Decision Latency: The Hidden Cost of Waiting for the One Person Who Knows

A purchase order sits unapproved for three days because the one manager who knows the supplier terms is at a trade fair. A customer quote waits until Thursday because only the head of sales can sign off a discount above 8 percent, and she is in back-to-back meetings. A shipment is held at the dock because nobody except the operations lead knows whether this particular exception is allowed, and he is on holiday. None of these decisions is hard. Each takes about ninety seconds to actually make. Yet each cost days, and behind each one a queue of downstream work stood still.

This is decision latency: the accumulated delay every time a routine decision has to wait on the one person who holds the context. It is not the thinking time. It is the waiting time. And because it hides inside the elapsed time of normal work, most companies never put a number on it, even as it drags out every cycle, stalls deals, and leaves paid people idle waiting for a yes.

The evidence that this is a first-order cost is now hard to ignore. West Monroe’s 2026 research found that 73 percent of leaders believe their organisation loses up to 5 percent of annual revenue to slow decisions and delayed execution - a hidden charge they named the “Slowness Tax”¹. This piece names the mechanism behind that number, models what it costs in euros and throughput, explains why the tools you already own do not remove it, and shows how a Company Brain plus AI employees that decide the routine cases immediately collapse it.

What Decision Latency Actually Is

Decision latency is the total elapsed time between the moment a decision is needed and the moment it is made and acted on. Borrow the word from computing, where latency is the delay before a transfer of data begins - not the size of the data, just the wait. In an organisation, the “data” is a routine judgment, and the latency is how long it sits in a queue before someone with the context finally makes the call.

• It is waiting, not thinking - for routine decisions the deliberation takes seconds; the delay is almost all queue time while the request waits for one specific person to get to it.
• It concentrates on one node - the delay clusters around the approver, the expert or the manager who holds context nobody else has written down, so the whole flow moves at the speed of that one person’s inbox.
• It is invisible in the numbers - the decision itself has no line item, so the days lost hide inside the elapsed time of the surrounding process and never show up as a cost.
• It is routine, not strategic - this is not about hard board-level calls that deserve deliberation; it is the thousands of small, rule-bound decisions that could be made instantly if the context were available.
• It is a queue, not an event - because decisions arrive faster than the one person can clear them, they back up, and every new request waits behind the ones already in line.

The distinction that matters is between the two clocks running on every decision: the deciding clock and the waiting clock.

Once you see the two clocks, the pattern is everywhere: the decision is easy, but it waits. The next section walks through why.

The Anatomy of a Stalled Decision

Every stalled routine decision follows the same shape. A request is ready, but the context needed to decide it lives in one person’s head, so the request waits for that head to become available. Here is the sequence, step by step.

1. The trigger arrives - an order, a quote, a payment, a leave request, a customer exception, a spec change. The work is ready to move.
2. The rule is ambiguous or unwritten - whoever is holding the work does not know if this case is within policy, so they cannot proceed on their own authority.
3. It routes to the one who knows - the request goes to the approver, the expert or the manager who carries the reasoning nobody else has, usually as an email, a Slack ping or a ticket.
4. The queue forms - that person already has a backlog, is in meetings, is travelling, or is simply overloaded, so the request joins a line behind others just like it.
5. Downstream work goes idle - the people who needed the answer to continue cannot, so their time is now blocked on a decision that has not been made.
6. The chase begins - reminders, follow-ups, “any update on this?” messages, a meeting scheduled purely to get an answer - coordination overhead that produces nothing.
7. The decision is finally made - in ninety seconds, days after it could have been, often with less context than if it had been handled fresh.
8. The cycle resets - and the next routine case starts the same journey, because nothing about the dependency changed.

The reason this is so common is that the surrounding work is exactly the kind that already eats knowledge-worker time. Asana’s Anatomy of Work Index found that 60 percent of work time goes to “work about work” - coordination, chasing updates, switching tools and clarifying status - rather than the skilled work itself⁷. Microsoft’s 2025 Work Trend Index found that 68 percent of workers lack enough uninterrupted focus time and 62 percent spend excessive time hunting for information⁸. A stalled decision drops straight into that soil and grows.

What Decision Latency Actually Costs

The cost lands in three places, and each is measurable if you look for it: cycle-time drag, stalled revenue, and idle downstream work. Layered on top is the coordination overhead of chasing the decision in the first place.

