@inproceedings{tao-etal-2026-structured,
title = "Structured Strategy Injection for Data Analysis Agents",
author = "Tao, Zifu and
Mo, Guozhao and
Zhou, Weixiang and
Lu, Yaojie and
Lin, Hongyu and
He, Ben and
Han, Xianpei and
Sun, Le",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1279/",
pages = "27100--27123",
ISBN = "979-8-89176-462-0",
abstract = "Data analysis agents often produce shallow or non-auditable summaries because they lack the explicit reasoning frameworks and validation habits used by domain experts, especially when evidence must be gathered across heterogeneous data sources. We propose structured strategy injection, which injects compact methodology specifications with explicit checkpoints into an agent{'}s controller context, turning expert workflows into executable plans and enabling compliance-based transparency. To make such methodology-driven planning executable and auditable in heterogeneous settings, we introduce Data2MCP, a router-based orchestration infrastructure that standardizes SQL databases, knowledge graphs, vector stores, and tables as MCP-compatible tools and routes queries dynamically for multi-source retrieval and join-like reasoning. We compare Data2MCP against a RAG baseline and DA-Agent across multi-source retrieval, DAComp, and BIRD MiniDev. In multi-source retrieval, Data2MCP outperforms DA-Agent, and strategy injection benefits both architectures. On DAComp, strategy injection improves DA-Agent from 50.84 to 57.17 (+6.33). The code is available at https://github.com/turambar928/Data2MCP."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tao-etal-2026-structured">
<titleInfo>
<title>Structured Strategy Injection for Data Analysis Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zifu</namePart>
<namePart type="family">Tao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guozhao</namePart>
<namePart type="family">Mo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weixiang</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yaojie</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongyu</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ben</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xianpei</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Le</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Data analysis agents often produce shallow or non-auditable summaries because they lack the explicit reasoning frameworks and validation habits used by domain experts, especially when evidence must be gathered across heterogeneous data sources. We propose structured strategy injection, which injects compact methodology specifications with explicit checkpoints into an agent’s controller context, turning expert workflows into executable plans and enabling compliance-based transparency. To make such methodology-driven planning executable and auditable in heterogeneous settings, we introduce Data2MCP, a router-based orchestration infrastructure that standardizes SQL databases, knowledge graphs, vector stores, and tables as MCP-compatible tools and routes queries dynamically for multi-source retrieval and join-like reasoning. We compare Data2MCP against a RAG baseline and DA-Agent across multi-source retrieval, DAComp, and BIRD MiniDev. In multi-source retrieval, Data2MCP outperforms DA-Agent, and strategy injection benefits both architectures. On DAComp, strategy injection improves DA-Agent from 50.84 to 57.17 (+6.33). The code is available at https://github.com/turambar928/Data2MCP.</abstract>
<identifier type="citekey">tao-etal-2026-structured</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1279/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>27100</start>
<end>27123</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Structured Strategy Injection for Data Analysis Agents
%A Tao, Zifu
%A Mo, Guozhao
%A Zhou, Weixiang
%A Lu, Yaojie
%A Lin, Hongyu
%A He, Ben
%A Han, Xianpei
%A Sun, Le
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F tao-etal-2026-structured
%X Data analysis agents often produce shallow or non-auditable summaries because they lack the explicit reasoning frameworks and validation habits used by domain experts, especially when evidence must be gathered across heterogeneous data sources. We propose structured strategy injection, which injects compact methodology specifications with explicit checkpoints into an agent’s controller context, turning expert workflows into executable plans and enabling compliance-based transparency. To make such methodology-driven planning executable and auditable in heterogeneous settings, we introduce Data2MCP, a router-based orchestration infrastructure that standardizes SQL databases, knowledge graphs, vector stores, and tables as MCP-compatible tools and routes queries dynamically for multi-source retrieval and join-like reasoning. We compare Data2MCP against a RAG baseline and DA-Agent across multi-source retrieval, DAComp, and BIRD MiniDev. In multi-source retrieval, Data2MCP outperforms DA-Agent, and strategy injection benefits both architectures. On DAComp, strategy injection improves DA-Agent from 50.84 to 57.17 (+6.33). The code is available at https://github.com/turambar928/Data2MCP.
%U https://aclanthology.org/2026.findings-emnlp.1279/
%P 27100-27123
Markdown (Informal)
[Structured Strategy Injection for Data Analysis Agents](https://aclanthology.org/2026.findings-emnlp.1279/) (Tao et al., Findings 2026)
ACL
- Zifu Tao, Guozhao Mo, Weixiang Zhou, Yaojie Lu, Hongyu Lin, Ben He, Xianpei Han, and Le Sun. 2026. Structured Strategy Injection for Data Analysis Agents. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 27100–27123, Budapest, Hungary. Association for Computational Linguistics.