McDonald's × IBM AI drive-thru — the largest NA QSR AI deployment that ended in 2024

NA-US (McDonald's pilot in 100+ US drive-thru locations; IBM operated the technology stack from McD Tech Labs acquired Dec 2021)

Content

McDonald’s × IBM AI drive-thru was the largest NA quick-service restaurant (QSR) AI deployment involving voice-based conversational ordering at the drive-thru. The system used automated order-taking (AOT) technology built by IBM at McD Tech Labs (acquired by IBM from McDonald’s in December 2021) and deployed across 100+ McDonald’s US drive-thru locations starting in 2021.

Timeline (per multiple sources, 2021–2024):

• 2021: McDonald’s acquires Apprente (voice-AI startup), renames it McD Tech Labs
• 2021 (December): IBM acquires McD Tech Labs from McDonald’s; partnership announced
• 2021–2024: AI drive-thru pilot in 100+ restaurants in NA
• June 2024: McDonald’s announces end of partnership with IBM; voice ordering AI shut off at all pilot locations “no later than July 26, 2024”

Why this unit matters — even though it failed:

• It was the largest publicly-documented QSR AI deployment in NA by a single brand
• It had a 3-year pilot period, not a one-off test
• The pilot was extended across 100+ stores rather than held to a small test bed
• McDonald’s did not declare AI-driven ordering a dead-end — they ended the IBM partnership specifically, with reports of subsequent partnerships with Google for AI drive-thru replacing the IBM role
• The pilot ending is not a verdict on consumer-facing AI in QSR — it is a verdict on one specific vendor build and integration

Reasons cited in coverage (multiple sources):

• Order accuracy at noisy drive-thru (SOTA voice models at the time did not yet match the noise-tolerance required)
• Edge cases (special instructions, substitutions) were frequent enough to slow service
• Customer experience degradation when the AI mis-ordered
• Workers’ and franchisees’ reports of cost-to-serve issues when the AI failed

What this unit is doing in the taxonomy

Anchors the consumption × generative AI / voice × consumer-facing cell with a discontinued status — a category of unit the field guide previously did not have. Distinct from:

• Walmart Sparky (walmart-sparky-ai-shopping-assistant.md) — still deployed/piloting
• Instacart Cart Assistant (instacart-cart-assistant.md) — still deployed
• Loblaw PC Express in ChatGPT (loblaw-pcxpress-chatgpt.md) — still deployed

Why it matters for talks

• This is the most visible “AI deployment that didn’t work out” unit in the NA food-system. Worth surfacing as a Negative Result because:
  • It tests the argument that vendor-reported AI deployment scale is not the same as deployment sustainedness
  • It is on-record evidence that voice-based consumer AI in noisy environments was not mature in 2021–2024 — 2026-vintage voice models (Whisper-class accuracy improvements) may now be viable but that is the next test
  • The IBM partnership specifically ending — not McDonald’s AI strategy ending — is a load-bearing distinction; McDonald’s reportedly moved toward Google in subsequent announcements
• Workforce framing. QSR AI drive-thru has been framed as a labour story by some media (displacement concerns) and as an efficiency story by others (faster service, fewer mistakes). The pilot ending complicates the labour-displacement concern by demonstrating that the technology itself wasn’t ready — though it does not resolve the policy question of what happens if/when it works.
• Negative-result units are first-class objects in the field guide. Honest industry reading requires recording what failed, on what scale, after what duration — not just what is currently live.

Critical context

• The unit is discontinued as of mid-2024 — not in 2026. Activity-status should be tagged discontinued and freshness follow-up questions should ask whether successor AI programs at McDonald’s have advanced.
• McDonald’s has not made a public statement rejecting AI in drive-thru; subsequent reporting points to Google partnership activity. Worth checking as of 2026 verification.
• The 100+ restaurants figure is consistently reported but the specific store list and which markets got the AI was never publicly disclosed.
• Customer impact during the pilot (mis-orders, longer wait times, complaint rates) is not quantitatively reported; coverage is qualitative.
• The acquisition-by-IBM-then-IBM-ended-by-McDonald’s sequence is an unusual partnership structure; “tech-lab → vendor → partner-end” is a structural pattern worth noting for AI procurement in NA food retail.
• Worker / labour union concerns were raised during the pilot period; the discontinuation may have been partly tactical (cost / accuracy) rather than a labour settlement.