Complex search goals are decomposed into a sequence of micro-steps that can be partially executed, suspended, and resumed across sessions. The system tracks the user's position in the decomposition and ranks content for the current micro-step rather than the abstract goal.
Patent Overview
- Inventor
- Jaime Teevan, others
- Assignee
- Microsoft Corporation
- Filed
- 2014-09-12
- Granted
- August 28, 2018
The Challenge
The Challenge
A goal like 'switch our team to a new project management tool' is not a query. It is a multi-week decision process that fragments into sub-questions about feature comparison, pricing, migration, training, and rollout. The challenge: decompose the goal into micro-steps the user can act on in short windows of attention, track which steps are done, and rank content for the open step rather than the abstract goal.
- Abstract Goals Are Not Queries — Per goal, the user's actual intent is too large to be a single query, so it must be broken into smaller actionable steps.
- Attention Windows Are Short — Per session, users have brief windows of focus, often interrupted, that fit micro-steps but not entire goals.
- Progress Is Invisible — Per goal, the system cannot tell which micro-steps the user has completed and which remain open.
- Resumption Is Lossy — Per resumed session, the user often re-investigates already-answered sub-questions because the engine does not surface their completed work.
- Ranking Targets The Query, Not The Step — Per query, the ranker reads the literal string and not the micro-step it represents inside the larger goal.
Innovation
How The System Works
The system accepts or infers a higher-level goal, decomposes the goal into a sequence of micro-steps appropriate to that goal type, tracks the user's progress through the steps across sessions, and ranks content for the open step the user is currently working on rather than for the abstract goal.
- Capture Or Infer The Goal — Per user, the higher-level goal is either explicitly entered or inferred from a cluster of related queries.
- Decompose The Goal — Per goal type, the system applies a decomposition template that breaks the goal into ordered or partially-ordered micro-steps.
- Surface The Step List — Per goal, the user sees the micro-step list and can mark progress, reorder, or annotate the steps.
- Detect The Active Step — Per session, the active micro-step is inferred from the current query and recent activity.
- Rank For The Active Step — Per query, the ranker conditions on the active step so the SERP serves the micro-step rather than the abstract goal.
- Persist Step State Across Sessions — Per goal, step completion and partial state persist so the user resumes where they left off.
- Suggest The Next Open Step — Per goal, after a step is completed or set aside, the system suggests the next open step to keep the goal moving.
Tasks Resume At The Micro-Step, Not The Goal
The patent's load-bearing idea is that real work happens in micro-steps, not in goal-sized chunks. The system that tracks the user's position in the decomposition serves the next step every time the user returns, rather than restarting the goal from scratch.
Step-Conditional Ranking
Per goal, the active micro-step conditions ranking so results match the user's current open step. Per session, resumption picks up where the prior session left off.
- Goal Decomposition — Per goal type, a template breaks the goal into ordered micro-steps.
- Progress Tracking — Per goal, step completion persists across sessions and devices.
- Step-Conditional SERP — Per query, results are matched to the active micro-step.
Technical Foundation
Technical Foundation
The patent specifies goal capture, decomposition templates, step tracking, active-step inference, step-conditional ranking, and cross-session persistence.
- Goal Capture And Inference — Per user, goals are entered explicitly or inferred from clustered queries that the system recognizes as belonging to a larger goal.
- Decomposition Templates — Per goal type, templates encode the canonical sequence of micro-steps with optional, required, and parallel step relationships.
- Step Persistence Store — Per goal, step status, partial annotations, and intermediate artifacts persist across sessions and devices.
- Active Step Inference — Per session, the current query and recent activity are matched against the decomposition to identify the active step.
- Step-Conditional Ranker — Per query, the ranking function reads the active step as input alongside the query string.
- Next-Step Recommender — Per goal, completed steps unlock recommended next steps drawn from the decomposition template.
The Process
The Process
From a captured or inferred goal, the system decomposes it into micro-steps, tracks step completion across sessions, infers the active step at query time, and ranks results for the active step.
- Identify The Goal — Per user, the higher-level goal is captured explicitly or inferred from a query cluster.
