Estimation, planning, and honest commitments

Estimates are unreliable by nature. At the start of a piece of work you know the least you ever will, the happy path is the easiest one to imagine, and the hidden work often outweighs the visible. So the competency is not producing accurate estimates. It is managing the uncertainty honestly, which means estimating in ranges with named assumptions, slicing work small so errors compound less, de-risking before committing, and keeping the line between an estimate and a commitment sharp enough that everyone knows which one they are hearing. Delivery credibility is built or destroyed here. When an optimistic estimate posing as a commitment is defended until the deadline and then missed, the cost goes beyond the slip. It also spends the trust that would have made the next honest forecast believable. The examinations test this because the pressure to give a confident date never lets up at any scope, and the defensible answer is nearly always the one that names what is not yet known.

1. An estimate is not a commitment, and both are said aloud. "Our best estimate is this, and what we can commit to is that." The gap between the two is the buffer against uncertainty. Naming it, where it could have been hidden, is what builds trust.
2. Ranges, not points. A single date at the start of uncertain work is a range in disguise. State the range and the assumptions that set its width.
3. Slice small and de-risk first. Small pieces are estimated more accurately and their errors compound less. A time-boxed spike on the fuzziest part narrows the range faster than any amount of planning.
4. History beats hope. How long similar work took in the past is a better forecast than how long anyone hopes this piece will take.
5. Re-forecast the moment reality diverges, and say so. Surprises sprung at the deadline are how trust dies. The honest form of a slip is early disclosure with options on date, scope and resourcing.
6. Do not guess under pressure; be decisive about the process instead. If you are asked for a date on the spot with no estimate, commit to the date by which a credible estimate will exist, name the key unknowns, and commit to scope. Being decisive about how the answer will be produced is different from inventing the answer.

Estimation is one of the better-studied problems in both software engineering and cognitive psychology, and this unit has one research-grade model alongside its practitioner ones.

The planning fallacy research

Daniel Kahneman and Amos Tversky, Intuitive Prediction: Biases and Corrective Procedures (1979) found that people predicting how long a task will take systematically underestimate it, even when they know similar tasks overran in the past. They reason from the specifics of the case in front of them and ignore the distribution of similar cases. The corrective the authors propose is the outside view, which forecasts from the reference class of comparable work, and it is the psychological basis for principle 4. The effect is well established and has been replicated across domains. It is why asking people to try harder does not fix estimation.

The cone of uncertainty practice

The cone of uncertainty, given its standard software form in Steve McConnell, Software Estimation: Demystifying the Black Art (2006), is the observation that estimates are least accurate at the start of work and narrow as reality is discovered. The working practices of this unit follow from it: estimate in ranges, slice work small, de-risk the fuzziest parts first, and re-forecast as the cone narrows. It narrows only if someone does the discovering. Uncertainty nobody examines stays wide, and the model offers no excuse for failing to re-forecast once more is known.

Cycle time and historical forecasting practice

This practice bases forecasts on measured data about how long work takes, in place of how long people estimate it will take. It is the operational form of the outside view. History predicts only work that resembles the past, so novel work still needs de-risking spikes. A cycle-time distribution is itself a range, and quoting its average as a date reinvents the original error.

Brooks's law practice

Frederick P. Brooks Jr., The Mythical Man-Month: Essays on Software Engineering (1975) observed that adding people to a late software project makes it later, because the existing people have to train the new ones and the number of communication paths multiplies. This is why "add people" is rarely the honest option when a commitment is at risk, and why the options left are date, scope and, occasionally, resourcing applied early instead of late. The law's scope is narrower than its fame suggests. It applies to work that cannot be partitioned, and the judgment lies in telling that work apart from work that can.

Batch size and predictability practice

Donald G. Reinertsen, The Principles of Product Development Flow: Second Generation Lean Product Development (2009) argues that predictability is a property of the delivery system rather than of effort. Smaller batches and limited work in progress reduce variance in delivery times more reliably than pressure does. So a team producing wildly inconsistent estimates gets engineering help (smaller slices, historical data, forecast reviews) and no exhortation.

The estimate with its assumptions

Any estimate longer than a few days is given as a range, with the two or three assumptions that set its width written beside it. When an assumption turns out to be wrong, the range moves and the reason is plain to see. An estimate whose assumptions nobody can state is an impression.

The commitment, buffered and explained

The outward promise is derived from the estimate with a deliberate, stated buffer, and the stakeholder hears both numbers. The commitment also states the scope it covers, the known risks, and how variance will be communicated. A commitment is a shared understanding of those four things. It is no promise to hit a date whatever happens.