The three cost channels

• Cycle-time drag - every process that contains an approval step inherits the latency of that step, so quotes, orders, hires and releases all run slower than the work itself requires. In a distribution network, an average decision latency of 48 hours against markets that move 0.5 percent a day produces continuous value erosion from timing alone¹².
• Stalled revenue - deals, quotes and orders that wait on a sign-off are not just delayed, they are at risk, because the customer is also talking to someone faster. West Monroe ties up to 5 percent of annual revenue to exactly this kind of slow decision and execution¹.
• Idle downstream work - when a decision blocks the next step, the people and machines waiting on it are paid to stand still, and that idle capacity never comes back.
• Coordination overhead - the chasing itself burns time. Asana found teams lose 103 hours a year each to unnecessary meetings and 209 hours to duplicated work, much of it the follow-up and status-checking that a stalled decision generates⁷.
• Decision debt - when the wait gets too long, people stop asking and start guessing, which trades latency for error and rework later.

What the research puts on the table

Bain’s survey of nearly 800 companies found a 95 percent correlation between organisations that make and execute decisions well and those with top-tier financial performance⁵. Decision effectiveness is not a soft metric; it tracks the financial result. And speed is a core component of it.

Why Decision Latency Compounds

A single delayed decision is an annoyance. The reason decision latency is a strategic problem is that it compounds - across a chain, across a queue, and across time. Three mechanisms turn a small per-decision wait into a structural drag.

1. It stacks along the chain

• Sequential waits add up - a process with three approval steps, each waiting on a different busy person, does not wait once; it waits three times, in series.
• Handoffs reset the clock - each time the work crosses to a new decider, it rejoins the back of a new queue, so the latency is not shared, it is summed.
• The critical path runs through the slowest node - the whole process moves at the speed of its most overloaded approver, no matter how fast everyone else is.

2. It multiplies across the queue

• Arrival outruns clearance - when routine decisions arrive faster than one person can clear them, the queue grows, and average wait grows with it - basic queueing behaviour.
• One absence spikes the backlog - a two-day trip does not add two days of delay; it adds two days of arrivals on top of the existing backlog, and the queue takes far longer than two days to drain.
• Batching makes it worse - overloaded deciders batch approvals into a weekly slot, which turns a potential same-day answer into a guaranteed multi-day wait for everything.

3. It erodes as knowledge concentrates

• The node gets narrower over time - as the business grows more complex, more of the exceptions route to the one person who has seen them before, deepening the dependency.
• Turnover is a cliff, not a slope - when the person who holds the reasoning leaves, latency does not rise gradually; the queue simply stops until someone slowly relearns the rules.
• Deloitte finds managers already saturated - nearly 40 percent of a manager’s time is consumed by administrative work and firefighting, so the node has no spare capacity to absorb more¹⁸.

This is why you cannot fix decision latency by asking the bottleneck to try harder. The problem is structural: one node, rising volume, no home for the reasoning outside a person’s head. The tools most companies reach for do not touch that structure, which is the next section.

Why Wikis, BPM Tools and Copilots Do Not Fix It

Most companies have already bought tools that were supposed to speed up decisions. Latency persists because each category solves a different problem than the one that actually causes the wait. The wait is caused by reasoning living in one head; these tools store outcomes, route tasks, or draft text - none of them holds the judgment to decide.

What each category actually does

The pattern behind the failure

• Outcomes are not reasoning - a wiki records that a 12 percent discount was approved for one customer last year; it does not tell you whether 11 percent is fine for a different customer today. The next decision still needs the person.
• Routing is not deciding - a workflow tool is excellent at getting the request to the right inbox and terrible at emptying that inbox, because it was never designed to make the judgment, only to move it.
• Drafting is not acting - a copilot can suggest “this looks approvable,” but someone still has to agree and then key it into the ERP, so the person stays on the critical path.
• Visibility is not capacity - a dashboard that flags 40 overdue approvals tells you the node is overloaded; it does not add any capacity to the node.
• More people is not less dependency - every human you add is another head that can be travelling, overloaded or gone next quarter.

This is not an argument that these tools are useless. A workflow tool is genuinely valuable for routing and audit; a wiki is useful reference. The point is narrower: none of them removes the dependency on one person for the decision itself, which is the thing that creates the latency.

