Progressive Estimation

What question does this solve?

The estimate that usually carries the most weight is the one made at the start, when the least is known. A buyer wants to know what something will cost before authorizing it; a provider cannot know that well until some of the work of understanding has been done. The model needs a way to let forecasts follow what has been learned, without turning each forecast into a promise.

Definition and role

Progressive Estimation is the mechanism for updating estimates as context, artifacts and evidence change. It keeps forecasts tied to the conditions known and to the decisions that must be made, and preserves the difference between a forecast and a commitment. In progressive consumption, the work that was bought produces information that can support the next offer: estimating becomes part of the route between learning and deciding how far to continue.

An estimate expresses an expectation about work under certain conditions. It can concern effort, time, cost or risk, and these are not equivalent. Effort is the work needed. Time includes availability, sequence and waiting. Cost depends on the resources used and how they are consumed. Risk expresses relevant possibilities of deviation and their consequences. Less effort in one activity does not show a proportional reduction in time or total cost.

The mechanism does not assume a complete estimate before any work. An initial demand may allow estimating only a bounded exploration, with the rest left as a broad range, a scenario, or a question with no basis for a forecast yet.

Why it matters for the hypothesis

The economic hypothesis begins with the cost of obtaining information useful for deciding. If AI makes it cheaper to clarify rules, try solutions and assess dependencies, it could make that learning viable for demands whose cost of clarification used to be disproportionate. Estimation is where that learning turns into something a buyer can decide on.

How it works

What the estimate refers to

The Delivery Unit gives the reference of the demand, and the Stage Contract bounds a transformation and its conditions. An estimate must say which result and scope it refers to. A forecast for implementation may assume clarified rules and available access; if those do not exist yet, the forecast must make the dependency explicit. A number on its own hides conditions that can change what it means.

Work already done must be kept apart from work remaining. Revising the forecast to finish does not erase earlier consumption; the total combines what happened with what is still expected. The reference can change legitimately: if a feature is added, the new estimate concerns a different scope, and comparing it with the old one must acknowledge that rather than present every variation as an execution error.

Information behind the forecast

The mechanism uses Evidence relevant to the question estimated: clarified rules, identified constraints, evaluation results, observed consumption. Information must be related to its context; an earlier execution can be a useful reference without showing that the current demand needs the same effort. A measure of activity does not replace an estimate of delivery: knowing how long a tool took to produce an output says nothing, alone, about the effort to prepare, evaluate, correct and deploy it. The depth of support should follow the decision, and a lack of basis should stay visible.

What a usable estimate contains

Information	Purpose
Result and scope	Identify the work the forecast refers to.
Dimension estimated	Separate effort, time, cost and exposure to deviation.
Assumptions	Make explicit the conditions taken as true.
Dependencies	Recognize factors that affect starting, executing or accepting.
Range or scenarios	Represent relevant variation without false precision.
Available basis	Relate the forecast to relevant information and experience.
Consumed and remaining	Keep history apart from the forecast of what is left.
Conditions for revision	State which findings or changes call for re-examining it.

An unexplained range can be as unhelpful as a single number: what matters is what would move the work towards each scenario and what remains unknown. A numeric range should not be given statistical meaning it does not have; without a basis for probability, it is a scenario or a working interval.

Updating on relevant events

Revision should follow changes that affect the decision, such as clarifying a rule, confirming a dependency or evaluating an alternative. It is not necessary to redo every estimate after each activity; updates that do not change the understanding add cost without improving the decision. People or systems can produce and revise forecasts within the applicable conditions, and the basis and scope must stay understandable even when revision is automated. A new estimate does not silently replace the old one: keeping earlier revisions shows how understanding evolved and separates changes of scope, of conditions and of execution performance.

More information can widen uncertainty

A discovery can reveal alternatives not yet examined or a more complex dependency, and the forecast may rise or widen. That does not necessarily mean estimation failed; it may mean a constraint is now recognized. The mechanism must allow that without pressure to show artificially growing confidence. The reverse also happens: a resolved constraint can rule out scenarios. Progress should be judged by the basis available, not by the direction of the revision.

Moving along the pipeline only strengthens the basis for a price when it produces knowledge relevant to the scope offered. The position of a stage is not a measure of confidence. A better-informed offer can have a higher price, a wider cost range or stricter conditions. When uncertainty remains relevant, one option is to authorize more investigation; others are to reduce scope or stop. The estimate informs those alternatives; it does not choose between them.

Human, automated and hybrid execution

The composition of execution affects preparation, production, supervision and evaluation, and the Execution Policy and Routing help identify the conditions a forecast assumes. Agents may reduce production effort and require more configuration or review; a human alternative may use existing context and depend on limited availability. The comparison must consider the route to an accepted output. Being able to start many executions does not show a proportionally shorter time: dependencies, evaluations and authorizations still condition continuity.

Not the same as price or commitment

Concept	What it is	What it does not do
Estimate	A forecast of effort, time, cost or risk under stated conditions	It does not set what is charged, and it does not authorize anything
Price	The commercial terms the provider is willing to assume	A new forecast of consumption does not change it automatically
Commitment	What the parties have actually assumed, including time	An estimate of time is not a commitment to a deadline
Authorization	A decision at a Commercial Gate about continuing	An authorized spending limit does not show the work will fit inside it

If the forecast shows that an authorized limit is insufficient, the situation must be raised before anyone assumes permission to exceed it. In an internal operation the same holds for budget and capacity: an updated forecast can support a decision about priority or resources without being an implicit authorization.

In the running example

A barbershop asks for booking, reminders and advance payment. The provider first offers a bounded discovery. It can estimate that advance without presenting a firm construction price for everything that still needs clarifying.

Requirements and prototype reveal which scheduling rules are needed; the technical investigation identifies how to integrate the existing calendar. Together they make it possible to estimate booking and an alternative with reminders; advance payment still depends on clarifications of its own. The provider uses these forecasts to form offers with defined scope and conditions, and can assume a price for booking even though its internal cost may vary. The forecast supports that choice; it does not replace it.

A constraint identified later may call for updating the remaining forecast. That does not change the price already assumed automatically: first one recognizes what the agreement covers. The estimate keeps the history of what was learned, consumed and authorized, instead of rewriting the past with each new piece of information.

Boundaries and failure modes

Learning costs money. Investigating in order to estimate consumes resources. The service has to decide how much learning is useful before offering a commitment: enough to choose, without prolonging discovery in search of certainty that cannot be had. When clarifying a demand requires initial work out of proportion to the advance being bought, the service may need a larger engagement, more exposure, or no offer at all. The AI hypothesis is that this boundary could shift as producing and evaluating information gets cheaper, without assuming that progressive estimation was technically impossible before. What should be measured is the cost of resolving a question and the effect of that information on the decision, including discarded attempts and review.

History helps only with context. Forecasts can improve with records of earlier routes if differences of scope and context are recognized; reusing an average of tasks to promise a new delivery is not enough. Demands that ended in discovery, and unsuccessful attempts, must also inform what is learned. Calibration, the representation of dependencies and the conditions under which AI lowers the cost of estimating without harming decisions remain open. The model prescribes no statistical method and guarantees no cost or time.

Connected concepts

• Economics overview
• Next: Progressive Commitment, what the parties assume on the basis of an estimate
• Pricing, the commercial decision an estimate informs
• Evidence, the information behind a forecast
• Glossary