Back office ops · Production

How Moveworks built Brief Me: an agentic document intelligence feature for the Moveworks Copilot

The problem

Without an AI assistant, synthesizing information across multiple enterprise documents for tasks like competitive comparisons requires manual searching, scanning, reading, and writing—a process that can take hours.

Workflow diagram · grounded in source
1
User uploads sources
trigger
“enabling employees to upload PDF, Word, and PPT files into chat and interact with the content inside”
2
Online ingestion pipeline
ai_action
“it fetches the sources. Next, it chunks the files into smaller segments using proprietary techniques. Then, it generates metadata from the sources. Following that, it embeds the chunks. Finally, it performs indexing. These steps—fetching…”
3
Query rewriting for multi-turn context
ai_action
“The Reasoning Engine examines up to n previous turns in the conversation history. It identifies and prioritizes the most relevant turns to generate a new query with full contextual awareness of the chat history. We utilize GPT-4o with in…”
4
Operation planning (SEARCH vs READ)
ai_action
“This module is the planning step to determine what kind of operations need to be taken over provided sources. The operation planner uses a combination of in-context learning (ICL) and Supervised Fine Tuning (SFT) methods (in progress) to…”
5
Hybrid search and chunk ranking
ai_action
“we use the following two methods to perform search: Embeddings-based retrieval of snippets relevant to the query. This has the advantage of higher recall and generalizability. Keyword-based retrieval of snippets relevant to the query. Th…”
6
Map-reduce parallel output generation
ai_action
“we introduced a novel map-reduce based algorithm to process contextual data”
7
Paragraph-level citation output
output
“Brief Me, however, is capable of providing citations at the paragraph level, enabling users to place greater trust in the generated responses. We employ a combination of LLM and heuristic-based methods, such as n-gram matching, to attrib…”
Reported outcome

Brief Me achieves 97.24% correct-action rate and 97.35% correct-resource rate in internal evaluations on 2,200 real-usage query-response pairs, and provides paragraph-level citations enabling users to place greater trust in generated responses.

Reported metrics
Action prediction accuracy97.24
Resource prediction accuracy97.35
Search query quality93.80
Response completeness97.98
Show all 11 reported metrics
Action prediction accuracy97.24
Resource prediction accuracy97.35
Search query quality93.80
Response completeness97.98
Response groundedness89.21
Retrieval Precision at 365.11
Ingestion pipeline P90 latency<10 seconds
Employee time saved on information synthesissave employees significant time
Manual information synthesis time (without AI)hours
Embedding model training dataset sizeApproximately 1 million query-document pairs
Evaluation dataset size2,200 query-response pairs
Reported stack
Moveworks CopilotBrief MeGPT-4oBM25MPNetOpenSearch
Source
https://www.moveworks.com/us/en/resources/blog/how-we-built-brief-me-for-productivity
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Brief Me achieves 97.24% correct-action rate and 97.35% correct-resource rate in internal evaluations on 2,200 real-usage query-response pairs, and provides paragraph-level citations enabling users to place greater tr…

What tools did this team use?

Moveworks Copilot, Brief Me, GPT-4o, BM25, MPNet, OpenSearch.

What results were reported?

Action prediction accuracy: 97.24; Resource prediction accuracy: 97.35; Search query quality: 93.80; Response completeness: 97.98 (source-reported, not independently verified).

How is this back office ops AI workflow structured?

User uploads sources → Online ingestion pipeline → Query rewriting for multi-turn context → Operation planning (SEARCH vs READ) → Hybrid search and chunk ranking → Map-reduce parallel output generation → Paragraph-level citation output.

Related back office ops cases