At a restaurant technology conference this week, Burger King said people were driving away rather than ordering through its AI voice bot. The company is now creating a simpler way for them to reach a human.
If the broader figures contradicted it, this could be dismissed as one chain's poor test. They don't. Retailers scaling store-intelligence technology rose from 42% to 60% in a year, yet the share of gross sales lost to in-store inefficiencies climbed from 5.5% to 6.4%, Coresight Research found in a survey of 200 U.S. retail decision-makers.
On September 15, a shopping-center REIT described cutting analyst underwriting from a day or two to about ten minutes. AI is producing value in paperwork, but it still struggles at the counter.
Store tech spread faster than store losses fell
Coresight's State of In-Store Retailing 2026 first came out in May, and Coresight recapped it on September 23. Simbe and RELEX Solutions, two vendors that sell store and supply-chain software, sponsored it. It puts the annual cost of in-store inefficiency at $196 billion. The loss rate rose for the second year in a row, from 4.5% of gross sales in 2024 to 5.5% in 2025 and 6.4% in 2026. Retailers reporting at least slight challenges reached 90%, up from 88%.
Within the respondent group, 86% said manual work had declined by an average of 14%. Coresight's cited example, Schnucks Markets, also reported a 30% reduction in out-of-stock items. The survey can't establish that the technology changed the loss rate: losses are self-estimated, the sponsors sell the products, and two lines rising together don't show that one caused the other. Deployment counts alone say nothing about results.
What this means: Before scaling store tech, compare loss rates in stores using it before and after rollout with rates in stores that don't. Without that comparison, you don't know whether to expand it.
The drive-thru is where customers push back
For customers, the drive-thru speaker is AI's most visible public test. Restaurant Business reports that Burger King continues testing voice AI at about 1,500 locations, and is making the bot one way to order rather than the only way. At that same conference, Technomic presented data showing that 46% of consumers find ordering from a human appealing, while 23% feel the same about an AI bot and 46% find the bot unappealing.
McDonald's took a different approach. On September 23, it announced plans to spend $8.5 billion over the next decade modernizing its restaurants. CFO Ian Borden said the voice assistant takes orders with 90% accuracy in English and Spanish and could free at least 50 labor hours per week at each restaurant. He said those hours would support hospitality and food prep, not cuts. The stock fell nearly 5% that day, its biggest decline since April 2025.
The Fed's September Beige Book makes the pressure clear: "heightened price sensitivity among customers was putting a limit on their ability to pass through input price increases." When price increases won't hold, labor hours become the expense to reduce. That makes the drive-thru speaker an obvious place to begin, even though it's also the job 46% of customers say they don't want a bot doing.
What this means: If AI goes at the counter, first create a fast path to a person. Then treat drive-offs and wait time as costs alongside the labor hours saved.
The paperwork dividend
At the BofA real estate conference on September 15, Brixmor's chief investment officer reported that the company can run a broker's pitch through its model and rank it against the entire portfolio in about ten minutes. An analyst previously needed a day or two for that work. He estimated legal savings of about $50K per acquisition. CEO Brian Finnegan said the REIT has reduced outside legal spending by half.
The return comes from paperwork rather than customers. That's the paperwork dividend: AI produces its quickest gains when a document contains a correct answer. In its 2026 landlord whitepaper, Re-Leased, a lease software company, says lease administration and abstraction is the leading use, cited by 60% of respondents. A case study on GGP, covering about 100 U.S. retail centers, shows AI finding exclusives and co-tenancy clauses across hundreds of lease templates. That had previously required one-off legal research. For every AI tool it accepts, GGP turns away two or three others.
RCLCO's 2026 CRE C-suite survey found that among 156 senior leaders, only 3% report no meaningful AI use. Employee productivity ranks as the top goal at 85%, while only 15% expect staff cuts.
What this means: Begin with lease abstracts, underwriting, and site screens, where you can compare the output with the source. Review catches a wrong clause. A customer who drives off never tells you why.
Hard data, soft data
The Federal Reserve Bank of San Francisco offers one reason AI pays on documents and stalls with people. Its analysis of 1,006 banks, which hold more than 87% of U.S. banking assets, found that banks with heavier AI use earned about 0.38 points more in return on assets. It also found that greater AI use was associated with a smaller share of small-business lending. The authors connect that result to AI's advantage with "hard" data like credit scores and financials over the "soft" relationship knowledge those loans require, while making clear that these findings are correlations.
Retail and real estate split along the same line. Leases, rent rolls, traffic counts, and demographic profiles are measurable inputs. A customer deciding whether to talk to a speaker is soft information. Site screening belongs with the hard inputs, so GrowthFactor evaluates a location against the data first and leaves the final call to the person who knows the market.
What this means: Group AI projects by their inputs first, rather than by vendor. Document work should already be earning. Customer-facing work needs a before-and-after number before it expands to more stores.
-Andrew
Founding Team Member, GrowthFactor