Fabric Data Agent performance is not the same via Copilot Studio agent

[kevinvu99]

Good Morning, 

Long time follower, first time poster. I have been self-teaching Fabric Data Agent (FDA) and Copilot Studio (CS) to complete a project. I have been fortunate in that I have been successful in connecting my CS agent (CSA) to my FDA. I am currently in the testing phase. 

Background: I am creating a chatbot (in CSA) that is an orchestrator that knows which FDA to route the user prompt to and get answers from the specific FDA. I gave my FDA specific instructions on how to respond to user prompts. For example, do step 1, 2, then 3. When I tested my FDA in Fabric, the agent performed very well and followed all three steps, >95% of the time. But when I tested the same FDA via my CSA, the CSA would follow steps 1 and 2 and maybe forget to perform step 3, and various other combinations of steps 1, 2, and 3 failing. The CSA results are very inconsistent, less than ~30% of the time. 

My question (and seeking knowledge from the community) is, have you seen different performance from your FDA versus your FDA via CSA? If so, can you share your experience and any possible tips I can try to get better performance from my CSA? I am aware of the new feature to publish FDA directly to the end users, but my goal with CSA is to be a one-stop-shop chat agent.

[YASAV9930]

<div class="pWvJNd" data-processed="true" data-complete="true">
<div class="mZJni Dn7Fzd" dir="ltr" data-container-id="main-col" data-xid="VpUvz" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQ3KYQegYIAQgAEAE" data-processed="true">
<div class="Y3BBE" data-sfc-cp="" data-hveid="CAEIARAA" data-processed="true" data-complete="true">Data science is a transformative, multidisciplinary field focused on extracting meaningful knowledge and insights from data to facilitate better decision-making and prediction. It is far more than just analysis; it's a blend of computer science, statistics, machine learning, and domain expertise.<button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="8be1f4a7-42b4-4e75-aea8-8effbb48feba" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_b_C5gNJc;class=qhXR1_b_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgBEAE"></button></div>
<div class="Y3BBE" data-sfc-cp="" data-hveid="CAEIAhAA" data-processed="true" data-complete="true">Here are some key thoughts and perspectives on data science:<button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="81a11b63-6dbc-4e8e-80be-5e25b77c2518" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_f_C5gNJc;class=qhXR1_f_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgCEAE"></button></div>
<div class="Fsg96" data-sfc-cp="" data-complete="true" data-processed="true">&nbsp;</div>
<div class="otQkpb" role="heading" aria-level="3" data-animation-nesting="" data-sfc-cp="" data-complete="true" data-processed="true" data-sae="">Core Philosophy<button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="79bb3c18-904e-4ea7-bb39-a79331f2ffcf" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_k_C5gNJc;class=qhXR1_k_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgDEAA"></button></div>
<ul class="KsbFXc U6u95" data-processed="true" data-complete="true">
<li data-hveid="CAEIBBAA" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Data as the New Raw Material</strong>: Data is widely considered a valuable asset that needs refining. Organizations leverage data to gain a competitive edge, improve efficiency, and innovate.</span></li>
<li data-hveid="CAEIBBAB" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">The Power of Evidence-Based Decisions</strong>: A fundamental principle is moving away from decisions based purely on intuition ("gut feeling") toward those supported by empirical evidence derived from data analysis.</span></li>
<li data-hveid="CAEIBBAC" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">The Scientific Method Applied to Data</strong>: At its core, data science is the application of the scientific method&mdash;forming hypotheses, designing experiments, collecting data, analyzing results, and iterating&mdash;within a data-intensive environment.</span></li>