The spike before the plan

Before committing to work with a wide range, the team runs a time-boxed investigation of the fuzziest part, with a defined question and a fixed end. It produces a narrower range, or the discovery that the work is something other than it seemed. A week spent learning that a quarter's estimate was out by a factor of two is the best week of the quarter.

Forecast reviews

At a fixed cadence, the team compares its forecasts with its actuals and asks why they differed. Chronic over-commitment and chronic sandbagging are handled the same way, because neither kind of forecast carries any information. Both are recalibrated against historical data. A team that delivers exactly to plan every quarter, with no misses, is more likely to be sandbagging than excelling.

The early slip conversation

In the week a commitment becomes doubtful, the stakeholder hears about it, along with the cause and the options: move the date, reduce the scope, or add resourcing where it will help. The conversation happens while there are still choices to make. A slip disclosed late turns a delivery problem into a trust problem.

At team scope the object is the team's own forecast and commitment. The manager states ranges, buffers deliberately and re-forecasts in the open. At organisational scope the object becomes commitments that depend on several teams. There the uncertainty has to be aggregated honestly (adding up best cases will not do), and the truthfulness of rolled-up status becomes the thing to manage. Calibrating forecasts across teams is a system to run, and a speech will not achieve it. At executive scope the object is the organisation's planning cadence and the credibility of its commitments to the business and the board. That means staged commitments with checkpoints, credibility repaired through a sequence of smaller verifiable deliveries in place of bolder promises, and the flagship call between date and quality made on total cost over time. How Judgment Scales sets out the pattern, and DE-4 Managing scope, risk, and dependencies across workstreams covers the reporting system that carries status upward.

• Gives ranges with named assumptions and separates the estimate from the commitment. team
• Buffers the commitment deliberately and says so. team
• Slices work small and de-risks the fuzziest part with a time-boxed spike before committing. team
• Re-forecasts early and openly when reality diverges, with options. team
• Declines to give a date on the spot with no estimate, and says what it would take to give one.
• Failure mode: treating an estimate as a promise. team
• Failure mode: silent padding, so that nobody knows what the number means. team
• Failure mode: committing to the optimistic number to dodge a hard conversation. team
• Failure mode: going quiet as the timeline slips. team
• Defines a department commitment as scope, known risks and a variance protocol, and guards that definition under pressure. org
• Aggregates uncertainty across several teams honestly instead of adding up best cases. org
• Treats chronic over-commitment and chronic sandbagging as the same calibration failure and recalibrates both against history. org
• Improves predictability through batch size and work-in-progress limits instead of pressure. org
• Failure mode: letting a commitment mean "guaranteed date". org
• Failure mode: guessing under executive pressure and calling it decisiveness. org
• Failure mode: praising the sandbagger's streak. org
• Failure mode: demanding better estimates without changing how work is sliced or measured. org
• Gives the board a staged commitment: near-term milestones firm, later phases as ranges that tighten at named checkpoints, unknowns stated plainly. exec
• Rebuilds credibility after repeated misses by diagnosing the systemic cause and delivering smaller, verifiable commitments in sequence, without promising bigger. exec
• Discloses a major miss early with options, while choices still exist. exec
• Makes the flagship date-versus-quality call on total cost over time, decides explicitly, and communicates the reasoning. exec
• Reads execution health as whether the organisation reliably converts commitments into delivered outcomes, and not as activity. exec
• Failure mode: giving the board false precision because it asked for a date. exec
• Failure mode: re-promising bigger after each miss. exec
• Failure mode: sitting on a known miss until it is undeniable. exec
• Failure mode: holding a date by shipping something that will fail at scale. exec

Honesty versus confidence. Stakeholders want a date and read a range as evasion. A point estimate sounds confident and is usually wrong. The way through is to be confident about the process (the checkpoints, the spikes, the date by which the range will narrow) while staying honest about the number.

Buffer versus credibility. A commitment with too little buffer gets missed. One with too much looks like sandbagging, and if it is always hit, that is what it is. Naming the buffer resolves this, because the stakeholder sees both numbers and the reason for the gap.

Slicing versus coherence. Small pieces estimate well but can lose sight of the whole. Large pieces keep the shape of the work and estimate badly. The judgment is to slice for estimation while planning for the outcome, and to keep the connection between the two views visible.

Early disclosure versus premature alarm. A slip disclosed as soon as it becomes possible spreads alarm about risks that may never materialise, and one disclosed only once it is certain comes too late for options. The rule is to disclose when the probability passes the point at which the stakeholder would want to act, and to say plainly that it is a probability.

Date versus quality. Holding a launch date by cutting quality and slipping the date to protect quality are both real options with real costs. The call is made on total cost over time, whatever this quarter's optics. The failure the examinations present most often is the launch that ships on its date and then fails at scale.

