Most lists come together through ordinary search, and that is the right starting point. But some audiences depend on a calculation a search box cannot do, like counting things within a group, comparing two numbers, working out a figure the data does not store directly, ratios, change over time, or the makeup of a team. The Advanced Dataset Creator handles those: you describe the audience in natural language, the agent does the calculation for you, shows you its plan in natural language, and builds the list only after you approve, so you get audiences a normal search cannot reach while staying in control the whole way. The Advanced Dataset Creator runs in the CLI, inside your AI assistant, so it is a command-line workflow rather than something you open in the web app.
Search and filters cover most list building and are faster for a straightforward audience, an exploratory list you are still shaping, or a one-off pull. Reach for the Advanced Dataset Creator when your audience depends on a count, compares two numbers, or needs a value the data does not store directly. The tell is simple: when you catch yourself trying to fake a calculation with a long chain of filters, switch to the Advanced Dataset Creator.
It can count or total something across a group, then keep only the groups that cross a threshold you set. For example, count the people in a role at each company and keep the companies with more than five. It can also pin a condition to an exact value or range, so a number lands exactly where you want it. Stating the threshold plainly helps the agent build the right thing the first time.
Some of the most useful audiences depend on a figure no single piece of the data holds, such as revenue per employee, the share of headcount in engineering, or the ratio of account executives to sales development reps. The agent works that figure out from the details the data does provide and adds it as a new column, which you can then filter, sort, and rank on like any other. A value you define once is reused every time you re-run the list.
It works through a simple back-and-forth. You describe the audience and its conditions, the agent shows you its plan in natural language, and the technical version underneath if you want it, and the list is built only after you approve. Read the plain-language plan first, because it usually tells you whether the logic is right without reading anything technical. The review step lets you catch a misunderstanding before a single record is built, and it gradually builds your sense of what the tool can do next time.
A list like this is worth keeping, because the data behind it keeps changing. Saving your work keeps the whole setup, including the conditions and any values you defined, so you can run it again later and get current results. Compare a fresh run with an earlier one to see what changed in your market, and give each saved list a clear, descriptive name so it is easy to find months later. When you are ready, download the result as a spreadsheet, choosing which rows and columns to include.
These are real customer searches that ordinary filters cannot express. Describe the figure in plain terms and let the agent work it out.
Companies with more than 5 people in RevOps roles whose tenure is less than one year.US companies using Pardot with fewer than 5 marketing employees.Companies where most of the engineering team has been there less than 18 months.Companies where hiring for AI and ML titles is accelerating, with the last 30 days more than double the prior 30.Companies where the ratio of account executives to sales development reps is between 1.5 and 3, and engineering is more than 40 percent of all employees.Take my uploaded list of 1,500 companies and add how many software engineers each has in North America and Europe, and whether that count exceeds five.