<li data-hveid="CAEIBBAD" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Models are Useful Abstractions</strong>: As the statistician George E. P. Box famously stated, "All models are wrong, but some are useful." Data science models simplify complex realities to provide predictive power and useful approximations of the world.</span><button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="c8361c2b-a1b2-4458-bded-7a9ccd838e91" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_10_C5gNJc;class=qhXR1_10_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgEEAQ"></button></li>
</ul>
<div class="Fsg96" data-sfc-cp="" data-complete="true" data-processed="true">&nbsp;</div>
<div class="otQkpb" role="heading" aria-level="3" data-animation-nesting="" data-sfc-cp="" data-complete="true" data-processed="true" data-sae="">The Role of the Data Scientist<button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="3c9742a7-c0df-4c32-9f71-4a60190fcd3b" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_15_C5gNJc;class=qhXR1_15_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgFEAA"></button></div>
<ul class="KsbFXc U6u95" data-processed="true" data-complete="true">
<li data-hveid="CAEIBhAA" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">A "Full-Stack" Role</strong>: Data scientists often require a broad range of skills, encompassing the technical ability to program and manage databases, the analytical rigor of a statistician, and the communication skills of a storyteller to explain findings to non-technical stakeholders.</span></li>
<li data-hveid="CAEIBhAB" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Curiosity as the Driving Force</strong>: The most effective data scientists possess an innate curiosity. They constantly ask "why?" and are skeptical detectives, ensuring the integrity of the data and the validity of their conclusions.</span></li>
<li data-hveid="CAEIBhAC" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">The 80/20 Rule of Wrangling</strong>: A common truth in the field is that a significant majority of time (often cited as 80%) is spent on data preparation&mdash;cleaning, transforming, and organizing raw data before any meaningful modeling can begin.</span></li>
<li data-hveid="CAEIBhAD" data-sae="" data-complete="true"><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Impact Through Communication</strong>: Technical skills are only half the battle. The ability to translate complex algorithms and statistical results into a clear, actionable business narrative is arguably the most valuable skill a data scientist can possess.</span><button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="049582ab-24b0-4ef9-92ac-282861ff1e52" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_1l_C5gNJc;class=qhXR1_1l_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgGEAQ"></button></li>
</ul>
<div class="Fsg96" data-sfc-cp="" data-complete="true" data-processed="true">&nbsp;</div>
<div class="otQkpb" role="heading" aria-level="3" data-animation-nesting="" data-sfc-cp="" data-complete="true" data-processed="true" data-sae="">Impact and the Future Landscape<button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="2d732125-41b4-458a-adce-3a46cb3b1f57" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_1q_C5gNJc;class=qhXR1_1q_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgHEAA"></button></div>
<ul class="KsbFXc U6u95" data-processed="true" data-complete="true">
<li data-hveid="CAEICRAA" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Ubiquity and Pervasiveness</strong>: Data science is no longer confined to tech companies; it is transforming traditional sectors like healthcare, finance, agriculture, marketing, and government services.</span></li>
