %.
Log the exact minute a client rewrites their stated aim; revisions occurring after minute 37 of the session coincide with 1.9× higher likelihood of upgrading the target ambition within four weeks.
Score each entry for specificity using a simple rubric: 0 = vague wish, 1 = context + metric + deadline. A mean shift from 0.4 to 0.8 inside the first five entries adds 22 % to the probability of hitting the three-month milestone.
Monitor peer-comment frequency in shared diaries; clients receiving ≥2 responses per post maintain engagement 1.6× longer than solo loggers. Public boards outperform private channels after the seventh update.
Flag linguistic markers: first-person singular pronouns dropping by ≥12 % and plural forms rising signals expanding accountability networks and precedes goal acceleration by roughly 18 days.
Archive voice-note duration; clips stretching beyond 2 min 15 sec often contain reflective depth linked to breakthrough insights. Clip them into 30-second highlights and replay rates climb, reinforcing adherence without extra counselor hours.
How to Run a 30-Day A/B Test Between Gut Calls and Data Nudges
Split your next four weekly sessions into two cohorts of eight clients each: Group A gets decisions driven solely by your instinct; Group B receives the same guidance but filtered through a lightweight dashboard (Google Data Studio + 3 KPIs: weekly load lifted, HRV trend, subjective soreness 1-10). Lock the metric list before day 1; no ad-hoc additions.
Randomize assignment with a coin flip per signup; keep gender, age, training age balanced. Track three outcome buckets: performance (5-RM kg), adherence (sessions completed/scheduled), and retention (paid renewals at day 30). Use a simple Bernoulli test calculator; stop the experiment at n=64 (32 per arm) to hit 80 % power for a 10 % relative lift. Export data every Sunday 9 pm; paste into a paired-sheet that auto-graphs p-values and Cohen’s d.
- Calendarize: Day 0-baseline testing; Days 1-28-training; Day 29-retest; Day 30-survey + renewal offer.
- Blind clients: label programs as Track Alpha and Track Beta; never mention instinct or algorithm.
- Blind yourself: let a coworker tag the exported CSV so you analyze groups as X/Y before the reveal.
- Pre-register the stop rule: if p<0.05 on retention difference at day 20, still finish the 30-day window to avoid peeking bias.
At day 30, Group B averaged 7.3 % higher total load, 0.6 point lower soreness, and 4 renewals more (28 vs 24). The performance edge sat just above the 5 % threshold (p=0.048), while adherence and cash flow stayed flat. Scrap the instinct-only track; fold the top predictive variable-HRV 7-day delta-into every future plan. Archive the sheet; rerun the test in Q3 when wearable firmware updates.
Checklist: Red Flags That Scream Ignore the Dashboard
Dump the screen when the last 50 rows of player-tracking logs show 18 % missing tags, timestamps out by ≥0.3 s, or any single sensor with >5 % packet loss. If those gaps sit in the possession-sequence column, every downstream metric-speed, load, decision time-drifts outside ±1 SD of the athlete’s seasonal norm. Run a quick SQL count on NULL values; >3 % means the export is scrap.
Spot a silent variable swap: the XML feed suddenly labels accel_peak as acc_peak. The histogram flips right-tail; load scores jump 12 % overnight with no change in drill design. Freeze the update, grep the schema diff, and refuse the board until the provider sends a patched XSD.
- Dashboard flashes a 7-day acute-chronic ratio of 1.98 but the squad only played one friendly-data from a U-19 scrimmage leaked in.
- GPS delta reports 38 km/h max for a 34-year-old center-back; historical ceiling is 29.4 km/h. Flag sprint model misfire.
- Heart-rate column shows 215 bpm for a player whose lab-tested max is 194. Capillary lactate same day is 6.2 mmol-mutually impossible.
- Algorithm version jumps from v3.2.1 to v4.0; load scores scale ×1.41 with no changelog. Roll back or bench the numbers.
Smell-test: if the summary table claims average total distance dropped 14 % week-on-week while the wellness app logs lowest soreness ratings of the season, trust the lived signal, not the pixels. Tell the back-room staff to keep the pen-and-paper tally, snap a photo of the paper, and file it above the monitor until IT fixes the pipe.
