Data Agent

The Data Agent is a powerful feature in Assist 2.0 that enables deterministic, code-based data analysis. Unlike standard AI responses, which can produce inconsistent results on numerical work, the Data Agent uses Python code executed in a sandbox to perform precise calculations — so your analysis is accurate and reproducible every time.

What is the Data Agent?

The Data Agent allows you to:

Perform deterministic calculations using Python code instead of relying on AI interpretation
Eliminate AI hallucinations when working with numerical data and complex analysis
Get consistent, reproducible results every time you run an analysis
Analyse structured data from CSV files, spreadsheets, and other data sources

When to use the Data Agent

Use the Data Agent when you need:

Precise mathematical calculations (ROI, ROAS, conversion rates, etc.)
Data aggregation and statistical analysis
Performance rankings and comparisons
Budget or resource allocation recommendations based on metrics
Any analysis where accuracy is critical

Typical use cases include marketing performance analysis, sales reporting, financial summaries, customer metrics breakdowns, and any other workflow where the underlying numbers need to be exact.

How the Data Agent fits into a template

A Data Agent doesn't live on its own — it's wired into a template and agent that together define:

Welcome message: What end-users see and what data they should upload
Agent instructions: The agent's role, capabilities, and the specific triggers that tell it to invoke the Data Agent (e.g. "when the user uploads data," "when calculations are required")
Output format: How results are presented back to the user in plain language
The Data Agent itself: The Python code that runs in the sandbox to do the actual analysis

When a user uploads a file and asks for analysis, the agent recognises the trigger, calls the Data Agent, and the code runs deterministically against the uploaded data. Because it runs in a sandbox, the same input will always produce the same output.

Creating a Data Agent

In Assist 2.0 you don't need to hand-write Python or paste it into a template field. You describe what you want in chat and the system builds the Data Agent, template, and agent for you.

Step 1: Prepare sample data

Create a sample CSV or spreadsheet with the structure you want to analyse. Include realistic column names and a few rows of representative data — the system uses this to understand the shape of your data and test the Data Agent against it.

Step 2: Start a new chat

Open a new chat in your workspace
Click the attachment button (+ icon)
Upload your sample data file
Wait for the file to attach

Step 3: Describe what you want

Type a prompt describing the analysis you need. Be specific about:

What calculations you need
Which metrics to analyse
What recommendations or insights you want surfaced

Assist 2.0 will plan the work, build the Data Agent, test it against your sample file, and wire it up to a reusable template and agent.

Note: If you'd rather review the plan before anything is created, switch the chat into Plan Mode before sending your message. You'll get a proposed plan you can revise and accept.

Step 4: Review and save

Check the generated Data Agent against your requirements
Run it once with your sample file to confirm the output looks right
The template and agent are saved into your workspace automatically — you'll find them under Library → Templates and Library → Agents

Heads up: Building a new Data Agent may be restricted to admin users in your workspace. If you don't see the option, ask your Assist administrator.

Using an existing Data Agent template

If a Data Agent template already exists in your workspace (built by you, a teammate, or installed from the Marketplace):

Open the Library from the left nav
Go to the Templates tab
Find the template you want to use
Click into it to start a chat, or open it in edit mode to see how it's configured

Tip: If the template you want isn't in your workspace, check the Marketplace tab — you may need to install it first.

Example: Campaign Performance Analyzer

To make this concrete, here's one example of a Data Agent template — the Campaign Performance Analyzer, which analyses marketing campaign data and provides budget recommendations.

What the user uploads

A CSV or spreadsheet containing:

Campaign names — the name or identifier for each marketing campaign
Ad spend — total amount spent on each campaign
Revenue generated — total revenue attributed to each campaign
Optional metrics — impressions, clicks, conversions, or other performance data

What the Data Agent does

Reads and processes the uploaded data file
Calculates key metrics (ROAS, total spend, total revenue)
Identifies top and bottom performers
Generates budget reallocation recommendations
Returns structured data for the agent to explain in plain language

What the output looks like

The agent returns a report in three sections:

RESULTS SUMMARY

Total Campaign Spend: £18,900
Total Revenue Generated: £45,230
Overall ROAS: 2.39

KEY INSIGHTS

Top performers:

Email Newsletter: ROAS 7.00 (£400 spend → £2,800 revenue)
Google Search - Brand: ROAS 4.20 (£2,500 spend → £10,500 revenue)
Meta - Retargeting: ROAS 4.20 (£1,800 spend → £7,560 revenue)

Underperformers:

TikTok Awareness: ROAS 0.60 (£1,500 spend → £900 revenue, net loss £600)
LinkedIn - B2B Leads: ROAS 1.01 (£2,800 spend → £2,828 revenue)
Meta - Prospecting: ROAS 1.20 (£3,500 spend → £4,200 revenue)

ACTIONABLE RECOMMENDATIONS

Increase budget for top performers:

Email Newsletter: £400 → £500 (+£100)
Google Search - Brand: £2,500 → £3,125 (+£625)
Meta - Retargeting: £1,800 → £2,250 (+£450)

Reduce budget for underperformers:

TikTok Awareness: £1,500 → £750 (-£750)
LinkedIn - B2B Leads: £2,800 → £1,400 (-£1,400)
Meta - Prospecting: £3,500 → £1,750 (-£1,750)

Financial impact:

New total budget: £16,175 (down from £18,900)
Capital freed for reinvestment: £2,725

The same pattern — structured upload, deterministic calculation, plain-language summary with recommendations — works for any domain. Sales pipeline analysis, customer churn breakdowns, finance reports, survey results: if it's tabular and needs accurate maths, a Data Agent can handle it.

Best practices

For template creators

Use clear, descriptive welcome messages that tell users exactly what data format to upload
Define specific triggers in the agent instructions for when to call the Data Agent
Test your Data Agent with sample data before deploying the template
Structure your output into clear sections for easy reading
Include enough context in the agent instructions so the agent can explain results in plain language

For template users

Prepare your data in the correct format (CSV or spreadsheet)
Include all required columns as specified in the template instructions
Use consistent data formatting (same currency, date formats, etc.)
Review the results to ensure they make sense for your business context
Take action on the recommendations provided

Troubleshooting

Issue: The Data Agent doesn't activate

Ensure the agent instructions clearly specify when to call the Data Agent. Check that you've uploaded data in the expected format and column names.

Issue: Python code errors

Verify your data file has the correct column names. Ensure there are no missing values or formatting issues. Try the Data Agent against the original sample data first to isolate whether it's a data or code issue.

Issue: Results don't match expectations

Review the Data Agent's calculation logic to understand how figures are produced. Check your input data for accuracy and confirm the code is reading the correct columns.

Key takeaways

✅ The Data Agent provides deterministic analysis using Python code instead of AI interpretation
✅ Eliminates AI hallucinations for numerical and data-driven tasks
✅ Works for any tabular analysis use case — marketing, sales, finance, customer metrics, and more
✅ You can build your own Data Agent by describing what you want in chat with a sample file attached
✅ Structured outputs make it easy to understand and act on insights
✅ Reproducible results ensure consistency across multiple analyses

Next steps

Browse the Library → Templates tab to see if a Data Agent template already exists for your use case
Create a custom Data Agent for your own analysis needs by uploading sample data and describing what you want
Share useful Data Agent templates with your team via the workspace Library

For more information about creating custom templates or advanced Data Agent configurations, contact your Assist administrator.