Imagine asking, “What is our cost of acquisition per channel?” Or, “Are high-usage customers more likely to stay on paid plans?” And just literally getting the answer. Let’s build a data analyst agent.

Creating a data analyst agent is surprisingly simple. We’ll start by connecting our database. I’m going to use BigQuery, but you can use Snowflake, Salesforce, Airtable, whatever you consider a database. Then I’ll authenticate it, and that’s it.

I can already ask my agent questions like, “What is our MRR growth over the last three months?” And our Gumloop agent will start exploring and analyzing our database. AI is exceptional at this type of work. And come back to us with an answer. This is a data analyst agent, and it’s ready to go.

Now, it’s not the most efficient data analyst agent. If I look at the tool calls between my question and its answer, the agent is spending a lot of time and tokens and effort figuring out our database. It’s listing all the tables, the schemas. It’s going to do that in every conversation because agents have no memory. We can fix this by creating a skill, a reusable set of instructions so the agent knows what it’s looking at, simply by asking, “Write me a skill about our BigQuery setup.” It’ll explore the tables, relationships, and write itself a guidebook on what it understood, which it can reuse in the next question or any agent can reuse.

Here’s that same question again. “What is our MRR growth over the last three months?” Notice how much more efficient it is. It started by reading its skill and instantly knew where to go.

Two more tips to getting the most out of your data analyst agent. First, give it more tools. Let’s give it Google Sheets so it can spit analysis out to a sheet or compare a forecast that lives in a spreadsheet to actual data in our database. Same with email, so it can automate reports for you with something like, “Every Friday at 9 a.m., send me a report of our revenue over the last week broken down per channel.”

Final tip, bring your agent into Slack so everyone can finally start asking the questions they’ve always wanted to. In fact, not ask questions, but get answers.