<li data-hveid="CAEICRAB" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">The Rise of AI and Machine Learning</strong>: The field is heavily intertwined with advances in Artificial Intelligence and Machine Learning, with increasing specialization in areas like Deep Learning, Natural Language Processing (NLP), and computer vision.</span></li>
<li data-hveid="CAEICRAC" data-complete="true" data-sae=""><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Ethical Imperatives</strong>: As algorithms drive more critical decisions in society&mdash;such as loan approvals, hiring decisions, and criminal justice&mdash;discussions around ethics, algorithmic bias, fairness, and accountability are becoming central to the practice of data science.</span></li>
<li data-hveid="CAEICRAD" data-sae="" data-complete="true"><span class="T286Pc" data-sfc-cp="" data-complete="true"><strong class="Yjhzub" data-complete="true">Continuous Learning</strong>: The data science landscape is constantly evolving with new tools (e.g., Python, R, SQL), frameworks, and cloud platforms (AWS, Google Cloud, Azure), requiring practitioners to be perpetual learners.</span><button class="rBl3me" tabindex="0" data-amic="true" data-icl-uuid="e4a9b183-db5b-48c2-96bc-bf98477a8652" aria-label="View related links" data-wiz-attrbind="disabled=qhXR1_2g_C5gNJc;class=qhXR1_2g_UpSNec;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQye0OegYIAQgJEAQ"></button></li>
</ul>
<div class="Fsg96" data-sfc-cp="" data-complete="true" data-processed="true">&nbsp;</div>
<div data-xid="Gd7Hsc" data-sfc-cp="" data-wiz-uids="qhXR1_2k" data-complete="true" data-processed="true">
<div class="DBd2Wb" data-complete="true" data-processed="true">
<div data-complete="true" data-processed="true">
<div data-complete="true" data-processed="true">&nbsp;</div>
</div>
<div class="zkL70c" data-complete="true" data-processed="true">
<div class="" data-complete="true" data-processed="true">
<div class="csTa2e" data-sfc-cp="" data-wiz-uids="qhXR1_2n,qhXR1_2m,qhXR1_2o,qhXR1_2p" data-signal-inputs="" data-complete="true" data-processed="true">
<div id="shrproxync41acHdMZyI4-EP-bi8sAg_1" class="bQ0Yzc" data-sfc-cp="" data-wiz-uids="qhXR1_2q,qhXR1_2r,qhXR1_2s,qhXR1_2t" data-signal-inputs="N7abZd=a5f0he/TVP8Qe;" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQp4QQegYIAQgKEAA" data-hveid="CAEIChAA" data-complete="true" data-processed="true"><button class="eGAasd" aria-label=" Share" data-processed="true"></button>
<div data-wiz-uids="qhXR1_2v" data-signal-inputs="iWO5td=WPf93c/TVP8Qe;fmcmS=WPf93c/Zxo6Vb" data-complete="true" data-processed="true">&nbsp;</div>
<div data-wiz-uids="qhXR1_2x" data-signal-inputs="iWO5td=WPf93c/moMghe;fmcmS=WPf93c/wQ7Xrd" data-complete="true" data-processed="true">&nbsp;</div>
</div>
</div>
</div>
<div class="VlQBpc" data-wiz-uids="qhXR1_2y,qhXR1_2z,qhXR1_30" data-complete="true" data-processed="true"><button class="ya9Iof" aria-label="Positive feedback" aria-pressed="false" data-snt="1" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQzO0OegYIAQgKEAY" data-hveid="CAEIChAG" data-processed="true"></button><button class="ya9Iof" aria-label="Negative feedback" aria-pressed="false" data-snt="-1" data-ved="2ahUKEwiBoN6_lKyRAxUcxDgGHXkcD4YQze0OegYIAQgKEAc" data-hveid="CAEIChAH" data-processed="true"></button>
<div data-wiz-uids="qhXR1_32,qhXR1_33,qhXR1_34" data-signal-inputs="QZcSBe=RjGxw/TVP8Qe;j0Ymx=RjGxw/Zxo6Vb" data-complete="true" data-processed="true">&nbsp;</div>
</div>
</div>
</div>
<div data-zzy="1" data-processed="true">&nbsp;</div>
</div>
<div class="Fsg96" data-sfc-cp="" data-complete="true" data-processed="true">&nbsp;</div>
<div data-subtree="aimba" data-complete="true" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="6" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="7" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="8" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="9" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="10" data-processed="true">&nbsp;</div>
</div>
<div class="SGF5Lb" data-processed="true">
<div data-container-id="11" data-processed="true">&nbsp;</div>
</div>
</div>
<div class="" data-processed="true" data-complete="true">&nbsp;</div>

