In recent years many large organizations have invested heavily in AI: deploying copilots, running proofs of concept and scaling pilots across functions. McKinsey’s 2026 Global Survey on AI found that 80 percent of respondents report improved individual productivity from AI, yet only 37 percent attribute any EBIT impact to their organization’s AI use; 6 percent qualify as high performers, assigning at least 5 percent of EBIT to AI and reporting significant impact.
The disconnect is less about AI’s technical sophistication and more about where it is applied. Most deployments speed up individual workflow steps. A larger, underexploited opportunity lies at the interfaces between those steps — the handoffs where work moves from one team, function or system to the next.
The coordination‑interface tax: large, fragmented and often invisible
Coordination — aligning, verifying, reconciling, waiting and handing work between steps — consumes an estimated 35–60 percent of total work time in knowledge‑intensive organizations, a convergent range derived from multiple studies. Yet the costs this creates are often invisible because no function owns the interfaces. Demand planners own forecasts, supply planners own allocations, but nobody is accountable for the gap in between where forecasts wait to be reconciled with capacity.
The coordination tax appears in three places:
- Visible costs: dedicated coordination headcount and governance (planning teams, PMOs, review forums). In manufacturing and industrial firms this typically runs about 3–5 percent of revenue.
- Financially visible but often unaddressed costs: cycle time and working‑capital impacts associated with interface latency (longer inventory holding, later receivables, committed but unused capacity).
- Invisible opportunity costs: revenue never captured because the organization could not coordinate quickly enough (products launched late, lost market share). This last component is the largest and does not appear on financial statements.
Because traditional cost programs focus on the visible slice, they miss most of the coordination tax.
Measurement and a concrete industrial case
McKinsey measured interface latency in a seven‑step demand‑to‑production workflow at a disguised Fortune 500 industrial manufacturer; six coordination interfaces sit between the steps. Interface latency exceeded processing time by nine to 36 times. The workflow’s end‑to‑end cycle typically took about one and a half to two and a half weeks while the actual work inside each step consumed half a day to a day — the rest was coordination.
For that single workflow McKinsey estimated an annual coordination tax of $140 million to $240 million, comprising latency‑driven working capital costs, capacity underutilization and revenue foregone because of delayed response to demand signals. The visible portion — people dedicated to coordination — accounted for less than 7 percent of the total; the remaining 93 percent manifested as latency and opportunity costs that did not show up on the income statement.
Why agentic AI matters at interfaces
Previous automation waves (quality programs, lean, BPR, ERP, RPA) improved tasks within steps but could not reliably perform the judgment required at interfaces. Agentic AI is the first generation of technology, in the authors’ experience, that can automate much of the verification, reconciliation and routing that previously demanded human coordination. Unlike RPA — which executes fixed rules inside a step — agentic systems can reason across boundaries, evaluate exceptions, apply contextual rules and escalate only what truly requires human judgment.
The mechanism operates on three practical levels:
- It substitutes a human coordinator whose availability and calendar determined latency with a machine that can act at high speed, collapsing days into hours or minutes.
- It automates routine verifications and constrained reconciliations within agreed business rules while routing genuine exceptions, with context, to humans.
- Each automated handoff generates data that improves subsequent decisions, producing a compounding performance effect.
A practical scenario: instead of arriving on Monday to a stale forecast awaiting manual review, a demand planner finds a reconciled plan in which the system has already ingested new demand signals, matched them against capacity and inventory, flagged three exceptions and staged a revised schedule for the remaining 94 percent of items. The planner spends time on exceptions and judgment rather than routine reconciliation.
Measured outcomes at the manufacturer and wider evidence
At the manufacturer that redesigned its interfaces and deployed agentic AI:
- Planning cycle time fell tenfold from 30 days to 3 days.
- 80 percent of planning became touchless.
- More than 50 siloed planners consolidated into roughly 10 end‑to‑end planners overseeing the system; over 150 team members were upskilled into product management, data science and AI engineering roles.
- Some 75 spreadsheets and legacy tools were replaced by a single platform; manual late‑night allocation adjustments became a five‑minute optimizer run.
- Finance‑validated structural profit impact reached billions within the first two years of the enterprise‑wide transformation (inclusive of platform consolidation and the tenfold clock‑speed improvement).
The authors emphasize that the result was operating leverage rather than mass layoffs: organizations redeploy and upskill people so that SG&A falls as a percentage of revenue because the same workforce produces more revenue, not because the total number of employees necessarily shrinks. Public and private examples (referenced in practitioner literature) show throughput gains and redeployment rather than wholesale headcount reductions.
Academic field evidence aligns with this view. A randomized field experiment by researchers at INSEAD and Harvard Business School assigned 515 start‑ups to two groups given identical AI tools: one group redesigned workflows across handoffs, the other applied AI inside existing steps. Firms that redesigned across interfaces generated 1.9 times more revenue and were 11 percentage points more likely to acquire paying customers.
How to capture the opportunity: organization before technology
The authors argue that interface redesign cannot be delegated to a single function (finance, IT or tech): it requires a COO‑sponsored, cross‑functional team accountable for an end‑to‑end workflow. The process should be sequential: map tasks at each interface, decide which require human judgment and which can be automated, redesign the flow, then introduce technology. Deploying AI on top of unchanged coordination patterns simply adds a tool without removing the source of delay.
Typical interface redesigns take six to 12 months. Investment typically allocates roughly 70 percent to people and process redesign, 20 percent to technology and data integration, and 10 percent to AI model development. Where the coordination tax is large, organizations can expect measurable payback in one to two years, consistent with broader findings on tech and AI transformations.
Conclusion
Every major technology wave reduced the cost of something important; AI is lowering the cost of coordination. For a century organizations structured themselves around the assumption that cross‑functional coordination is slow, expensive and bounded by human bandwidth — and designed layers of meetings and approvals accordingly. Agentic AI changes that economic constraint: by enabling fast, automated, judgement‑aware coordination at interfaces, organizations can redesign workflows around a dramatically lower coordination cost. Those that do so may gain durable cost and speed advantages, because AI‑mediated interfaces improve with each transaction they process.
This article was written by Ali Sankur (McKinsey Chicago), Bernhard Mühlreiter (McKinsey Vienna) and Steffen Fuchs (McKinsey Dallas); edited by Larry Kanter (McKinsey New York).
