Act-On AI Data Analyst Best Practices

  • Updated

How to ask better questions and get more useful answers from your marketing data

The Act-On AI Data Analyst is a conversational AI that lets marketers query their Act-On data using plain language. Instead of navigating the report builder, you type a question and get an immediate, visualized answer drawn from your live data.

This guide explains how to phrase your questions for best results. The AI Data Analyst is designed to understand marketing terminology and Act-On's data model, but a few simple practices will help you get to the right answer faster.

To use the AI Data Analyst, type your question into the prompt at the top of the Analytics page or any specific Liveboard page.

Why this matters: Good prompts are the difference between a usable answer and a plausible-looking wrong one. The AI Data Analyst is fast, but it can only be as accurate as the question you ask - and a slightly off answer can quietly get baked into dashboards and reporting. Thirty seconds spent framing the question well is cheaper than catching the mistake downstream.

How the AI Data Analyst Interprets Your Questions

When you submit a prompt, the AI Data Analyst translates your natural language into a structured query against your Act-On data. It uses context clues in your phrasing to determine:

  • Which data model to query (Email Performance, Contacts, Forms, Webinars, etc.)
  • Which metrics to calculate (opens, clicks, sent counts, bounce rates, etc.)
  • Which dimensions to group by (program, message, date, segment, industry, etc.)
  • Which filters to apply (date ranges, specific programs, contact fields, etc.)

The more specific and grounded in Act-On terminology your prompt is, the more accurately the AI Data Analyst can translate it into results.

Tip: Use Act-On terminology. Terms like 'sent', 'open', 'click-through rate', 'Automated Program', 'message', 'segment', and 'lead score' map directly to Act-On's data model. Using familiar Act-On language helps the AI Data Analyst resolve ambiguity faster.

Writing Clear, Effective Prompts

Be specific about what you want to measure

Vague questions produce vague results. Include the metric, dimension, and time frame when possible.

Example PromptWhy It Works
Show me email performance for this yearThe AI is trained to understand what key metrics contribute to email performance and will show you overall results.
What was the click-through rate for each email in Q1 2025?Clear metric (CTR), dimension (email), and period (Q1 2025)
Which automated programs had the highest open rates last month?Specific metric, specific asset type, relative time reference
How many net new contacts were added per week in Q1?Specific measure with appropriate time granularity

Name the program, message, or segment when you know it

Referring to a specific Act-On asset by name scopes the query and eliminates ambiguity across your database.

Example PromptWhy It Works
Show performance for the Onboarding Nurture programNames the program directly so no guessing is needed
What is the click rate for the March Newsletter send?Pins to a single message, not all sends
How many contacts engaged with the Q2 New Features campaign?Queries a named campaign rather than all assets

Specify date ranges explicitly

The AI Data Analyst handles relative expressions like 'last quarter' and 'this month', but explicit ranges eliminate ambiguity when comparing periods. If you don't specify a date, you may get a large dataset of results that can take a long time to load.

Example PromptWhy It Works
Compare open rates in January vs February 2025Two explicit months for a clean comparison
Show daily sends from March 1 to March 31Exact bounds for a trend chart
What was my contact growth last quarter?Relative expression - works well for quick checks
How did click rates change week over week in Q4 2024?Specific quarter with weekly granularity

Specify how you want results grouped

Tell the AI Data Analyst which dimension to break results down by. This prevents it from collapsing all data into a single aggregate when you want a breakdown.

Example PromptWhy It Works
Show open rate by message subject line for the Welcome SeriesGroups by a specific dimension within a named program
Break down form submissions by landing page last monthDimension (landing page) and metric (submissions) are both clear
Show click-through rate by industry segment for the product launch campaignCombines campaign scope with a contact-level dimension
What is the average lead score by contact lead source?Scores a key metric across a meaningful dimension

Use Auto Mode for Cross-Model Queries

Auto Mode lets the AI Data Analyst automatically select the most relevant data model for your question. It also supports cross-model queries, allowing you to pull from multiple data sources in a single prompt - for example, combining email engagement data with form submission data.

