McKinsey: The Real AI Opportunity Is in the Gaps Between the Work
Photo: ZHENYU LUO on Unsplash
$140-240M
a year is what McKinsey found one Fortune 500 manufacturer loses to coordination, not tasks, in a single demand-to-production workflow.
Interface latency across six handoffs ran 9-18 days against 12-24 hours of actual work. Redesigning the interfaces, not adding AI to the existing steps, cut a comparable planning cycle from 30 days to three in a separate McKinsey case study.
The week in three lines
- McKinsey put a figure on the cost of coordination between workflow steps, not the steps themselves. At one manufacturer, a single demand-to-production workflow lost 9-18 days to handoffs against 12-24 hours of actual processing time, an estimated $140-240 million a year. Redesigning the interfaces cut a comparable planning cycle from 30 days to three.
- McKinsey and BCG reached the same conclusion from different angles this week: the value sits in redesigning the structure around the work, not in adding a chatbot to work that stays the same. Gartner's new warehouse AI maturity map backs this up from the supply chain side. Most warehouse AI today sits on the bottom two of Gartner's four rungs, some way short of the semi-autonomous and physical-agent tiers the McKinsey and BCG case studies describe.
- MIT Sloan Management Review raises the counter-question to that argument: if AI increasingly takes on the coordination work and frees people for judgement calls, that only holds if the judgement is still real, not just harder to check.
The papers
The real AI opportunity sits in the gaps between the work
McKinsey mapped a demand-to-production workflow at a large manufacturer and found six handoffs between steps were losing 9-18 days to coordination against 12-24 hours of actual work, an estimated $140-240 million a year. In a separate case, redesigning those interfaces cut a comparable planning cycle from 30 days to three and made 80% of planning touchless.
Gartner Supply Chain · Supply chainGartner maps four stages of warehouse AI: most are stuck on the bottom two
Gartner set out four AI trends in warehousing as a maturity ladder, running from traditional optimisation and generative AI through to semi-autonomous agents and physical AI agents combining robotics and sensors. It frames warehousing as at an inflection point, driven by labour constraints, lower-risk capital models and AI technology reaching operational maturity.
BCG · GovernanceBCG's answer to unreliable AI agents: build an operating system around them, not a bigger model
BCG describes 'harness engineering', a five-part operating system for AI agents covering specifications, rules, an audit trail, shared context and quality gates. Using the approach, BCG built an agentic advisory platform for a bank that tripled the time advisers spent with clients and lifted wealth-adviser revenue productivity by more than 30%.
MIT Sloan Management Review · WorkforceAI is making it harder to tell who is actually good at their job
Drawing on expert interviews for a joint study with Axialent, MIT Sloan Management Review argues that generative AI can make work look polished regardless of the underlying skill of the person producing it, creating a 'capability mirage' where organisations look competent while real skill quietly erodes. The researchers say this is early evidence from expert interviews, not a measured outcome.
Also published
- McKinsey Technology Trends Outlook 2026 — McKinsey, 15 Sep 2026
McKinsey's flagship annual scan of 14 technology trends, not a single finding. Two figures stand out: more than three-quarters of cybersecurity vulnerabilities are now classified zero-day, and physical AI is arriving first in manufacturing and logistics.
- Gartner Forecasts Worldwide AI Spending to Grow 49.5% in 2026 — Gartner, 16 Sep 2026
A $2.7 trillion global AI spending forecast for 2026. Useful market context rather than a finding to act on directly.
- Gartner Predicts 60% of Organizations That Ignore Data Governance Culture Challenges Will Fail to Govern AI Successfully by 2027 — Gartner, 21 Sep 2026
A prediction built on a real survey of 223 data and analytics leaders, finding cultural resistance beats funding constraints as the reason governance programmes fail. Adjacent to this week's governance stories but about data culture specifically.
What nobody is saying
Two threads from this week's research do not meet each other. McKinsey and BCG both argue that AI's real value sits in redesigning the structure around the work, whether that's McKinsey's 'interfaces' or BCG's 'harness', so that machines handle coordination and people are freed for judgement. Both assume the human judgement being freed up is real and durable. MIT Sloan Management Review's research from the same week argues the opposite is happening: generative AI is making it harder to tell whether anyone's judgement is actually intact, because polished AI-assisted output no longer reliably signals underlying skill. Neither McKinsey's nor BCG's case studies say where that freed-up judgement is meant to come from, or how an organisation would notice it eroding.
Want this applied to your business?
We turn research like this into a transformation plan, not another slide deck.