A senior manager who leads four teams is in a leadership meeting when a vice-president asks them, on the spot, to commit the group to a launch date. The initiative was scoped in a slide deck a week ago and no team has estimated it. The vice-president wants a date for a customer conversation that afternoon and says, in front of the room, that a good leader should be able to give one.

There are two easy answers, a guess and a refusal. A guess delivered with confidence becomes the commitment the moment it leaves the room, and it will be defended against the evidence for the next quarter. A refusal ("we can't commit without estimating") is true, but it sounds unhelpful and leaves the vice-president with nothing to tell the customer.

The senior manager is decisive about the process and leaves the number open. They explain that the initiative has not been estimated, so a date given now would be a guess passed off as a plan. They promise a credible estimate, as a range with its drivers, by the end of the following week, and they commit today to the scope of the first phase, which is understood well enough to start. They name the two unknowns that will set the width of the range (a dependency on a platform team, and a data migration nobody has sized) and say what the group will do in the coming week to narrow both. For the customer conversation they give the vice-president a sentence that is true: the first phase is committed and starts now, and the full timeline will be set within two weeks.

There is a plan behind the answer. The four teams run a one-week spike on the migration and hold a scoping conversation with the platform team. The estimate that comes back is a range with three assumptions attached. The commitment offered to the vice-president is buffered against that range, states the scope it covers and the risks, and says how variance will be communicated. The vice-president gets a date within two weeks, and it is a date that carries information.

The senior manager does not let the pressure in the room push them into inventing a number. The honest answer costs a moment of discomfort in a meeting. The other answer costs a quarter spent defending a date nobody believed, followed by a launch that either slips or ships broken.

• DE-1 Prioritisation under constraint: the sequencing a commitment is made against, and the trade-offs that move when it slips.
• DE-4 Managing scope, risk, and dependencies across workstreams: the dependencies and risks that widen a range, and the status reporting that carries a slip upward.
• SV-2 Communicating and partnering across functions: explaining a range and its drivers to a stakeholder who wants a date.
• CC-5 Values, integrity, and ethical judgment under pressure: declining to shade a forecast under pressure, and absorbing the cost of declining.
• DE-5 Designing how work flows: batch size and flow as the system properties that make forecasts believable.

1. What is the difference between an estimate and a commitment, and why say both aloud?
   Answer

   The estimate is the best forecast. The commitment is what is promised outward, with a deliberate buffer. Saying both names the buffer, so nobody mistakes one for the other, and that visible gap is what builds trust.
2. Why is a single date at the start of uncertain work "a range in disguise"?
   Answer

   At the start you know the least you ever will, so the range is at its widest. A point estimate hides that width without removing it. State the range and the assumptions that drive it.
3. A vice-president asks you on the spot to commit your group to a date for unestimated work. What is the credible response? org
   Answer

   Commit to a date by which a credible estimate will exist, name the key unknowns and what will narrow them, and commit to the scope that is understood. Be decisive about the process and honest about the number.
4. One team consistently over-commits and misses; another quietly sandbags and always hits. Why are these the same problem, and what fixes both? org
   Answer

   Both produce forecasts that carry no information. Both are fixed by recalibrating against historical data, using smaller slices, cycle-time distributions and forecast reviews. A speech for one team and congratulations for the other fix neither.
5. Why does adding people to late work usually make it later, and what does that leave as options?
   Answer

   The existing people have to train the new ones and communication paths multiply, so the work slows down before it speeds up. What remains is date, scope, or resourcing applied early enough to help, disclosed while there are still choices.
6. What makes a staged commitment more credible to a board than a single date, and what does repeated re-promising cost? exec
   Answer

   Near-term milestones are committed firmly, later phases are given as ranges that tighten at named checkpoints, and unknowns are stated plainly. That is the only kind of forecast a sophisticated audience should believe. Re-promising bigger after each miss spends the credibility that would have made the next honest forecast believable.
7. The flagship launch can hold its date by cutting quality or slip it. What is the basis for the call? exec
   Answer

   Total cost over time, whatever this quarter's optics. A launch that fails at scale costs more than a slip. Decide explicitly and communicate the reasoning.

• Daniel Kahneman and Amos Tversky, Intuitive Prediction: Biases and Corrective Procedures (1979): the planning fallacy and the outside view.
• Steve McConnell, Software Estimation: Demystifying the Black Art (2006): the cone of uncertainty; ranges and slicing.
• Frederick P. Brooks Jr., The Mythical Man-Month: Essays on Software Engineering (1975): adding people to late work.
• Donald G. Reinertsen, The Principles of Product Development Flow: Second Generation Lean Product Development (2009): batch size and variance; predictability as a system property.