Example PromptWhy It Works
Show contacts who attended a webinar in Q1 and have not opened an email in the past 60 daysCombines event data with email engagement data across models
Show contacts in the Enterprise segment with more than 10 form submissions but no media downloadsCross-model: forms, media and contact data combined
Which automated programs are driving the most SMS replies and email clicks?Bridges program engagement with multiple engagement types

Auto Mode tips

  • Auto Mode works best when the overlap between available data models is low.
  • If your question spans models with overlapping fields (e.g., contacts appear in both email and scoring models), specify which model you want to query to improve accuracy.
  • Ask the AI Data Analyst to describe the available data models if you're unsure which one to use: "What data models do you have access to?"
  • Manual model selection improves accuracy when models share common dimensions like contact ID, segment, or date.

Use Deep Dive Mode for Complex Analysis

Deep Dive Mode is designed for multidimensional investigations that would take multiple manual queries to answer. Instead of asking a series of individual questions, you describe a high-level goal and the AI Data Analyst proposes a structured analysis plan for you to review and approve before running.

How to use Deep Dive Mode

  1. Ask a high-level analytical question - describe the outcome you want to understand, not the specific query.
  2. Review the proposed analysis plan. The AI Data Analyst will outline the steps it intends to take.
  3. Add or remove steps as needed. Tailor the plan to focus on what matters most.
  4. Approve the plan to begin the analysis.
  5. Receive a comprehensive report with findings, visualizations, and a summary narrative.

When to use Deep Dive Mode: Deep Dive Mode works best for:

  • Quarterly marketing performance reviews
  • Root cause analysis (e.g., why did open rates drop in March?)
  • Multidimensional investigations spanning programs, segments, and time periods
  • Campaign post-mortems comparing multiple sends and conversion paths
Example PromptWhy It Works
Analyze email engagement trends across all nurture programs in H1 2025 and identify risk factors for low performanceHigh-level goal with scope - ideal for Deep Dive Mode
Investigate why form submission rates dropped in Q2 compared to Q1Root cause framing that benefits from a multi-step plan
Give me a comprehensive view of the enterprise segment: database growth, engagement health, and scoring trends over the past 6 monthsMultidimensional - spans contacts, email, and scoring models

Deep Dive Mode tips

  • Always review the analysis plan before approving. Add specific dimensions or comparisons you need.
  • Remove irrelevant steps to save time - fewer steps means faster results.
  • Do not close the tab while Deep Dive Mode is running to ensure the analysis completes.
  • Expect longer processing time; thorough multi-step analysis takes more time than a single query.

Brainstorm with the AI Data Analyst

You don't need to know exactly what to ask. The AI Data Analyst can suggest relevant metrics, questions, and analysis angles when you provide context about your role or current focus area. Use this to discover questions you hadn't thought to ask.

Example PromptWhy It Works
I manage email marketing for a SaaS company. What metrics should I be tracking to evaluate program health?Role context produces a tailored list of relevant KPIs
I'm preparing for our Q2 marketing review. What questions should I be able to answer about our nurture programs?Meeting context surfaces the right analysis agenda
I'm trying to understand why our lead scoring isn't working well. What should I look at?Problem statement prompts diagnostic question suggestions
I own lifecycle marketing for our financial services vertical. What engagement trends should I monitor?Industry + role context generates focused recommendations
Tip: Provide role and context for better suggestions. The more context you give - your role, the vertical you support, the business question you're trying to answer - the more relevant the AI Data Analyst's suggestions will be. Think of it as briefing a new analyst before handing them a project.

Verify and Review Results

Review the reasoning behind an answer

After the AI Data Analyst returns a result, you can expand the reasoning section to see exactly how it interpreted your question. Reviewing this before acting on numbers is a good habit, especially for unfamiliar queries.

