"Pipeline is up this month" is usually not good news. It's congestion.

An action is a signal only when it is more expensive for someone who isn't going to buy. Your own actions cost the same either way, so they carry nothing; the buyer clearing calendar time, naming a number or pulling colleagues in has a real price, and only pays off if they intend to proceed. Meanwhile "we're very interested" is an intention — and the gap between intention and behavior has been measured over and over.

• Hard evidence · intentions move a lot, behavior moves a little: Webb & Sheeran (2006, Psychological Bulletin) meta-analyzed 47 experimental interventions and found that raising intentions by a medium-to-large amount (d ≈ 0.66) produced only a small-to-medium change in actual behavior (d ≈ 0.36). Peer-reviewed meta-analysis — strong. Translated: using "they said they want it" as a stage criterion means forecasting from a measure known to shrink by half.
• Hard evidence · a signal must cost the pretender more: Spence (1973, Quarterly Journal of Economics) — a signal separates real from fake only when it is costlier for the fake. So "they agreed to a meeting next week and put you in front of finance" is worth many times "they said they were very interested": the first costs calendar time and internal political capital. The work later won a Nobel Prize in economics.
• Moderate strength · small commitments predict larger ones: Freedman & Fraser (1966, JPSP) — the foot-in-the-door effect, where agreeing to a small request raises compliance with a larger later one. Burger's (1999, Personality and Social Psychology Review) meta-analysis confirms the effect is real but modest and condition-dependent. Use it as a direction (ask for one small commitment in the meeting), not as a lever. Cross-ref psychology.

Boundary: costly actions can mislead too. Some buyers convene a three-way call purely to collect the third quote their process demands, and you are the stalking horse. Add a second test: are they willing to let you meet other people on their side, and to tell you the timeline and the alternatives? A deal where one person controls all contact and volunteers nothing deserves a discount no matter how deep the stage. Spotting a real champion is covered in the piece on mapping the decision chain.

Drop an entire class of outcome from the denominator and the estimate is biased upward by construction. This is textbook survivorship bias: you counted the ones that finished the process, and the ones that didn't finish carry the most important information. Layered on top is base-rate neglect — when forecasting a single deal, people substitute "this customer seems really warm" for "historically only two in ten get past this stage."

• Hard evidence · representativeness crowds out base rates: Kahneman & Tversky (1973, Psychological Review) showed that when people are given an individuating description that "looks like" a category member, they will very nearly ignore how common that category actually is. Peer-reviewed, heavily replicated. The sales version: every line in your pipeline is one of these resemblance judgments, and your stage conversion rate is the base rate you're ignoring. Cross-ref psychology.
• Classic case, not an experiment · survivorship bias: Wald (1943), working for the wartime Statistical Research Group, analyzed the distribution of hits on returning aircraft and argued the armor belonged precisely where returning planes showed no holes — the planes hit there didn't come back. The original memoranda are real, and Mangel & Samaniego (1984, JASA) give a systematic account. The popular retelling has been dramatized, so use it as an intuition pump, not as evidence.
• Moderate strength · the outside view beats the inside view: Flyvbjerg (2006, Project Management Journal) argues for reference-class forecasting on large projects — predicting from the historical distribution of similar projects rather than from the details of this one — and reports substantially reduced systematic optimism. The theoretical root is Kahneman and Tversky's inside/outside view distinction. Observational work plus a methodological argument, not randomized: credible in direction, discounted in force.

Boundary: a cohort has to mature before it can be read, so a business with a three-month cycle waits at least three months. This is a retrospective tool, not a weekly dashboard. And when the sample is tiny — five leads in a month — no ratio deserves belief at all; look at absolute counts and days-in-stage instead. Some of that no-decision pile is also made of deals you should never have taken, which is the subject of the piece on qualifying.

Two errors stack. First, using an expected value as a prediction on a small sample amounts to pretending variance doesn't exist. Second, the probabilities you feed it were never calibrated — people are chronically over-precise, stating more confidence than their hit rate justifies. Fix neither and the weighted pipeline is just your mood, carried to two decimal places.