[YASAV9930]

Data science is a transformative, multidisciplinary field focused on extracting meaningful knowledge and insights from data to facilitate better decision-making and prediction. It is far more than just analysis; it's a blend of computer science, statistics, machine learning, and domain expertise.

Here are some key thoughts and perspectives on data science:

Core Philosophy

• Data as the New Raw Material: Data is widely considered a valuable asset that needs refining. Organizations leverage data to gain a competitive edge, improve efficiency, and innovate.
• The Power of Evidence-Based Decisions: A fundamental principle is moving away from decisions based purely on intuition ("gut feeling") toward those supported by empirical evidence derived from data analysis.
• The Scientific Method Applied to Data: At its core, data science is the application of the scientific method—forming hypotheses, designing experiments, collecting data, analyzing results, and iterating—within a data-intensive environment.
• Models are Useful Abstractions: As the statistician George E. P. Box famously stated, "All models are wrong, but some are useful." Data science models simplify complex realities to provide predictive power and useful approximations of the world.

The Role of the Data Scientist

• A "Full-Stack" Role: Data scientists often require a broad range of skills, encompassing the technical ability to program and manage databases, the analytical rigor of a statistician, and the communication skills of a storyteller to explain findings to non-technical stakeholders.
• Curiosity as the Driving Force: The most effective data scientists possess an innate curiosity. They constantly ask "why?" and are skeptical detectives, ensuring the integrity of the data and the validity of their conclusions.
• The 80/20 Rule of Wrangling: A common truth in the field is that a significant majority of time (often cited as 80%) is spent on data preparation—cleaning, transforming, and organizing raw data before any meaningful modeling can begin.
• Impact Through Communication: Technical skills are only half the battle. The ability to translate complex algorithms and statistical results into a clear, actionable business narrative is arguably the most valuable skill a data scientist can possess.

Impact and the Future Landscape

• Ubiquity and Pervasiveness: Data science is no longer confined to tech companies; it is transforming traditional sectors like healthcare, finance, agriculture, marketing, and government services.
• The Rise of AI and Machine Learning: The field is heavily intertwined with advances in Artificial Intelligence and Machine Learning, with increasing specialization in areas like Deep Learning, Natural Language Processing (NLP), and computer vision.
• Ethical Imperatives: As algorithms drive more critical decisions in society—such as loan approvals, hiring decisions, and criminal justice—discussions around ethics, algorithmic bias, fairness, and accountability are becoming central to the practice of data science.
• Continuous Learning: The data science landscape is constantly evolving with new tools (e.g., Python, R, SQL), frameworks, and cloud platforms (AWS, Google Cloud, Azure), requiring practitioners to be perpetual learners.

[YASAV9930]

Data science is a multidisciplinary field focused on extracting knowledge and insights from data to support better decision-making. It combines elements of statistics, computer science, machine learning, and domain expertise to solve complex problems across various industries. 

[YASAV9930]

Data science is a multidisciplinary field focused on extracting knowledge and insights from data to support better decision-making. It combines elements of statistics, computer science, machine learning, and domain expertise to solve complex problems across various industries. 

[kevinvu99]

First, thank you for the reply. Some parts of my FDA instructions are numbered list and some parts are bulleted list. I have examples for some use cases, but not all. So for the most part, your suggestions are already in the system prompt. However, I will go through the instructions again and make them more structured where warranted.

Second, for clarification, comparing between structured output format (step1, step 2, step 3) versus using numbered list, which one would normally yield better results in your experience?

Third, for confidentiality, I can't be more specific, so I'll use this example. Let's say Step 3 is for the agent to "Always respond at the end with 'This is a generated AI response.'" In FDA, it would always respond with that statement, but in CSA, it would be a hit-and-miss.

[v-priyankata]

Hi @kevinvu99 

Thank you for reaching out to the Microsoft Fabric Forum Community.

A Fabric Data Agent can behave one way when you test it directly in Fabric, and then act differently when it’s called through Copilot Studio. The main reason is that Copilot Studio doesn’t pass the user’s message through verbatim, it often rewrites, shortens, or interprets the prompt before sending it to the Data Agent. When that happens, some of your critical instructions can get lost, which is why the agent might skip step 3 or produce inconsistent results.

To improve this, it helps to make your FDA instructions very strict and self-contained. Use a fixed, structured output format (for ex: Step 1 Step 2 Step 3) and include clear example prompts and example responses inside the FDA’s instruction block. That way, even if Copilot Studio modifies the user’s query, the Data Agent still understands and follows the full 3-step workflow.

If I misunderstand your needs or you still have problems on it, please feel free to let us know. 

Thanks.