Week 1: Notice 01: Directory Assistance
The worked example. Directory assistance was automated decades ago, so this week teaches the four-part notice format on an industry whose outcome we already know, before you have to judge one that's still live.
22 February 2027
This week is the exception to how the rest of the course works: instead of auditing a live pitch, it audits one that already ran to completion. Directory assistance was automated out of existence between the 1980s and the 2000s, so every claim below can be checked against what actually happened rather than argued about in the abstract. That’s the point of putting it first — the notice format below (pitch, migration path, breaking changes, rollback plan) is introduced here on a case with a known answer, so you can calibrate what a honest breaking-changes section looks like before you have to write one about something still in progress.
Outline#
- why this week’s notice describes something that already happened, not a live proposal
- reading the notice below in two registers: the pitch and migration path are written the way the original sales case actually sounded; breaking changes is not
- what “no rollback plan” looks like when it’s true, not just pessimistic
- what the audit session is looking for when you write your own notice later in the course
The pitch
Directory assistance is a solved problem wearing a human costume. An operator's entire function -- match a spoken name to a listing, read back a number -- is a lookup, and lookups are what computers are for. Automated directory assistance systems already handle the majority of call volume in every market that deploys them, at a fraction of the marginal cost, with zero hold time and zero attitude. This is not disruption; it is correction. Retiring the human tier doesn't remove a service, it removes the delay.
Migration path
Callers are routed through speech recognition to a structured lookup against the same listings database operators used, with a text-to-speech readback and an SMS fallback for ambiguous matches; the operator queue becomes a low-traffic escalation path for the recognizer's low-confidence cases only.
Breaking changes
The unsatirized part. What the migration actually costs, plainly.
- Economic
- Directory assistance employed roughly 100,000 operators across North America at its peak; automation and the rise of search engines collapsed that number to near zero within two decades, with almost none of the difference absorbed by adjacent telecom roles. The jobs didn't move, they stopped existing, and the workforce was disproportionately women who had entered clerical telecom work as one of the few stable office-adjacent careers open to them.
- Functional
- An operator disambiguated names a database gets wrong: misheard spellings, recently changed numbers, businesses known by a nickname instead of their listed name, and callers who didn't know the exact information they were asking for and needed a human to ask a clarifying question back. Automated systems handle the clean majority of queries well and silently fail the messy remainder, which simply stopped being served rather than being visibly declined.
Rollback plan
None exists at scale. The operator workforce, the training pipeline, and the physical exchange infrastructure that supported it were fully decommissioned; reconstructing it would mean rebuilding an occupation from zero, not restoring a paused one.
Teaching team
- Marisol Quaye — convenor