• Hard evidence · the law of small numbers: Tversky & Kahneman (1971, Psychological Bulletin) showed that people wrongly expect small samples to display the statistical properties of large ones, and so badly underestimate random fluctuation. "Eight coin-flip deals should give four wins" is the textbook form of that error. Peer-reviewed, classic.
• Hard evidence · stated probabilities run high: Lichtenstein, Fischhoff & Phillips (1982), reviewing the calibration literature, found confidence routinely exceeds accuracy. Moore & Healy (2008, Psychological Review) separate this into three distinct things — overestimating yourself, believing you're better than others, and stating intervals that are too narrow — and identify over-precision as the most stubborn of the three. Peer-reviewed, strong.
• Hard evidence · only feedback fixes it: long-running work by Murphy, Winkler and colleagues on weather forecasters found their probability-of-precipitation forecasts remarkably well calibrated — on the days they call 70%, it rains roughly seventy percent of the time. They are one of the few well-calibrated groups in the literature, and the difference isn't intelligence: it's that they get scored, explicitly, every single day. Which is why point 4 above isn't bureaucracy — it's the one route to accuracy the literature actually supports.
• Contested · where the optimism comes from: Sharot, Korn & Dolan (2011, Nature Neuroscience) reported that people absorb good news more readily than bad when updating beliefs, and tied that asymmetry to signal tracking in the inferior frontal gyrus. The method has drawn substantive criticism — Shah and colleagues (2016) argued the updating paradigm can generate the asymmetry as a statistical artifact. Treat it as a hypothesis, not a settled mechanism. Cross-ref psychology / neuroscience.

Boundary: weighted pipeline isn't useless. With a large enough sample — dozens or hundreds of deals — and probabilities drawn from historical stage conversion rather than typed in by hand, it's a reasonable capacity-planning tool. It goes wrong when it gets treated as the answer to "how much money will arrive this quarter." It's a measure of book size, not a commitment.

Stock is the result; flow is the cause. And the reason a pipeline only ever swells is that two forces push the same way: status quo bias on the buyer's side, where doing nothing is always the safest option, and sunk cost on yours, where the more you've invested the less you can bear to mark it dead. Both say "leave it in," so absent a deliberate purge, a pipeline drifts steadily toward being a graveyard.

• A mathematical theorem, not psychology · Little's Law: Little (1961, Operations Research) proved that in a steady-state queueing system, L = λW. This isn't an empirical regularity; it holds unconditionally given the steady-state assumption, regardless of the business. So "the total went up" must decompose into "more arrived" or "things move slower." There is no third explanation.
• Hard evidence · status quo bias: Samuelson & Zeckhauser (1988, Journal of Risk and Uncertainty) ran a series of experiments showing that merely labeling an option as the current state significantly raises how often it's chosen — and the effect strengthens as the number of alternatives grows. The buyer's default is always "do nothing," which is why the biggest competitor often isn't a competitor. Peer-reviewed, well replicated.
• A single fMRI study (discount accordingly) · overriding a default takes extra effort: Fleming, Thomas & Dolan (2010, PNAS) found heightened activity in the subthalamic nucleus (STN) and inferior frontal cortex when participants rejected a default option, suggesting that departing from the status quo recruits additional inhibitory control. One study, modest sample — a mechanistic clue, not a settled finding. Cross-ref neuroscience.
• Hard evidence · why you won't mark it dead: Staw's (1976, OBHDP) escalation-of-commitment experiments and Arkes & Blumer's (1985, OBHDP) work on sunk cost both show that resources already spent push people to keep investing in a deteriorating choice. On a deal you've worked for eleven months, the hard part isn't judging whether it's dead — it's admitting those eleven months are gone. Peer-reviewed, well replicated.

Boundary: Little's Law assumes steady state, so a team with heavy seasonality, or one that has just changed how it sells, will get misleading answers in the short run. And "mark dead" doesn't mean delete the person: it means remove the deal from the forecast, not blacklist the buyer. Some come back six months later on their own — reviving them is covered in the piece on silence and dead deals, and what a valuable follow-up cadence looks like is in the piece on follow-up discipline.