How do you communicate an inconclusive spike result to leadership?

Basepath Analytics builds a pitch-recommendation tool for minor league clubs. Teodor Marchetti is the AI PM who ran a two-week spike for the Tallow Creek Anchors and has to walk into General Manager Renata Solis's office with an answer that is not a clean yes or a clean no.

How to answer this, stage by stage

Nobody is scoring whether you know the word "inconclusive." They are scoring whether you can hand someone in charge a decision they can actually use when the data will not commit to an answer.

Let's learn

Picture a tool that watches a pitcher's last twenty pitches and suggests the next call: fastball away, curveball in the dirt, whatever the data favors.

Before Basepath's tool, the Tallow Creek Anchors' pitching coach made every call by memory and gut, drawing on maybe 200 pitches he could actually recall clearly from a hitter's last at-bat against the club. The new tool draws on every pitch that hitter has ever seen against anyone, thousands of them, and returns a call in under a second.

Four honest reasons a spike can come back unclear, none of them "we ran a bad test."

Here's the turn: the two-week spike came back split. Strong agreement with the coaching staff on fastball counts. Weak, noisy agreement on breaking-ball counts, on a sample too small to say whether that gap was real or just forty at-bats of bad luck. Teodor's habit up to now had been simple: spikes were either clearly good or clearly bad, and he'd walk in and say so. This one would not sort itself that way.

At its worst, Teodor rounds this into "it works" and the Anchors run it into a real September call that costs a game, or he rounds it into "it doesn't work" and the club walks away from a tool that would have been solid with three more weeks of breaking-ball data.

What I would leave alone: I wouldn't build a special escalation process for every close spike result. Most spikes in this shop land clearly one way or the other; forcing a heavy "confidence band" ritual onto all of them would slow down the easy calls to protect against the rare hard one.

The lesson: the promise that every result would be clear was never true, and the day it wasn't true, the fix wasn't a better spike, it was a better sentence for reporting the honest one.

Now here is the same thing as a story

The short version above is what you'd say defending your report in a five-minute hallway conversation. Read this one for what it felt like the week "inconclusive" almost got treated as a synonym for "no."

Teodor is good at reading a room before he's good at reading a stat sheet. Two years running spikes for minor league front offices taught him that a GM's first question after any pitch is never about the model. It's "so do we use it or not."

For most of that two years, the answer was easy either way. A tool would clear 90 percent agreement across the board and Teodor would say so, plainly, and the club would adopt it. A tool would flop under 60 percent everywhere and he'd say that too, and the club would pass. Clean, both times.

Past spikes landed in a corner. This one landed in the middle, and the middle has no ready-made sentence.

The pitch-recommendation spike for the Anchors did not land in a corner. It landed in the middle. Strong on fastballs. Shaky on breaking balls, on a sample too small to call real.

Then, three days before the readout meeting, a colleague on the eng side leaned over Teodor's desk and said, half joking, "you're not really going to tell Renata it's inconclusive, are you? Just call it a win on fastballs and leave it there."

It was tempting. Rounding up would have made for an easier fifteen minutes in Renata's office. Teodor didn't do it. He printed the split, the sample sizes, and one line about what would close the gap, and walked in with the honest, harder-to-say version.

Two ways to hear the same word. One costs a week. The other costs the whole idea, quietly.

Renata read the printed sheet twice. Her first question was the dangerous one: "so is it ready or not." Teodor didn't flinch into either answer. He said: ready for fastball counts, not yet proven on breaking balls, and here's the exact test that would tell us which in three weeks, at the cost of running it through one more September series.

Same word, three different meetings. Only one of them ends with a decision anyone can actually stand behind.

The real cost was never going to be the fifteen minutes in that office. It was going to be whichever club, months from now, hears "inconclusive" from some other AI PM, reads it as failure, and never finds out the tool was three weeks of data away from working.

Four parts, every time. Skip any one of them and "inconclusive" turns back into a coin flip for whoever hears it.

Renata funded the three-week follow-up. It came back clean: 88 percent agreement on breaking balls too, once the sample was big enough to trust. The tool shipped for the whole rotation, not just fastball counts, and nobody had to pretend the first spike said something it didn't.

What I'd tell my past self, the one who almost took the eng colleague's advice: the easy sentence was never the honest one, and Renata could handle the harder sentence just fine. She couldn't have handled finding out later that I'd rounded it for her.

PICK, the pick that survives the follow-up questionNot a script for hedging every result. PICK is what tells you exactly which kind of wrong read costs more, and to build your report around avoiding that one.

The recap, one line per letter: position is report the range, never the rounded verdict. Impact is Renata's lost week versus a shelved good idea or an underprepared launch. Cost asymmetry is optimizing against the shelved idea, because it never shows up on any dashboard. Kill criteria is the 200-pitch follow-up spike that would turn "inconclusive" into a real yes or no.

And if you want to be sure it really works, try it somewhere elseSame four letters, a veterinary imaging tool instead of a pitch call. Different flip family entirely, same discipline about the word "inconclusive."

Dr. Priya Balan runs diagnostics at Windle Hollow Animal Diagnostics, piloting an imaging tool that flags likely tumors on X-rays for a second look. A ten-case spike came back split: strong agreement with radiologists on soft-tissue masses, weak on bone lesions, on only three cases either way. Mapped onto PICK: position is report the split, not a blended accuracy number. Impact is a vet who trusts a rounded-up "it works" missing a bone lesion, against a clinic that abandons a genuinely useful tool over three unlucky cases. Cost asymmetry favors naming the bone-lesion gap loudly, since a missed lesion costs an animal's health, not just a clinic's convenience. Kill criteria is a twenty-case follow-up read specifically on bone lesions before the tool touches anything but soft tissue.

The same four honest reasons apply here too: small sample, real but noisy signal, mismatched test conditions, uneven strength across categories.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "report the range and the gap, name the next test, never round it to a verdict," and stop.
Cost: there's no budget for a proper follow-up spike. Say so honestly, and scope the cheapest version, ten more at-bats logged during games already being played, rather than skipping the follow-up entirely.
The result is a clear improvement, for real: a spike that comes back strongly positive everywhere still deserves the same discipline, since "clearly great" on forty samples can still hide a rare failure mode nobody tested for.

Where people run it wrong.
They round an unclear result up to look decisive, or down to look cautious, instead of reporting what was actually found.
They give leadership a feeling instead of a number and a next step.
They let "inconclusive" become a permanent state with no test that would ever resolve it.

How to use it live. The moment an interviewer asks how you'd communicate an unclear result, buy two seconds by asking yourself: what specifically would leadership do differently if I rounded this up versus down? Naming that gap out loud is the whole answer.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0