The 45-Minute Task That Should Take 10: A Field Theory of Business Operations Waste
Every operator knows this one. A task that should take ten minutes — reconciling a transaction, sending a follow-up, filing a document, chasing an invoice — somehow eats forty-five. This essay maps where the extra thirty-five minutes go, by category and by vertical, and what changes when each is automated.
What the extra 35 minutes actually contains
It is not that the task is hard. The task is exactly what it sounds like — ten minutes of actual work. What stretches it to forty-five is a stack of small frictions, each one innocent in isolation, ruinous in aggregate. Naming them precisely is the first move, because the right automation depends on which friction is dominant.
The five categories below come from auditing how owners actually spend their time, not from a productivity book. We've run this audit informally on dozens of multi-entity operators and small-business owners. Every single one of their forty-five-minute tasks decomposes into some combination of these five categories. The ratios differ; the structure is identical.
The five categories of friction
Listed in order of how much time they typically eat in a forty-five-minute "ten-minute task." Numbers below are observational, drawn from operator interviews and our own engagement baselines.
You stop, you swap tools, you forget what you were doing
The task starts in your inbox. It moves to QuickBooks. It needs a number from your CRM. The number is wrong, so it goes to a spreadsheet. The spreadsheet sends you back to the inbox to find the original email. Every tool switch costs 30–90 seconds of "where was I?" Multiply by six switches and you've lost ten minutes before you've done the work.
What changes when automated: The task happens inside a single view. The data comes to you, not the other way around. McKinsey Global Institute's frequently-cited 20% knowledge-work-search figure (a full day per week) is mostly this category.
You don't have the one number you need to finish
You're reconciling a transaction. The vendor name is ambiguous. The invoice is in a folder somewhere. You can't remember whether this charge was a reimbursable or a job cost. You text your assistant. They reply twenty minutes later. The task was ten minutes of work and an hour and twenty of waiting.
What changes when automated: The information exists, indexed and findable, before you start. Document intelligence pulls vendor / amount / category off the invoice at the point of receipt; the missing field is never missing.
The same data, typed three times
A new tenant. You type the name into the lease tracker. You type the same name into QuickBooks. You type it into the CRM. You set up a recurring invoice. Each entry is a minute; together it's seven, and you'll catch a typo six weeks from now when the rent doesn't match the lease.
What changes when automated: One source of truth. The new tenant gets entered once, and the data flows. This is mostly a software win; AI helps when the data sources are messy and need normalization.
You're not choosing — you're remembering which way you chose last time
"What expense category does Home Depot go to?" "Which entity gets this charge?" "What's our standard reply to a tenant late-payment first notice?" These aren't decisions. They're recall. But they take five minutes of mental load because the rules live in your head, not in the system.
What changes when automated: The categorizer remembers. The reply template surfaces. You spend zero seconds choosing, two seconds confirming. Brynjolfsson, Li & Raymond's 2023 NBER paper documented this exact mechanism in customer support — the gain came mostly from removing the "what would I say here?" cognitive load.
The mistake costs ten minutes; finding it costs an hour
You typed the wrong amount. You picked the wrong entity. You forgot to attach the receipt. Each one is a five-second mistake at the time. Each one becomes an hour at month-end when reconciliation finds it and you have to retrace which transaction it was. RAND's research on operational software failure consistently finds that error-correction work is 5–10x the size of the original error work.
What changes when automated: The error is caught at the point of entry, not three weeks later. Confidence scoring on transaction categorization, validation on form fields, anomaly detection on amounts. The 10-minute task stays 10 minutes because nothing leaks into the cleanup pile.
The shadow cost that's not on the clock
You don't actually do the 45-minute task right now. You think about it five times today. Each time, ten seconds of "I need to remember to do that." Multiply by six pending tasks and you've burned an hour of attention without touching any of them. This is the cost no productivity tool measures and the one most worth eliminating.
What changes when automated: The task is either handled in the background or surfaced when it's ready to be done in two minutes. You stop holding it. This is the actual goal: less work that requires your attention.
The pattern by vertical — five concrete examples
The same five frictions, dressed differently in each industry. Each example below is a real ten-to-forty-five-minute task, mapped to the categories above and the automation that actually fixes it.
Property management — Late-rent first notice
The 10-minute version: Send a courteous payment reminder to the tenant who hasn't paid by the 5th.
The 45-minute version: Check rent roll. Identify who's late. Look up tenant in the lease tracker. Confirm grace period. Find the unit-specific account ledger. Compose the message. Send through the right channel (email or text, by tenant preference). Log the touch. Set a reminder to follow up. Realize you missed two tenants. Start over.
Dominant friction: Context switching, missing info, re-entry.
What automates it: Late-rent detection runs daily. A draft notice with tenant name, amount, lease grace period, and contact channel is queued for owner approval each morning. Click to send. Logging is automatic. More on property management AI here.
Multi-entity owner — Categorizing the week's transactions
The 10-minute version: Review last week's transactions across business accounts and code them.
The 45-minute version: Log into QB for entity A. Code. Log into QB for entity B. Code. Hit a Home Depot charge — was that the rental property or your house? Find the receipt. It's in a pile. Skip it for now. Code the rest. Log into the personal Mercury account. Code. Forty-five minutes later, you've coded most of it and pushed the receipts pile to next week.
Dominant friction: Context switching, decisions-that-aren't-decisions, missing info.
What automates it: A categorizer that knows your entities and your history pre-codes the high-confidence majority. Exception queue surfaces the ambiguous ones with the receipt already attached (matched by amount + date + vendor). Ten minutes of review, no log-ins, no piles. More on the multi-entity problem here.
