The data room has been three weeks away for four months. Not because you do not want the round. Not because the money is unclear. Because building it is work, and work is the thing you are actually discounting.
Deadlines belong on the task, not on the payment. That is the whole operating conclusion, and the evidence behind it is more specific than the usual advice about discipline.
Why these three models
The decision is where to put a commitment device so it binds something. The features that fire are a systematic deviation between what you intend and what you do, a quantity that accumulates while you do not act, and an outcome you will be tempted to judge far too early.
Three lenses, three error structures. The present-bias split produces an equilibrium answer about which domain the bias lives in. The backlog produces a complex answer about why the thing gets worse while you are deciding. The luck-skill ratio produces a random answer about how long you must wait before the result means anything. The second and third disagree with the first in a useful way: one says fix your intentions, the others say your intentions were never the binding constraint.
1. The split: money is not where the bias is
Present bias is usually presented as a single trait. You discount the future steeply, so you delay. The correction that matters is that the effect is not uniform across what is being delayed, and the popular version has been quietly measuring the wrong thing.
Augenblick, Niederle and Sprenger ran the comparison directly, pairing effort choices with a companion monetary discounting study on the same subjects. Their finding, stated in their own abstract: they confirm very limited time inconsistency in monetary choices, while subjects show considerably more present bias in effort.1 The same people, the same session, two domains, and the bias lives in one of them.
The meta-analytic picture agrees on direction. Across 62 papers and 81 estimates, the present bias parameter comes out at 0.82 for monetary rewards and 0.66 for nonmonetary ones, and once selective reporting is corrected for, the monetary estimate rises to 0.87.2 A parameter of 1.0 would mean no present bias at all. So the honest reading is that the money effect is real and mild, the effort effect is materially larger, and the literature has been overstating the first.
There is a second finding in the effort paper that founders should take personally. Present bias in the allocation of work has predictive power for demand for a meaningfully binding commitment device.1 In other words, the people who most need a deadline are the ones who will pay for one, and they can be identified in advance by how they schedule work rather than by what they say about money.
Present bias over effort, the domain lens
- Assumes: the discount applied to a near-term cost differs by whether the cost is money or work.
- Fits because: the delayed thing here is a task, not a payment.
- Breaks when: the delay is genuinely a financing decision, where the money parameter is the right one and it is mild.
- Evidence: grade B. Robust direction, contested magnitude, and the money side shrinks under publication-bias correction.
- Counteracts: the assumption that a bigger payoff will fix a delayed task.
- May reinforce: excusing avoidance that is really about not knowing how to start.
2. The backlog: undone work is a stock
The second lens explains why the delay costs more than the delay.
Undone work is not an event that has failed to happen. It is a stock, and a stock changes only through its flows. It rises with everything that arrives and falls only with what you actually complete. You cannot clear a stock by resolving to clear it, and you cannot clear it by working harder for one day if the inflow is unchanged.
This is the lens people are measurably worst at. Highly educated adults are often unable to infer the behaviour of even simple stock and flow systems, and the failure survives every explanation offered for it: experiments show the persistent poor performance is not attributable to an inability to interpret graphs, to contextual knowledge, to motivation, or to cognitive capacity.3 The authors name it stock-flow failure and treat it as a fundamental reasoning error.3
Applied to the diligence pack, the arithmetic is unforgiving. Every week you delay, the stock grows by whatever new month of accounts, contracts and metrics arrived. The task you are avoiding is strictly larger than the task you avoided last week, which raises the effort cost, which is the exact quantity the first model says you are over-discounting. The bias and the backlog compound each other.
Stock and flow, the accumulation lens
- Assumes: undone work accumulates and drains through identifiable flows, and is conserved.
- Fits because: the cost of the task is not constant while you decide.
- Breaks when: the backlog decays on its own, which occasionally it does. Some of what you have not done has stopped mattering.
- Evidence: grade A. Structural, and the failure to reason about it is one of the better replicated results in the stack.
- Counteracts: treating a delayed task as the same size it was when you first delayed it.
- May reinforce: panic about a stock that is genuinely draining without you.
3. The ratio: how long before the result means anything
The third lens governs what happens after you finally do the work, and it is the one that quietly undoes the other two.
Outcomes mix skill and chance in a ratio, and the ratio sets the sample size you need before a result carries information. Where luck dominates, a single outcome tells you close to nothing, and the required sample is large. Where skill dominates, a short run is already informative.
The founder failure here is symmetrical with procrastination and less discussed. You build the data room, you send it to four investors, three pass, and you conclude the data room was not the problem. In a domain with this much variance, four is not a sample. You have paid the effort cost and then thrown away the information by judging too early, which makes the next effort cost feel less worthwhile, which feeds straight back into the first model.
Luck-skill continuum, the patience lens
- Assumes: the mix of skill and chance in a domain is stable enough to estimate.
- Fits because: you will judge the work by an outcome that arrives with heavy noise attached.
- Breaks when: the ratio itself shifts, which it does when a market changes character.
- Evidence: grade B. The framework is sound and the placement of any specific domain on it is a judgement.
- Counteracts: reading four rejections as a verdict on the artefact.
- May reinforce: waiting for a sample as a way of never concluding anything.
Where to put the deadline, cheapest first
- Move the deadline off the money and onto the task. “Close the round by March” binds nothing, because the money is not what you are discounting. “The three-year P and L is exported by Friday” binds. Cost: rewriting one line in your plan.
- Cut the task until it is smaller than your avoidance. The effort discount applies to the next step, not the project. One exported statement is a different psychological object from a data room.
- Freeze the inflow before you attack the stock. Clearing a backlog while it is still filling is the classic stock-flow error. Stop the new arrivals for a week, or accept that you are running to stay level.