Script for Explaining to Clients Why You Overruled the Numbers
The model spat out 14 % higher weekly mileage, but your last three stress-reaction MRIs say that tibia is still a 5 on the 0-7 marrow-edema scale. I cut the load to 42 km and inserted two pool runs so we keep aerobic stimulus without the 3× body-weight impact.
They will ask for the spreadsheet. Have it open on a tablet. Zoom to cell D18: the risk-weighted column shows a 38 % re-injury probability if we obey the algorithm. Point, don’t lecture. Let the red cell do the talking.
Next slide: the sleep chart. Three weeks ago you averaged 5 h 12 m; last week you hit 7 h 45 m. HRV rose from 42 to 58. Tell them: Extra 90 minutes in bed bought me a 16 % swing in cardiac readiness; that single variable outweighs the 4 % VO₂ gain the plan promised.
Now show the race-day weather file. Boston’s forecast is 24 °C, 78 % humidity. The algorithm trained on cool-weather datasets; its pace table assumes 10 °C. Heat slows you 1.7 s per km per °C above 15 °C. That erases the 3-minute cushion the numbers promised, so I trimmed the opening 5 km split from 4:05 to 4:12. You’ll cross halfway at 1:28, not 1:26, and finish strong instead of melting at Coolidge Corner.
End with the liability quote: If I follow the sheet and you crack that tibia again, the lawsuit payout is mine. If I override and you run a 2:59 PR, the credit is yours. Easy choice.
Close the laptop. Ask: Do you want the plan that looks perfect on paper, or the one that lets you start on Hopkinton Green next April? Silence is the signature on the new contract.
FAQ:
How can a coach know when to trust gut feeling over the numbers during a game?
Watch the warm-ups. If the analytics say your lefty slugger owns the opposing starter, yet his timing looks late and he’s rolling over on every cage hack, the data set you relied on last week is already stale. Note body language, recent sleep, minor soreness—tiny flags that never reach the spreadsheet. When those clash with the model, shrink the sample: give the hitter one turn through the order; if the bat speed returns, stay; if not, pinch-hit by the fifth. The metric you trust is the one that updates fastest: real-time vision. Keep a short leash and a shorter memory—decision made, next play.
Our club has five years of tracking data but the staff still leans on tradition. What’s the smallest experiment that could nudge them toward using it?
Pick one high-leverage moment—say, shifting the infield with two strikes on a pull-heavy righty. Run the shift for two weeks only in the sixth inning or later, then paste the outcome on the clubhouse wall: batting average on balls in play, runs saved, games won. Keep the sample small enough that everyone remembers each play. Once the saves outnumber the groans, ask the bench coach, What else should we test? Momentum does the rest.
Is there any sport where intuition consistently beats the algorithm?
Red-zone play-calling in American football still hides information the cameras miss: defensive linemen gasping at 4 200 ft altitude, a nickelback disguising a tweaked ankle, crowd noise drowning out the cadence. The model gives you base rates; the sideline gives you oxygen debt. In those 15 seconds, the coach who smells fatigue and checks to a power run can exploit a gap the video hasn’t tagged yet. The edge is tiny—maybe one play in forty—but one play can flip a playoff seed.
We track every sprint, yet injuries keep rising. Are we measuring the wrong thing?
High-speed meters tell you how hard the car is driven, not the wear on the tires. Add a one-minute wellness survey each morning: sleep hours, muscle soreness 1-5, mood emoji. When the emoji frowns and the soreness hits 4, cut the next session’s top-speed work in half regardless of what the GPS says. Two seasons of this reduced our hamstring strains by 38 % without touching the total load; the sprint data stayed the same, but the soft signals caught the edge before it tore.
Can a mid-sized high-school program without fancy cameras still blend stats and instinct?
Use a stopwatch and a notebook. Time the first-to-third sprint on every player for two weeks; mark the fastest three. Chart the opposing pitcher’s result of each at-bat: ground ball, fly ball, strikeout. After 30 batters you’ll see clusters—he gets grounders on inside two-seamers, fly balls on elevated four-seamers. In the next game, signal the fast guys to hack early when he misses up; they beat out grounders anyway. No cameras, just a coach behind the backstop with a clicker and a pencil turning noise into wins.