  1. Expand the reasoning section in the response.
  2. Check how the AI Data Analyst interpreted your question and which metric definitions it applied.
  3. Verify the data sources and models that were used.
  4. Review any key decisions made during analysis (e.g., how it handled date ranges or calculated a rate).

If you're unsure about a result

Ask the AI Data Analyst directly to explain or verify its work. You can use natural language to challenge any part of the response.

Example PromptWhy It Works
How did you calculate that click-through rate?Forces the AI to surface its formula and denominator choice
Why did you exclude those records from the count?Reveals any implicit filters applied during analysis
Double-check this calculationTriggers a re-verification pass on the most recent result
What metric definition did you use for open rate?Confirms whether unique or total opens were used as the numerator
Can you explain your analysis step by step?Full walkthrough of reasoning for complex or surprising results

Identify when a response isn't coming from your data

The AI Data Analyst can also answer general marketing questions from its training knowledge - not just from your Act-On data. While this can be useful for definitions or benchmarks, it's important to recognize when a response is general knowledge versus a calculation from your actual data.

Signs that a response may not be drawn from your data:

  • The answer sounds generic, with no specific numbers or date ranges
  • The response uses industry-average benchmarks rather than your account's metrics
  • The reasoning block shows no data queries were executed
  • The answer doesn't reference any of your specific programs, segments, or messages
To get data-grounded answers: Be more specific in your prompt - reference a named program, segment, date range, or metric. Ambiguous questions like "What's a good click rate?" will draw on general knowledge. Questions like "What was the click rate for the Q1 Demand Gen program?" will query your data.

Refining Your Results

The AI Data Analyst supports follow-up questions in the same conversation. You don't need to re-state your full question each time - you can narrow, expand, or pivot based on the answer you received.

Conversational follow-up examples

  • Now filter that to only include contacts in the Financial Services industry.
  • Break this down by week instead of month.
  • Show the same data but for Q4 2024 instead.
  • Add unsubscribe rate to this table.
  • Show only programs where click rate is below 1%.

Adjusting chart types

You can ask the AI Data Analyst to change the visualization format for any result.

Example PromptWhy It Works
Show this as a bar chart insteadSwitches from table to visual format
Display this as a trend line over timeReframes the view for time-series analysis
Give me this as a table I can downloadReturns a tabular format for export

Prompt Examples by Marketing Use Case

Email Performance

Use these prompts to analyze send-level and program-level email engagement.

Example PromptWhy It Works
What are the top 10 emails by click-through rate in the last 90 days?Ranks emails by a specific metric within a rolling window
Show send count, opens, and clicks for every message in the Q2 Demand Gen programFull engagement summary scoped to one program
What percentage of sends to the Healthcare segment bounced in 2024?Combines segment filter with bounce rate calculation
How did our opt out rate trend month over month in 2024?Tracks a deliverability health metric over time
Which email subject lines had the highest open rates last quarter?Useful for subject line A/B learnings

Automated Program & Nurture Analytics

Understand how contacts engage with your nurture workflows.

Example PromptWhy It Works
How many contacts were sent messages from the Trial Conversion nurture in Q1 vs Q2?Entry volume comparison across two periods
Which nurture programs have the highest email engagement in the last 60 days?Ranks programs to identify top performers

Contact & Segment Analysis

Analyze your database composition, growth, and quality.

Example PromptWhy It Works
How many contacts have been created month over month this year?Database growth trend for capacity planning
What percentage of my contacts have opened at least one email in the past 90 days?Active database health check
Show contact counts by industry for contacts added in the last 6 monthsNew contact acquisition breakdown by vertical
How many contacts have a lead score above 75 and are in the Evaluation stage?Sales-ready contact count for pipeline alignment

Form & Landing Page Conversions

Measure how effectively your content and forms are capturing leads.