- Buy the commitment, because you are the type who will use it. Demand for a binding device is predicted by how you allocate work.1 A named person expecting a named artefact on a named day is the cheapest version.
- Set the sample size before you send it. Decide now how many investor conversations constitute evidence about the artefact. Deciding afterwards guarantees you decide at the point it hurts most.
What to do this week
Do now, sized at one sitting, effect visible immediately. Take the thing that has been three weeks away for months and write down its next physical step, the one that takes under an hour. Put a name and a date on that step, not on the project. Reversible, free, and dominant across every scenario about why it has not happened.
Hedge, where the premium is the whole loss, and cover live before the next attempt. Tell one person outside the company what the artefact is and when it lands. If you were going to do it anyway you have spent one message, and that is the entire downside.
Defer and trigger, size fixed now. Do not restructure how you plan work this quarter. Pre-commit the trigger instead: the second time the same artefact slips a stated date, it gets cut in half rather than rescheduled. Decide now what half means, because a scope cut decided on the day of a slip is a scope cut decided under pressure.
Note the arrivals as well as the dates. The commitment device works this week; the compounding backlog it was aimed at has been growing for months and will not reverse on the same schedule.
What usually happens next
Run the break test before the base rate. Has a rule changed, has an actor entered or left, has a measurement become a target? If your investors have started asking for a standardised pack that did not exist last year, the task is not the one you have been avoiding and the whole delay was against the wrong object.
If nothing broke, name the shape. This pattern is goal-seeking with a moving goal: you close some of the gap, the gap grows, and the trajectory flattens well short of done. It is not a motivation curve and it does not resolve with a burst of effort, because the inflow was never addressed.
Subtract the counterfactual before you credit the intervention. If the pack finally shipped the week you hired someone, ask whether the hire cleared the backlog or whether the deadline you set at the same time did. Usually it was the named date, and the hire is expensive evidence for a cheap conclusion.
What this ensemble cannot see
All three models treat the avoidance as a discounting or accumulation problem. None of them can tell the difference between a task you are discounting and a task you do not know how to do.
That distinction matters, because the prescriptions diverge completely. Present bias wants a commitment device. Not knowing how wants a worked example or an hour with someone who has done it. Applying a deadline to the second produces a missed deadline and a demoralised founder, and nothing in this framework will tell you which one you are in. The test that does is embarrassingly simple: try to write down the next physical step. If you cannot, it was never procrastination.
There is a second limit. The evidence here is about the direction and rough size of a bias, and the money estimate in particular moves under publication-bias correction.2 Nothing in it licenses a number. Take the split, which is robust, and leave the parameter alone.
And one property none of these models contains: the artefact you keep avoiding is often the one whose contents you are least confident about. The avoidance is then carrying real information about the business, and clearing it efficiently would destroy the signal rather than act on it.
The one action that survives the ignorance: before the end of this week, try to write the next physical step of the thing you have been avoiding. If it takes two minutes, you have a deadline problem and the levers above apply. If you cannot write it at all, you have a different problem and the honest next move is to find one person who has produced that artefact before.
Who has to move
This only changes anything if the person avoiding the task reads it, and that person is usually the founder, who is also the only one with no external deadline. The cheapest first test is to give one artefact a named recipient and a date this week, and see whether it ships. If it does, the constraint was the deadline. If it does not, the constraint was never time and you have learned that for the price of one message.
Sources and notes
- Ned Augenblick, Muriel Niederle and Charles Sprenger, Working Over Time: Dynamic Inconsistency in Real Effort Tasks, NBER Working Paper 18734, 2013. https://www.nber.org/system/files/working_papers/w18734/w18734.pdf. The abstract states that experimental tests of dynamically inconsistent time preferences have largely relied on choices over time-dated monetary rewards, that several recent studies have failed to find the standard patterns, that the authors sidestep the confounds by investigating choices over consumption in the form of real effort in a longitudinal experiment paired with a companion monetary discounting study, that they confirm very limited time inconsistency in monetary choices while subjects show considerably more present bias in effort, and that present bias in the allocation of work has predictive power for demand of a meaningfully binding commitment device.
- Stephen L. Cheung, Agnieszka Tymula and Xueting Wang, Quasi-Hyperbolic Present Bias: A Meta-Analysis, IZA Discussion Paper 14625, 2021. https://docs.iza.org/dp14625.pdf. The abstract reports a meta-analytic dataset of estimates of the present bias parameter based on 62 papers with 81 estimates, finds people are on average present biased for both monetary rewards (0.82, with a 95% confidence interval of 0.74 to 0.90) and nonmonetary rewards (0.66, 0.51 to 0.85), notes substantial heterogeneity across studies, and reports evidence of selective reporting and publication bias in the direction of overestimating the strength of present bias, with the monetary estimate rising to 0.87 (0.82 to 0.92) after correction.
- Matthew A. Cronin, Cleotilde Gonzalez and John D. Sterman, Why don’t well-educated adults understand accumulation? A challenge to researchers, educators, and citizens, Organizational Behavior and Human Decision Processes 108(1), 2009, pages 116 to 130. Author copy: https://www.mit.edu/~jsterman/CroninGonzalezSterman061210.pdf. The abstract states that highly educated people are often unable to infer the behaviour of simple stock-flow systems, and that in a series of experiments persistent poor performance is not attributable to an inability to interpret graphs, contextual knowledge, motivation, or cognitive capacity.
A note on a number that is deliberately absent. It would be easy to write that people discount effort some multiple more steeply than money. The sources do not support a ratio, the money parameter itself moves under publication-bias correction, and the useful content is the split rather than its size. The direction is what changes where you put the deadline.
Joshua Agonya Pi’Rwot, Founder.