- Apply The Decomposition Template — Per goal type, the template generates a sequence of micro-steps.
- Persist The Step List — Per goal, the step list and progress status persist in a goal store across sessions.
- Receive A Query Inside The Goal — Per query, the system reads which goal context the query falls under.
- Infer The Active Step — Per query, matching against the decomposition identifies the active micro-step.
- Rank For The Active Step — Per query, the ranker promotes results that fit the active step.
- Update Progress And Suggest Next — Per session, step status updates and the next open step is surfaced for the user.
Quality Control
Quality Control
Step inference can mis-target the active step, decomposition can be too rigid for novel goals, and resumption can re-trigger steps the user already finished. The patent specifies safeguards to keep the decomposition useful.
- Decomposition Confidence — Per goal, when the template match confidence is low, the system falls back to generic task-aware ranking instead of forcing a decomposition.
- Step Inference Confidence — Per query, when active-step inference is uncertain, the ranker hedges between multiple candidate steps rather than committing to one.
- User Override — Per goal, the user can mark steps complete, skip steps, reorder steps, or replace the decomposition with a custom one.
- Template Adaptation — Per goal type, templates adapt based on aggregated outcomes so common variations update the canonical decomposition.
- Time-Decay Of Goals — Per goal, if no activity arrives for a long span, the goal is set aside so stale goals do not contaminate new queries.
Real-World Application
Microtask decomposition powers task-resumption features on web search and assistant surfaces. A user investigating a tool migration sees the comparison sub-step content during their lunch break, the pricing sub-step content during a later session, and the migration-guide sub-step content during a third session. The system holds the goal state and serves the active step each time.
- Goal-level Intent Granularity — The unit of intent is the goal, with the active step as the rendering context.
- Template-driven Decomposition — Goal types map to known decomposition templates.
- Step-conditional Ranking Output — Results are matched to the active micro-step within the goal.
Why Long Goals Lose To Decomposition
Per session, users do not have the attention to complete a multi-week goal in one sitting. Content that addresses the full goal in one monolithic page loses to content that addresses the active micro-step the user is on right now.
Why Resumption Is The Hidden Loyalty Lever
Per returning user, the property that surfaces their open step when they return earns the next click. Properties that restart the user from scratch lose the user to whichever property tracks their progress.
<\/section>What This Means for SEO
What This Means for SEO
Microtask decomposition means the engine does not rank for goals; it ranks for active micro-steps inside goals. Strategy has to publish content that matches the micro-step the user is currently working on, with explicit progression to the next step.
- Map The Decomposition, Not Just The Topic — For each major goal your audience pursues, list the micro-steps in order. Publish a page per micro-step rather than one monolithic page per goal. The engine increasingly routes step-level queries to step-fit pages.
- Step Headings Should Mirror The Decomposition — If the decomposition has steps like 'compare options', 'check pricing', 'plan migration', 'train the team', then page titles and H1s should mirror those step phrases. The engine matches queries to step language.
- Sequence Internal Links By Step Order — The page that serves step three should link to the page that serves step four. Internal navigation encodes the decomposition the engine is already trying to read across the user population.
- Make Resumption Easy For Returning Users — Returning visitors are mid-goal. A persistent navigation that surfaces 'where you left off' or a progress indicator across pages captures the resumption click the engine is also trying to serve.
- Each Micro-Step Page Should Stand Alone — Users land on step pages from search without prior context. The page must briefly orient the user to where the step sits in the larger goal, then deliver the step's content in full so it functions independently.
- Cover The Whole Decomposition, Not Just The Easy Steps — High-volume head queries map to the popular steps. Long-tail volume hides in the unglamorous steps. The site that covers every step of the decomposition earns the goal-spanning audience and the cross-step engagement signal.
- Goal-Spanning Engagement Compounds Authority — Users who complete multiple steps on the same domain accumulate the cross-page engagement that the engine reads as goal-completion authority. Covering the full decomposition is not content volume for its own sake. It is the path to being the goal-completion destination the engine learns to surface across the arc.