AI for the Arts: Your CRM is full of audience insights; AI can help you ask better questions.
How AI can help arts organizations move beyond traditional broad audience segmentation categories like “dance buyer” or “classical music buyer,” help uncover hidden categories, and create more relevant and targeted messaging, offers, and recommendations.
If you’ve worked in arts marketing for any length of time, you probably have a pretty good idea of who your audiences are. You have your subscribers and single-ticket buyers. Your dance people and your classical music people. Your members, donors, frequent attenders and the people you haven’t seen in a while.
But what if some of the most interesting audience segments are ones that are not so obvious?
Buried in your CRM are years of clues about how people actually behave: what they attend, when they buy, how often they come back, what they’re willing to spend, whether they stick to one genre or bounce happily between them. And there are probably patterns in all of that data that aren’t obvious to even the savviest marketer.
This is where AI can be genuinely useful. Instead of asking it to write yet another subject line (although I definitely use it for that, too), we can use it to help us ask better questions of data we already have and perhaps discover audiences we didn’t know were there.
What could AI actually uncover?
Imagine an organization exports three years of patron-level transaction data from Tessitura. You might include variables such as productions attended, genres/artists, ticket price, purchase date, days before performance purchased, day of week, seat location, number of tickets, discount usage, frequency of attendance, first/last attendance, ZIP code, membership/subscription status, and perhaps email engagement.
Traditional segmentation often starts with categories the marketer has already decided matter:
Subscribers / single-ticket buyers / members / lapsed patrons / dance buyers / music buyers.
AI can instead look for clusters of behaviors that occur naturally in the data. You might discover segments like:
“Last-minute cultural omnivores.” People who attend dance, theater and music performances, who typically purchase within seven days of a performance, attend 4–6 times per year, and almost never subscribe.
Or:
“Artist-driven buyers.” People who attend infrequently but consistently buy when internationally known artists or companies appear and are relatively insensitive to ticket price.
Or something much less intuitive:
“Thursday-night explorers.” Patrons who disproportionately attend Thursday performances, buy lower-priced tickets, frequently try unfamiliar artists, and have unusually high repeat-purchase rates.
Nobody necessarily thought to create those segments. The patterns emerge from the data.
So what tool would do this?
For a typical arts organization, I wouldn't start by purchasing a sophisticated AI platform. I'd start with something like ChatGPT's data-analysis capabilities or a similar secure analytical environment.
You could export an appropriately privacy-protected CSV from Tessitura and analyze it using ChatGPT. By “analyze” I mean that you'd ask it to perform exploratory analysis and clustering.
For example, you could ask:
“Analyze purchasing and attendance behavior and identify variables that appear to distinguish different types of patrons. Then use clustering techniques to determine whether there are naturally occurring audience segments in the data. Don't begin with predefined marketing segments. Describe the characteristics of each cluster and identify what most differentiates it from the others.”
You could then ask follow up questions:
What programming does Segment 4 mostly attend?
Which segments have the highest average annual ticket revenue?
Which people are most likely to attend something outside the genre of their first purchase?
What distinguishes people who make a second purchase within 12 months from people who don't?
That last question gets particularly interesting for audience development.
You can take it a step further with propensity modeling
Finding hidden audience segments is useful, but AI can also help answer a different and potentially more valuable question: who is most likely to do something next?
That’s the basic idea behind propensity modeling. Using past behavior, a model looks for patterns associated with a particular outcome (buying a ticket, returning for a second performance, subscribing, responding to an offer) and estimates which patrons are most likely to exhibit that behavior in the future.
Clustering answers a question like:
“What kinds of patrons do we have?”
Predictive modeling, on the other hand, asks:
“What is this particular patron likely to do next?”
Suppose you were presenting a contemporary dance company. Historical data could potentially be used to estimate which existing patrons are most likely to purchase tickets.
Instead of emailing 80,000 people, you might identify 12,000 patrons whose past behavior suggests an unusually strong affinity for that particular program.
And the predictors might surprise you. Perhaps attendance at certain music programs is a stronger predictor than attendance at other dance performances. Or people who attended a particular community event have unusually high conversion rates for contemporary dance.
That's precisely where AI becomes useful beyond traditional segmentation.
What would you actually ask AI to predict behavior?
Once you have a privacy-protected dataset, you could ask questions such as:
“Using historical patron behavior, build a model to identify which variables are most predictive of a patron making a second ticket purchase within 12 months of their first purchase. Evaluate the model’s performance, explain which factors are most predictive, and score patrons by likelihood of making a second purchase.”
Or, for a specific upcoming program:
“Using historical attendance and purchasing data, identify the behavioral characteristics associated with patrons who have purchased tickets to contemporary dance performances. Then estimate which current patrons are most likely to purchase tickets to the upcoming contemporary dance program. Explain which variables are driving the predictions and group patrons into high-, medium-, and low-propensity categories.”
These aren’t simply prompts asking AI to guess what someone will do. You’re asking it to use historical data to build and evaluate a predictive model and to show you which behaviors are actually associated with the outcome you care about.
Do you need special tools for propensity modeling?
At the more sophisticated end, organizations can use customer-data and predictive platforms. Salesforce has Einstein capabilities; Microsoft Power BI incorporates AI-assisted analysis; and Tableau increasingly incorporates AI into analytics.
But for many small and mid-sized performing arts organizations, you don't need an expensive platform to experiment with the concept. A CRM export + secure data-analysis environment + someone who understands both the organization's programming and its audiences could get surprisingly far.
There is an important caveat: I would not upload raw patron personal identifiable information (PII)—names, emails, phone numbers, addresses, payment information, etc.—into a consumer AI tool. The analysis should use anonymized IDs and only behavioral variables, with the organization's data/privacy policies governing what environment can be used.
Parting Thoughts
None of this replaces a good marketer’s understanding of their audience. AI can find correlations and patterns, but it doesn’t necessarily know why they exist or whether they make sense in the context of your organization, your programming and your community.
What it can do is help us move beyond the audience categories we’ve been using for years and start asking better questions of data we already have. For many arts organizations, that may be the most useful application of AI right now: helping us better understand the people we’re trying to reach and giving us a smarter starting point for deciding what to do next.