Federal contractor — Drafting a capability statement
The 10-minute version: Produce a one-page capability statement targeted at an upcoming Army solicitation.
The 45-minute version: Find the last cap statement. Update the NAICS codes. Find the right past-performance examples — open the CPARS database. Find the certifications page. Reformat. Cross-reference the solicitation's required experience. Realize you used the wrong agency's language. Start over with the right tone.
Dominant friction: Missing info, re-entry, decisions-that-aren't-decisions.
What automates it: A retrieval-augmented draft pulls your past performance, certifications, and NAICS coverage, then writes a first draft matched to the solicitation's language. You edit. Ten minutes of editing replaces forty-five of assembly. More on federal contracting AI here.
Independent contractor — Estimate from a site visit
The 10-minute version: Send an estimate to the homeowner you walked through this morning.
The 45-minute version: Open the laptop in the truck. Find the estimate template. Type the address. Type the scope. Look up materials prices. Calculate labor. Open the photos folder. Find the right photos. Drop them in. Email the estimate. Realize you forgot to attach the licensing certificate. Re-send.
Dominant friction: Re-entry, context switching, mental-tax.
What automates it: Voice-memo a scope walk-through at the end of the site visit. The system transcribes, drafts the estimate from a price book it already knows, attaches the photos you took on the phone, and queues it for your one-click review and send before you leave the driveway.
Fund manager — Quarterly LP letter
The 10-minute version: Update the LP letter with this quarter's portfolio summary and AUM change.
The 45-minute version: Pull the AUM number from the admin report. Pull P&L from the spreadsheet. Pull top-five positions from PMS. Open last quarter's letter. Copy the structure. Rewrite the commentary. Add the new positions. Run by counsel. Counsel finds a typo in the disclosure language. Rewrite.
Dominant friction: Missing info, re-entry, errors-caught-late.
What automates it: Templated letter with auto-populated numbers from admin + PMS + accounting. Commentary draft with the prior quarter's structure preserved. Disclosure language locked. You write the qualitative paragraph, not the spreadsheet recap. Ten minutes of judgment instead of forty-five of assembly.
Independent owner — Following up on a stalled deal
The 10-minute version: Send a follow-up email to the prospect who went quiet two weeks ago.
The 45-minute version: Try to remember the prospect's name. Search inbox. Find the thread. Re-read the last three exchanges to remember context. Try to recall what was promised. Realize you said you'd send the references; you didn't. Pull references. Compose the follow-up. Worry about tone. Edit. Send. Forget to log the touch in the CRM.
Dominant friction: Context switching, missing info, mental-tax.
What automates it: The system surfaces stalled deals on Monday morning, with a thread summary, last-committed-action, and a draft follow-up tailored to context. The references are already attached. You hit send. The CRM is updated automatically.
Why the math works out the way it does
The math behind why automation pays off here is unintuitive until you've run it. The ten-minute task happens dozens of times a week. If automation takes the average from forty-five minutes to twelve, you've recovered 33 minutes per occurrence. At forty occurrences a week — typical for an active multi-entity owner across all the categories above — that's twenty-two hours back. Per week. Per person.
Brynjolfsson, Li & Raymond's 2023 NBER paper put the average AI uplift in customer operations at 14% — a smaller number than the operator math implies, because their measure included the full task, not just the friction. The gap between "14% productivity gain" and "twenty-two hours a week back" is precisely the cognitive-load category — the already-mentioned mental tax that doesn't show up in any clock-on-the-wall measure. Both numbers are real. The bigger one is what owners actually feel.
Goldman Sachs' February 2026 10,000 Small Businesses Voices survey found 84% of AI-using small businesses cited "efficiency" as the top benefit. That's the wrong word. It's not efficiency they're describing. It's the elimination of the mental tax. The survey question didn't have a good label for it; the operators answered with the closest word available.
The five categories of automation that actually flatten the curve
Once you've named the friction, the automation pattern that fixes it is usually obvious. Five categories cover most of the field.
- Unified ingestion. Every receipt, every transaction, every message, every document lands in one place. The act of organizing — itself a friction — disappears. This is mostly software, not AI.
- Classification at point of entry. Transactions categorized, messages triaged, documents tagged the moment they arrive. The decision that wasn't really a decision is made by the system at confidence; humans see only the exceptions. This is where AI earns its keep.
- Drafting from context. The follow-up email, the late-rent notice, the capability statement, the LP letter — all drafted from data the system already has, queued for human edit. Replaces forty minutes of assembly with eight minutes of judgment.
- Search by meaning. "What's the renewal clause for the Plano lease?" returns the answer, not a folder. Cross-industry research consistently shows 40–60% reductions in search time on well-implemented deployments.
- Anomaly and exception detection. The system watches for the error you would have made — missing receipt, wrong entity, unusual amount — and surfaces it before it ships. Errors caught at entry cost a tenth as much as errors caught later.
What this means for AMG clients
The goal of applied AI in our engagements is not "more AI." It is the flattening of forty-five-minute tasks back into ten-minute ones, and ten-minute tasks back into background work. Most of the time, the right tool is a mix — some plain software for the structural friction, some AI for the read-and-write friction, all of it measured against the only number that matters: how much of your attention the task still requires.
If you'd like to walk through your own ten-minute-that's-actually-forty-five list, send a short note. We'll tell you which of the five frictions is dominant in your case, and which of the five automation categories would change the curve the most.
Have a 45-minute task in mind?
Send a short description. We'll tell you honestly which friction is doing the damage — and roughly what the ten-minute version would take to build.