Example PromptWhy It Works
Which landing pages had the most form submissions last month?Ranks conversion assets by volume
What is the submission rate for the Request a Demo form over the past quarter?Conversion rate for a specific high-value form
Show form submissions by source for the Webinar Registration page in MarchTraffic source attribution for a specific event

Webinar & Event Engagement

Measure attendance, engagement, and follow-through for virtual events.

Example PromptWhy It Works
How many registrants attended the March Product Webinar live vs on-demand?Attendance split for a specific event
What was the average attendance rate across all webinars in Q1?Benchmark across your event program
Which webinar topics had the highest attendee-to-registrant conversion rate?Identifies high-interest content themes
Show email open rates for post-webinar follow-up sends in Q1Measures effectiveness of event follow-up sequences

Lead Scoring & Lifecycle

Track how contacts progress through your scoring model and lifecycle stages.

Example PromptWhy It Works
What behaviors are most commonly associated with contacts who reach a lead score above 80?Behavioral pattern analysis for scoring refinement
Show average lead score by segment for contacts added in the last 90 daysNew contact quality by acquisition segment

What to Avoid

Avoid overly open-ended questions without context

Very broad questions without a scope can return too much data or an incorrect aggregation. Narrow the topic before asking.

Less effective → More effective

Instead of: "How are my campaigns doing?"

Try: "What are the click-through rates for email campaigns sent in Q1 2025?"

Instead of: "Tell me about my contacts."

Try: "How many contacts were added each month in 2024, broken down by lead source?"

Avoid mixing multiple unrelated questions in one prompt

Ask one focused question at a time. Chaining unrelated metrics into one prompt can produce partial or blended results.

Less effective → More effective

Instead of: "Show email CTR, form submissions, and lead score distribution for Q1."

Try asking these as three separate questions in the same conversation:

  1. Show email click-through rates for Q1 2025.
  2. How many form submissions did we receive in Q1?
  3. What is the lead score distribution across my active contacts?

Understand metric definitions before asking

Act-On metrics have specific definitions. Being aware of these helps you phrase your question correctly:

TermAct-On Definition
Open RateUnique opens divided by delivered messages (not total sends). Bounced sends are excluded from the denominator.
Click-Through Rate (CTR)Unique clicks divided by delivered messages. Different from click-to-open rate (CTOR), which divides clicks by opens.
Click-to-Open Rate (CTOR)Unique clicks divided by unique opens. Measures engagement quality among openers.
Bounce RateHard and soft bounces combined as a percentage of total sends.
SendA single outbound email delivery attempt to one contact address.
Automated ProgramA multi-step workflow in Act-On that triggers a series of emails and actions based on contact behavior or data.
Segment

In the context of Email performance: A dynamic or static recipient list of contacts matching defined criteria within Act-On.

Also, Contact Segment and Contact Subsegment are standard fields on the Contact model that could represent your overall market segments.

Lead ScoreA numeric value assigned to a contact based on behavioral and demographic scoring rules configured in Act-On.

When Results Don't Look Right

If the AI Data Analyst returns an unexpected result, try these approaches:

  • Rephrase with more specific language. Include the metric name, data model, and date range explicitly.
  • Check the query it generated. The AI Data Analyst shows the underlying query it used - review it to see whether it interpreted your question as intended.
  • Scope to a known asset. Narrow to a single program or message you know well and verify the numbers match what you see in standard Act-On reports.
  • Use the follow-up conversation. Ask "Why did you calculate it that way?" or "What metric definition did you use for click rate?" to understand the interpretation.
  • Report unexpected results. If the AI Data Analyst consistently misinterprets a question, share feedback so the underlying coaching can be improved.
Data freshness note: The AI Data Analyst queries data that is synced from Act-On on a scheduled basis. Very recent activity (within the last few hours) may not yet be reflected. For real-time send-level data, use the standard Act-On reporting interface.

Was this article helpful?

Have more questions? Submit a request