How AI Employees Collapse Decision Latency

The fix is not to make the one person faster. It is to remove the routine decisions from their queue entirely, so only genuine exceptions ever reach them. An AI employee grounded in a Company Brain does this by holding the reasoning, connecting to the systems where the decision has to land, and being available every minute of every day.

What an AI employee does that the tools cannot

• Holds the decision rules - the thresholds, the exception logic, the customer-specific and product-specific reasoning that used to live only in one person’s head, so it can apply them to a fresh case.
• Decides the routine cases immediately - the large majority of the queue is standard and within policy; the AI employee clears those the moment they arrive, at any hour, with no batching.
• Escalates only true exceptions - when a case falls outside the rules or crosses a risk line, it routes to a human with the full context already assembled, so the person spends their judgment, not their time gathering facts.
• Acts across the real systems - it writes the approval into the ERP, releases the order in the WMS, updates the CRM and notifies the customer, so the decision becomes a completed action, not a suggestion.
• Never travels, never batches, never sleeps - the single biggest driver of latency - the availability of one person - simply stops applying.
• Keeps an audit trail - every decision it makes is logged with the rule it applied and the data it used, which is better traceability than an email approval ever gave you.

Routine decided, exceptions escalated

The design principle is a clean split: the AI employee owns the routine, rule-bound decisions, and humans own the exceptions and the judgment calls. This is exactly the split the market is moving toward. Gartner predicts that by 2028, 15 percent of day-to-day work decisions will be made autonomously by agentic AI, up from essentially zero in 2024¹¹, and that 40 percent of enterprise applications will feature task-specific AI agents by the end of 2026, up from under 5 percent in 2025⁹.

Why the Company Brain Makes It Durable

Removing the routine decisions from one person’s queue only stays fixed if the reasoning has somewhere permanent to live. That home is the Company Brain: a living memory of how your company actually decides. Without it, any automation is brittle and any expert departure resets the latency.

What the Company Brain holds

• The thresholds - the discount limits, the payment ceilings, the approval bands that separate routine from exceptional.
• The exception logic - the “except for this customer,” the “never for this product line,” the “always check X before Y” that turns a rule into a real decision.
• The reasoning, not just the outcome - not only that a case was approved, but why, so the same logic applies correctly to the next, different case.
• The escalation map - which cases go to whom, and what context they need to decide fast.
• The corrections over time - every time a human overrides the AI employee, the reason is captured, so the rules get sharper instead of drifting.

Why this is the durable part

• It survives turnover - when the approver leaves, the reasoning does not leave with them, so latency does not spike into a cliff. This is the difference between a fix and a patch.
• It survives travel and overload - the context is always available, so a trade fair or a busy quarter no longer sets the speed of the whole process.
• It stays current - unlike a wiki that is written once and rots, the Company Brain updates as decisions are made and corrected, so it reflects how the company works now.
• It is portable across decisions - the same brain that clears discount approvals can ground order releases and invoice checks, because the reasoning is captured once and reused.
• It compounds - every decision handled adds to the record, so the system gets more capable and the human exception rate falls over time.

The Company Brain is what turns “we automated some approvals” into “decision latency is structurally low and stays low.” It is the difference between speeding up once and staying fast.

The Euro-and-Throughput Model

To make the cost concrete, here is a transparent, illustrative model for a single high-volume decision queue at a mid-sized company. The figures are inputs you can replace with your own; the method is the point.

The assumptions

• Volume - one decision queue receives 400 routine decisions per month (discount approvals, order exceptions, purchase sign-offs).
• Latency - average wait per decision is 2 working days before the one approver clears it.
• Idle downstream cost - each waiting decision blocks about 30 minutes of downstream work at a loaded cost of 60 euros per hour, so 30 euros of idle time per decision.
• Coordination cost - each decision generates about 15 minutes of chasing and status-checking, another 15 euros per decision.
• Stalled-revenue cost - roughly 1 in 20 of these decisions sits on a deal or order where the delay risks the outcome; call the expected loss 250 euros per affected decision.

The arithmetic

Around 276,000 euros a year, from a single decision queue, before counting the strategic cost of moving slower than competitors. Most mid-sized companies have several such queues running in parallel across sales, finance, operations and HR.

