Kooza, Cirque du Soleil’s touring acrobatics show, needs seven days to raise its blue-and-yellow Big Top. It travels with 56 artists from 19 countries, more than 1,200 handcrafted costume pieces and a live band. After drawing Singapore audiences back to Bayfront Event Space earlier this year, the show is now in Melbourne.
None of this is particularly subtle. The AI that Cirque has been building to run the operation, however, is very much so.
The Montreal-based entertainment group launched an AI-powered invoice assistant in January 2024, before SAP’s Joule assistant was widely available. The point was not to wait for a packaged AI product, but to solve a finance problem using technology Cirque already had in place.
Cirque processes more than 70,000 supplier invoices a year across 38 shows worldwide, drawing on artists and staff from 80 countries. About 40% of incoming inquiries to its accounts payable team were from vendors simply asking for an update on payment status, which was a repetitive task that took up staff time without adding much value.
“The time-consuming research required to identify the payment status of an invoice and its reason was overwhelming. We were looking for a more efficient way to handle this, and SAP Business AI provided us with a simply stunning answer,” says Philippe Lalumière, Cirque’s vice-president of information technology.
The tool does two things. First, it drafts proposed responses to incoming supplier inquiries by using data already in Cirque’s systems. Given the multinational supplier base, it also translates communications where needed.
See also: Should you get excited about OpenAI’s upcoming IPO?
Secondly, it uses sentiment analysis to triage urgency, surfacing the most pressing cases first without waiting for a human to read through a queue. While this is back-office work, it is what makes the project more useful as a business case.
Finance beat costume design
Before choosing the invoice assistant, Cirque held a cross-functional workshop that produced two AI proposals. One would have applied AI to costume design, using past designs to help suggest new ones. The other focused on invoices. Finance won.
See also: Global capitalism bets it all on AI future, alarming voters
“The primary objective wasn’t simply to build AI, but to deliver business value, with a business case,” says Lalumière. For a company whose product is spectacle, the choice is revealing. Cirque did not start with the most visible or creative AI use case. It picked work that could be measured, owned and improved.
The discipline has a recent history. Cirque went through a 20-month shutdown during the pandemic and emerged with a sharper focus on operational rigour. Part of that response was moving to RISE with SAP, an offering under which SAP handles infrastructure management, including cloud, network and system maintenance. That freed Cirque’s internal IT team to focus on business projects rather than upkeep.
Cirque has used SAP products since 2001 and now runs a broad stack across the company. Ariba covers procurement, including more than 18,000 costume items, and consolidates supplier logistics across sea, air and land.
Concur manages more than 7,000 staff trips a year. “With SAP Concur, our workload is now cut in half. The data flows easily and transparently through the system. We focus more on the control than the process itself,” says Chantale Périgny, Cirque’s accounts payable manager.
Meanwhile, SuccessFactors supports a workforce spread across contract types and jurisdictions. By migrating to SAP S/4HANA, Cirque managed to cut the number of cost objects per tour by nearly 80%, from about 140 per production to far fewer. Finance staff also moved from checking data to analysing it.
Besides that, the company gained faster and more reliable visibility into tour performance.
Built to keep control
To stay ahead of the latest tech trends, click here for DigitalEdge Section
The invoice assistant was developed as a three-party arrangement. SAP handled application development and AI expertise. Cirque contributed platform architecture on SAP Business Technology Platform. Serrala, the accounts payable platform the system draws on, enabled data integration through a direct API.
The setup was not designed to leave Cirque dependent on outside consultants. Knowledge was transferred in stages until the source code sat in Cirque’s own repository. One internal expert now owns the full technical stack, with no continuing reliance on SAP professional services.
With the system in Cirque’s hands, the finance team has been able to cut the time spent on supplier inquiries. Urgent cases that once took 48 hours to respond to are now handled in about one hour. Non-urgent cases are resolved 25% faster, while the team’s overall backlog has dropped by 25%.
Procurement has seen similar gains. Guided buying through SAP Ariba has made sourcing supplies for shows around the world up to five times faster than under the previous system, according to Stéphanie LeRoy, Cirque’s procurement senior director.
No AI roadmap without better data
Cirque is considering deeper use of Joule across S/4HANA, Concur, SuccessFactors, Ariba and SAP Analytics Cloud. It may also add targeted AI agents for specific internal processes.
However, Lalumière will not commit to a detailed AI roadmap until Cirque has completed an upgrade to the 2025 version of S/4HANA. That migration is, in his telling, a prerequisite to properly assessing what the platform can now do.
His reasoning has the bluntness of someone who has seen enterprise AI programmes fail quietly. “AI without reliable data sources remains a dead end,” he says.
That caution is what makes the case useful beyond Cirque. Many companies are still trying to prove that AI can do more than produce demos. Cirque’s example suggests the stronger starting point may be a narrow process with clean ownership, reliable data and a clear measure of whether the system works.
For now, the show goes on. Kooza runs in Melbourne until July 19, opens in Brisbane on Aug 1 and reaches Sydney on Oct 17. The Big Top takes seven days to assemble. Getting the data infrastructure underneath it right, it turns out, takes considerably longer and, by Lalumière’s own admission, is still in the works.
