Filly is a world-class SQL engineer, but isn't a mind reader; if you ask vague questions, you’re going to get boring answers.
The coolest part about Filament is that you can literally chat with your data. No code, no pivot tables, just a search bar. But we’ve noticed a pattern: most people approach that search bar with the same polite, vague hesitation they’d use with a stranger at a networking event. They ask things like, ‘How are our donations looking?’ and then get frustrated when Filly gives them a generic total that they already knew. We need to have a heart-to-heart: Filly is an AI assistant, and while it is incredibly fast, it has limited intuition about your specific mission. Think of Filly as a very literal, very intelligent intern (minus the career ambition).
The biggest hurdle to getting value out of Filly isn't your data literacy; it's your specificity. When you ask a vague question, the AI has to sift through your connected apps; Salesforce, Mailchimp, Google Analytics, and guess which table you’re actually interested in. To really get the most out of Filly, you need to understand the four-part harmony of data analysis: Measures, Metrics, Attributes, and Dimensions. Think of these as the building blocks of any question you could possibly ask. If your data is a library, the Measure is the total number of books, the Metric is the books borrowed per month, the Dimension is the Genre, and the Attribute is the Author’s name.
Here’s the breakdown of how these pieces actually fit together:
Measures are your raw, quantitative building blocks; the things that can be aggregated like ‘how much’ or ‘how many’. They are the basic counts or sums sitting in your database, like a total dollar amount of donations or the raw number of volunteer sign-ups. On their own, they tell you the scale of what happened, but they don't yet tell you if that result is good, bad, or indifferent.
Metrics take those raw measures and give them a job to do by adding context. A metric is usually a calculation, like taking your total donation measure and dividing it by the number of donors to get your average donation size, or tracking your sign-up count over time to find your monthly growth rate. This is where you move from just having numbers to having a pulse on your performance against a goal.
Dimensions are the qualitative buckets you use to slice and dice your measures and metrics. They provide the categories that allow you to organise your data into meaningful groups. These groupings have types, like geographical (country, city, region), time (year, quarter, month), or may be product related (campaign, team, individual). With dimensions, you can see exactly which specific segments of your mission are driving your results.
Attributes are the granular details about your measures, or that live inside your dimensions. While a dimension is the broad category (like supporter), an attribute is the specific characteristic of that supporter, like the supported details in Salesforce, or your tags in Raisely. Attributes are the fine-print details that allow you to filter your analysis down to a specific group, ensuring your outreach is targeted.
When you combine these into a prompt, you’re giving Filly a high-definition map to follow. Here is the basic structural formula for what makes a good prompt:
"Give me the [Metric/Measure] grouped by [Dimension] filtered by [Attribute] for [Timeframe] shown as a [Chart Type]."
If you stick to this structure, you stop being a user and start being a strategist. Here are three examples of how to put that into practice across your different data sources:
Donor Retention Deep-Dive: ‘Show me the average donation amount (Metric) grouped by State (Dimension) for all donors whose membership status (Attribute) is 'Active' during 2025 (Timeframe) as a Bar Chart.’
Email Engagement Audit: ‘Calculate the total click-to-open rate (Metric) grouped by Email Campaign Name (Dimension) for all campaigns with a 'Fundraising' tag (Attribute) sent in the last 90 days (Timeframe) as a Line Chart.’
Program Impact Snapshot: ‘Give me the total count of new volunteers (Measure) grouped by Source Channel (Dimension) where the Volunteer Interest (Attribute) is 'Environment' for Q4 (Timeframe) as a Pie Chart.’
By naming your measures and slicing them with dimensions and attributes, you’re providing clear direction to your intern (Filly), which we all know they need sometimes. You’re moving away from asking ‘What’s the vibe of our data?’ and moving toward knowing ‘What is the exact reality of our impact?’
The secret to mastering Filament is realising that your first prompt is just the start of the conversation. You don’t have to get the perfect chart in one shot. If Filly shows you something interesting but messy, say so. Refine the output; ‘That's too much noise, filter it to just the New South Wales region,’ or ‘Now change that to a line chart so I can see the trend.’ This iterative process is how real analysis happens. Remember, you (and Filly) aren't failing if the first chart isn't board-ready; you're just refining the brief.