The 90-Day Playbook to Cut Decision Latency

You do not fix decision latency across the whole company at once. You pick the worst queue, prove the model on it, and expand. Here is a focused 90-day sequence.

Phase 1: Find the latency (Weeks 1-4)

1. Week 1: Map the queues - list the recurring decisions that wait on one person. Sales discounts, order exceptions, purchase sign-offs, refund approvals, access requests. Rank them by volume times latency.
2. Week 2: Measure the wait - for the top queue, pull the real elapsed time from request to decision. Separate the deciding time from the waiting time so you can see how much is pure queue.
3. Week 3: Capture the reasoning - sit with the person who holds the context and write down the actual rules: thresholds, exceptions, what makes a case escalate. This is the start of the Company Brain.
4. Week 4: Set the escalation line - agree explicitly which cases the AI employee may decide and which must always go to a human, including anything regulated or high-risk.

Phase 2: Put an AI employee on it (Weeks 5-8)

5. Week 5-6: Ground and connect - load the decision rules into the Company Brain and connect the AI employee to the systems where the decision lands - CRM, ERP, email, the approval tool.
6. Week 7: Run in shadow - the AI employee decides in parallel while the human still signs off, so you can compare its calls against the expert’s on real cases.
7. Week 8: Tune the rules - every disagreement is a rule to sharpen. Feed the corrections back into the Company Brain until the routine cases match the expert’s judgment.

Phase 3: Go live and measure (Weeks 9-12)

9. Week 9: Hand over the routine - let the AI employee decide and act on standard cases, escalating exceptions with context attached. Keep the human in the loop only for the escalations.
10. Week 10-11: Watch the two numbers - track average decision latency and the share of cases decided without a human. Both should move fast.
11. Week 12: Report and expand - compare cycle time, throughput and idle downstream work against the week-2 baseline, then pick the next queue.

How Superkind Fits

Superkind builds custom AI employees for SMEs and enterprises, grounded in a Company Brain and connected to the systems you already run. The approach is process-first: we start from where your decisions actually wait, not from a generic product.

• We map the latency first - we go into your organisation and find the decision queues that wait on one person, before building anything.
• We capture the reasoning into a Company Brain - the thresholds, exceptions and escalation rules that lived only in one head become a living memory the company owns.
• The AI employee decides the routine cases - standard, in-policy decisions are cleared immediately, at any hour, and written back into your systems as completed actions.
• True exceptions escalate with context - your experts still make the hard calls, but they receive the case with the facts already assembled and the rule that flagged it.
• It connects to your real systems - CRM, ERP, WMS, email and your approval tools, so the decision becomes an action, not a suggestion.
• It survives turnover - because the reasoning lives in the Company Brain, latency stays low when the approver travels, is overloaded, or leaves.
• Outcomes, not seats - pricing is tied to the decisions handled and the latency removed, not to logins.
• More output without more headcount - the same team clears far more decisions, far faster, because the routine ones no longer queue behind a person.

Decision Framework: Is Decision Latency Your Problem?

Not every slow process is a decision-latency problem, and not every decision should be automated. Use these signals to tell whether this is your bottleneck and where to act.

The compliance line, kept simple

• Routine operational decisions - approving a standard order, releasing a within-policy payment, confirming a low-value request are generally minimal or limited risk under the EU AI Act²².
• Transparency where it talks to people - Article 50 requires disclosing that a person is interacting with an AI system where that is the case²¹.
• High-risk stays human - decisions that materially affect individuals, such as employment, creditworthiness or access to essential services, are high-risk; keep a human decider and use AI only to assemble the case²².
• Auditability helps - because the AI employee logs each decision with the rule it applied, you get a cleaner trail than an email approval ever provided.

Frequently Asked Questions

Related Articles

• Span of Control: How AI Employees Change the Manager-to-Report Ratio
• The Routine-Work Tax: Why Your Best People Spend Half Their Day Below Their Pay Grade
• The Bus Factor: When One Person Leaving Stalls the Whole Company
• The Cross-Department Handoff: Where Work and Knowledge Go to Die
• Process Debt: The Hidden Tax of Undocumented Workarounds
• The Last-Mile Problem: Why AI Pilots Impress in the Demo and Die at Handoff

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI employees that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach.