Success metrics your company has never measured

Operational KPIs for a small business are the numbers that show how work moves through the firm: how fast a request gets an answer, how often a file arrives complete, how much each person handles, and where work and money stall. Most firms with 5 to 30 people have never seen these numbers, because the work runs through inboxes, PDFs and spreadsheets that record the result and none of the steps. When AI does the admin work inside a system that logs each step, every step gets a timestamp, and the numbers appear for the first time.

Below are ten of them, with formulas, grouped into speed, completeness, capacity and leakage.

Why don't small firms measure their operations?

Because nothing records the steps. An inbox knows when an email arrived. It does not know when the request inside it was finished, who was waiting on whom, or how many times someone asked for the missing form. A spreadsheet holds whatever someone typed into it last.

There is no shortage of work to measure. Microsoft's 2025 Work Trend Index, based on Microsoft 365 signals, found the average worker receives 117 emails a day and is interrupted every two minutes during core work hours, 275 times a day. Asana's 2021 Anatomy of Work Index, a survey of more than 13,000 knowledge workers, found they spent 60% of their day on work about work, not the skilled job they were hired to do.

The same holds in the industries this post uses as examples. In a 2025 survey of 1,984 real estate industry professionals, run by the National Apartment Association and sponsored by AppFolio, leaders put 66% of their time into routine operational and reactive work and 16% into strategic work. In Liberty Mutual and Safeco's 2025 Independent Agents at Work study, 87% of agency frontline staff said their workload had gone up in the past year, and 51% reported feeling burned out.

The hours get spent. What the owner gets back is a feeling: busy, behind, stretched. You cannot manage a feeling.

Which firms already measure this?

Law firms that bill by the hour. Their billing software sits in the path of every hour, so they can see where the day goes. Clio, which sells law firm software, publishes benchmarks from its Legal Trends Report. Its 2026 figures: in an eight-hour day, lawyers record about 3.4 billable hours, invoice about 3.0 and collect about 2.8. The median firm carries 44 days of finished work not yet billed and 27 days of invoices not yet paid.

Those numbers exist because a system did part of the work and wrote it down. That is the argument of this post: measurement comes from the system that does the work, not from a person remembering to log it.

What changes when AI does the admin work?

An AI layer that does admin work in one system can record every step it takes. A request came in. A document was missing. A follow-up went out. A person approved the reply. Each of those is an event with a time on it.

Line the events up and you get cycle times, completeness rates, backlogs and drop-offs, with no timesheet.

Most companies adding AI skip this part. In McKinsey's State of AI survey, published in March 2025, respondents were asked about 12 practices for adopting and scaling generative AI. Tracking well-defined KPIs had the most impact on the bottom line, yet less than one in five said their organization did it. In Liberty Mutual's 2026 Independent Agency Growth Study, agents who used AI estimated they saved 4 hours a week. That is an estimate. A firm whose AI logs every step can count instead.

Ten operational KPIs, with formulas

Use the median for time metrics, not the average. One stuck file can drag an average for weeks. Watch the 90th percentile too: it shows how bad the slow cases get.

Speed

1. First response time. Time from a request arriving to the first real reply.

Formula: median of (first reply time minus received time).

Example: an insurance agency gets a quote request through its website at 4:40 pm on a Friday. If the first reply goes out Monday at 10 am, that prospect waited more than 65 hours.

2. Cycle time. Time from a request arriving to the work being done.

Formula: median of (done time minus received time), split by request type.

Example: a property manager runs at least three clocks: rental applications, maintenance requests and lease renewals. One average across all three tells you nothing.

3. Stall share. The part of cycle time spent waiting, and on whom.

Formula: time in a waiting status divided by total cycle time, grouped by who the item waits on (client, carrier, staff or approver).

Example: say a bookkeeping firm closes a client's month in 9 days, and 6 of them go to waiting for bank statements. Faster staff will not fix that. Asking earlier will.

Completeness

4. First-pass complete rate. Share of submissions that arrive with everything needed.

Formula: submissions complete on arrival divided by all submissions.

