A 3% versus 33% intervention gap turns drive-thru voice AI into a procurement problem

Food On Demand reported that Taco Bell has extended its voice AI to more than 890 U.S. restaurants across 38 states, deepening its work with Omilia ^[1]. The more useful number for anyone signing a contract came from Intouch Insight's 2025 Emerging Experiences Study, which sent mystery shoppers to Bojangles, Taco Bell and Wendy's and found employees intervening in 22% of interactions across 120 visits where voice AI was confirmed in use ^[3].

The averages hide the point. Bojangles, running Hi Auto's order taker, logged a 3% intervention rate against 30% at Taco Bell and 33% at Wendy's ^[4] - ten and eleven times the Bojangles figure respectively ^[10]. Shoppers calling the interaction "easy and smooth" ran 67% at Bojangles, 57% at Taco Bell and 23% at Wendy's ^[5], a 44-point spread between best and worst ^[12]. Intouch also recorded a 40-second wait-time improvement and 100% speaker clarity in the Bojangles visits ^[6].

Intervention rate is a labor line, not a satisfaction score. The whole operator case for voice AI is that the machine takes the ordering interaction so staff can move to food prep and hospitality ^[14]. At 3%, that trade holds. At roughly a third of orders, the store has swapped an ordering task for a supervision task, and the crew member has to stay within earshot of the speaker box anyway. Two brands in this sample are paying integration and licensing cost for a system that still needs a human on standby for one order in three ^[4].

Two caveats worth pricing in. First, the sample: 120 confirmed voice AI visits across three brands, reported as a combined total with no per-brand breakdown ^[16]. That is thin enough that a 3% figure could be a handful of clean visits rather than a settled operating characteristic. Second, the provenance of the framing. The account leans on vendor-supplied context: Hi Auto says its platform runs across roughly 1,000 QSR stores in the U.S., U.K., Australia and New Zealand and processes more than 100 million drive-thru orders a year ^[7], and Bojangles named the company its Legendary Supplier of the Year in 2025 for deploying the Bo-Linda order taker across more than 400 restaurants, the first time a technology partner had taken the award ^[8]. Those 400-plus stores are about 40% of Hi Auto's stated footprint ^[13], so the best-performing deployment in the study is also the vendor's most concentrated account. That is an argument for attention density as much as platform maturity.

None of which changes the operator's position. The difficulty is known and unglamorous: accents, interruptions, substitutions, modifications, changing inventory, ambient noise, and integration with the rest of the restaurant ^[9]. Somebody has now measured how differently vendors handle it in the field.

What to watch: whether Taco Bell's 30% falls as its 890-store estate matures ^[1][4], whether Omilia or Wendy's vendors publish intervention rates of their own rather than store counts ^[1][2], and whether Intouch repeats the study with per-brand sample sizes disclosed ^[3][16]. The procurement signal to watch for is intervention rate written into a service-level agreement instead of a press release.