Playbook No. 08 — Published Friday Oct 30, 2026
One repeatable setup that makes sure a call at 9 p.m., on a Sunday, or during your one staff meeting of the week is answered, captured in the caller's own words, and in the right person's hands the next morning — instead of dying in a voicemail box nobody checks.
Most missed calls are not a staffing problem. They are a design problem. The phone rings after hours, the caller reaches a generic voicemail greeting, leaves half a message (or none), and the organization learns about it — if it learns about it — days later, from a second, frustrated call. The fix is not an expensive answering service and it is not "check voicemail more often." It is a small, deliberate intake system: decide, in advance and in writing, what every after-hours caller will be asked, what counts as urgent, and who sees what by when. Build it once, test it monthly, and the phone stops being a leak in your mission.
Time to run it: about 2–3 hours to set up the first time, about 15 minutes a week to maintain — our editorial estimate, not a measured result. Cost: $0 to design and pilot on free trial units; ongoing voice costs depend on your provider and call volume (see this week's Free Fuel for the nonprofit pricing that applies to one major provider). Cadence: answers every call; reviewed weekly.
One page, five fields, nothing else: (1) caller's name, (2) callback number, read back to confirm, (3) what they need, in their own words, (4) how urgent it is, by the caller's own judgment and by your rules in Step 3, and (5) anything with a deadline — a court date, a benefit cutoff, a medication, a closing shelter bed. Rule for the whole playbook: the system captures; it never advises. An after-hours intake is a disciplined message-taker, not a counselor, not a screener, not a decision-maker. Writing that sentence on the intake card is the step that keeps the whole system safe.
The greeting does three things in under thirty seconds: names the organization, says plainly that the caller has reached an automated after-hours line and whether the call may be recorded, and makes the one promise you will actually keep — who will call back, and by when ("A member of our team reviews every message each morning and will call you back by noon on the next business day" — only if that is true). If your work touches emergencies, the greeting also says, before anything else, what this line is not for and where to go instead — in the US, 911 for immediate danger, and your sector's crisis line where one exists. A caller in crisis who reaches a cheerful robot and a callback promise has been failed by design, not by technology. Write the greeting for that caller first.
Three buckets, decided by you, not by the software: Urgent — the small set of situations that page a named human tonight (define them concretely for your mission; three to five situations is typical, and each one names who gets paged and who is their backup). Next business day — everything with a deadline inside a week, plus every first-time caller. Routine — information requests and the like, answered from the morning summary in Step 5. The technology's job is only to sort callers into the buckets by asking the intake-card questions and matching plain answers ("Is anyone in danger right now?" / "Does this need help before tomorrow morning?"). If the answers are unclear, the rule is simple: unclear goes in the higher bucket. Sorting is a judgment you make once, on paper, while calm — the system just executes it at 9 p.m.
Two acceptable builds, in order of preference. Build A — structured voicemail, no AI required: the greeting walks the caller through the five fields one at a time, the recording is transcribed automatically, and the transcript plus recording land in one named place (a shared inbox or intake folder) — never in a personal voicemail box. Modern phone platforms, including the one in this week's Free Fuel, list speech recognition, call recording, and text-to-speech among their standard voice features, so this build is usually configuration, not development. Build B — a conversational voice agent: the same five fields, the same greeting, the same triage rules, asked conversationally instead of by prompt. Build B is nicer for callers who struggle with voicemail menus — older callers, callers in distress, callers for whom English is a second language. It is also the build that can go wrong in more ways, so it inherits two extra rules: it may never improvise advice (Step 1's rule, enforced — if it doesn't know, it takes the message), and a human listens to a sample of its calls every week (Step 6). Whichever build you choose, the test is the same: call it yourself, at night, from a number it doesn't recognize, and see what lands in the morning pile.
Every after-hours message becomes one line in a single morning summary — caller, need in their words, bucket, deadline if any — sent to one named owner before the day starts. The owner works the Urgent leftovers first (there should be none — urgent pages a human the same night; if an urgent message is sitting in the morning summary, the triage or the paging failed and that is a Step 6 finding), then Next Business Day, then Routine. Callback promises are tracked against the greeting's promise: if you said noon, the summary flags anything unreturned at noon. This is where "never miss a caller" is actually kept — not in the answering, but in the owning.
Once a month, place two test calls: one routine, one that should trip the Urgent bucket. Check that both were captured completely, sorted correctly, and that the urgent one actually paged the human. Listen to a sample of real calls (with whatever consent your recording notice and your own policy require) and read the transcripts for the two failure modes that matter: a caller whose need was flattened into the wrong words, and a caller who needed a human and got a form. Fix the greeting, the questions, or the buckets — in writing, on the intake card — and test again. A system nobody tests is a voicemail box with better marketing.
The one action: this week, write the intake card — five fields, three buckets, the honest greeting — on one page, before you evaluate a single tool. Every tool decision after that is easy, because you'll know exactly what the tool has to do.
The steelman for the skeptics: "Our callers need a person, not a robot — automating the phone is automating away the mission." For some calls, that is exactly right, and this playbook agrees: emergencies are routed to humans and crisis lines in the greeting itself, urgent situations page a named human the same night, and the system is forbidden from advising anyone about anything. What it replaces is not a person. What it replaces is the unanswered ring and the unchecked voicemail box — the two outcomes in which the caller gets no one at all. The honest comparison is never "robot versus warm human at 9 p.m." It is "a disciplined message, owned by a named person by morning, versus silence."
Published Friday, October 30, 2026. The Issue #8 newsletter edition wraps this playbook and lands in inboxes separately — browse the newsletter archive. This page is the playbook itself, frozen as published: playbooks never change after they go up. If a better way appears, it gets a new number, not a rewrite.
Free forever. One playbook every Friday.
Back to the shelf · ← Playbook No. 07 — Boards, briefed · Playbook No. 09 — Choose the work first →
One tested AI playbook per issue — for grant writing, donor thank-yous, volunteer scheduling, board reports — you can put to work the same day you read it. Free. One email per issue. Unsubscribe anytime.