Example: rental applications that come in with ID, recent pay stubs and bank statements on day one. Every incomplete one starts a chase, and every stub still has to add up.

5. Chases per file. How many times someone had to ask for something.

Formula: follow-up requests sent divided by files completed.

Example: say an agency renews 300 policies a month and averages two chases each for updated driver and vehicle details. That is 600 messages a month someone writes by hand. Chasing is one of the five jobs teams still do by hand.

6. Approval edit rate. Share of AI drafts a person changed before approving them.

Formula: drafts edited before approval divided by drafts approved.

Example: a consulting firm has AI draft its project status emails, and a partner approves each one. A rising rate is a sign the setup needs an update. This number only exists when a person approves before anything goes out.

Capacity

7. Work handled per person. Completed items per full-time person per week, by type.

Formula: items completed divided by full-time staff divided by weeks.

Example: a property manager with 4 staff and 600 doors tracks maintenance requests per person by month. Track it per unit you are paid on too: requests per door, service tasks per policy, hours per client.

8. Backlog age. How old the open work is.

Formula: count of open items with no status change in N days, plus the age of the oldest open item.

Example: maintenance work orders with no update in 7 days. That list is the agenda for Monday's meeting, and the oldest item is the first call to make.

Leakage

9. Silent drop-off rate. Share of work that never got a next step.

Formula: items due with no contact and no status change in the window divided by items due.

Example: an insurance agency counts policies that reached 30 days before expiry with no renewal contact. Each one is a renewal left to chance.

10. Days to cash. How long finished work takes to become money.

Formula: unbilled days equal (value of finished work not yet invoiced divided by annual revenue) times 365. Unpaid days equal (value of open invoices divided by annual revenue) times 365.

Example: law firms call these lockup days. The method works for any firm that bills for work. A consulting firm with $1.2 million in revenue and $150,000 of finished, unbilled work has about 46 days of work sitting unbilled.

The ten at a glance

Metric	Group	Formula
First response time	Speed	Median of first reply minus received
Cycle time	Speed	Median of done minus received, by type
Stall share	Speed	Waiting time divided by cycle time, by who
First-pass complete rate	Completeness	Complete on arrival divided by all received
Chases per file	Completeness	Follow-ups sent divided by files completed
Approval edit rate	Completeness	Drafts edited divided by drafts approved
Work handled per person	Capacity	Items completed per full-time person per week
Backlog age	Capacity	Open items idle N days, plus the oldest
Silent drop-off rate	Leakage	Items with no next step divided by items due
Days to cash	Leakage	Unbilled or unpaid value divided by revenue, times 365

How do you measure admin work?

List the five requests that arrive most often: a quote request, an application, a renewal, a work order, a document request.
For each one, write down what "received" and "done" mean. If two people disagree, the metric will too.
Pick one metric from each group. Four numbers are enough to start.
Take the timestamps from where the work happens, not from memory. A shared inbox with statuses gets you part of the way. A system that does the steps records them as it goes.
Review monthly. Change one thing, then watch the same number the next month.

Do AI operations pay off for a small business?

Only if the AI shows you what it did. In the Census Bureau's Business Trends and Outlook Survey, 19.8% of US businesses reported using AI as of May 3, 2026. Between December 2025 and May 2026, use rose among firms with at least 20 employees and did not change significantly among firms with fewer than 20. Salesforce's survey of 3,350 leaders of businesses with 200 or fewer employees found growing firms were twice as likely as declining ones to have an integrated tech stack, 66% against 32%.

AI that writes faster emails is a faster inbox. AI that does the work in one system and records each step gives you numbers you can manage.

Where dailybuilt fits

dailybuilt is the thinking layer for your company. The dailybuilt team maps your admin work, sets up AI to do it, and a person on your team approves before anything goes out. Each type of work lives in its own custom collection, automations move it along on 31 event triggers plus schedules, waits and branches, and every change is written to an audit log with who did it and when. When the work runs through dailybuilt, that log is what the monthly numbers are built from, and you can export all of it at any time.

Each month you get a report of the work handled and numbers your firm never had. It is a monthly retainer, scoped on an intro call. See how it works for property managers and insurance agencies.