For public-facing offices in government
Most of what reaches a public counter, a phone queue or a web form is a status question, a form question, or a where-do-I-go question with a published answer. Handing that band over is a real reduction in load. Anything that decides something about a person is outside the band, permanently, and the boundary between them is written before anything runs.
The volume does not arrive evenly. It arrives on the day a deadline lands, the week a notice mails, the morning after a storm, the hour a news segment runs. Staffing was set against an average that never occurs, so the office is over-resourced on a Tuesday in February and under-water on the first Monday after a mailing.
A large share of what arrives has a published answer. Where do I file this. Did you receive it. What does the letter mean. Which office handles that. Am I in the right place. Staff who were hired to exercise judgement about hard cases spend the day repeating sentences that already exist in a manual, a website, or a notice that was written before the caller could find it.
The people on the other end are not customers and cannot leave. Somebody waiting on hold is waiting on the only office that does this thing. That changes the meaning of a long wait: it is not a lost sale, it is a person who took the morning off work and is now going to have to take another one.
And every contact leaves an obligation. What was said, to whom, on what date, by whom. When a complaint arrives six months later, or an elected official’s office calls, or a records request lands, somebody has to reconstruct a conversation from a note field written under time pressure while the next call rang.
Meanwhile the channels multiply and never retire. Phone, walk-in, mail, email, a web form, a chat widget somebody added, a text line from a grant, and a general inbox that three people have access to and nobody owns. The same question arrives five ways and gets five slightly different answers, all of them defensible and none of them the same.
The office cannot separate the band of contacts with a published answer from the band that requires a person exercising delegated judgement, so both are staffed at the same cost and both wait behind each other.
That mixing is the whole problem, and it is not a technology problem first. It is that nobody has written down which contacts have a settled answer, where that answer lives, and what happens when the answer runs out. Until that line exists on paper, automating anything is guessing where it goes.
So the first act is to draw the line rather than to install something. Take a sample of real contacts, sort them into the band with a published answer, the band that needs judgement, and the band nobody can classify, and count each. The third pile is usually the most interesting thing anyone learns that month, because it is where policy is actually being made by whoever picked up the phone.
Only the first band moves. It is answered from your published source, in your language, with the exact citation that supports it, and every answer leaves a record naming what was said and which source it came from. When the source does not settle the question, the contact stops and goes to a person — it is not resolved on a best guess, and the record shows that it stopped.
The second band never moves. A determination about eligibility, a benefit, a penalty, a status, or anything else that affects a person is exercised by your staff under their delegation. This is not a limitation we are working around; it is the correct arrangement and a vendor offering to take it over is offering you a problem.
Wait time on the published-answer band — measured by time to first substantive response by band, against your own baseline for the same weeks last year.
Consistency of the answer across channels — measured by a sample of the same question asked through each channel, scored against the published source by your staff.
Hours trained staff spend on published answers — measured by contact counts by band multiplied by handling time, sampled the same way before and after.
What the record shows six months later — measured by whether a contact can be reconstructed from its own record without asking the person who took it.
Surge absorption on deadline and mailing weeks — measured by queue depth and oldest-waiting age during a known spike, compared with the previous equivalent spike.
Visibility of the unclassifiable band — measured by a weekly count of contacts no written rule covered, which is a policy finding rather than a workload number.
Escalations arriving through an elected official’s office — measured by count of contacts that reached an elected office before the original request was resolved.
any determination about a person. No eligibility decision, no benefit calculation, no penalty, no status change, no exercise of discretion delegated to a public employee. Those are the job, they stay with your staff, and the boundary is written into scope before anything runs rather than described afterwards.
The answers come from your published sources — the notices you mail, the pages you maintain, the manual your staff already work from. Nothing writes a new answer. Where two of your own sources disagree, that disagreement is reported to you as a finding rather than resolved silently in favour of one, because a resolved contradiction is a policy decision and it is not ours to make.
Records are written into the system you already use for contacts and cases. No second record of a person is created, because a second record is what makes a records request or a complaint response impossible to answer cleanly.
Where an interface into a system does not exist, that is reported as a limitation. It is not worked around by driving a screen, which produces a fragile dependency on a layout and no reliable record of what happened.
Language access is a property of the arrangement rather than a feature added later, and any surface a member of the public reaches is built to be usable by a person using assistive technology.
The people in these queues did not choose to be in a database. Contact content stays inside your tenancy, is retained on your schedule rather than ours, and is exportable by you without asking us to produce it.
Operational access is not permission to train. What the public says to your office does not become material improving anything that serves another organisation, and that boundary is architectural rather than a promise in a document.
On assurance: an independent SOC 2 Type II attestation is in progress and no report exists yet, so nobody can be handed one. Where a state or federal privacy obligation applies to a specific programme, the applicable terms are agreed in writing rather than assumed, and we would rather find the mismatch during scoping than during an audit.
And on authority: every consequential action carries a record naming what was done, under which written grant, by which actor, at what time, with what result. An office answering to oversight needs that record to exist during the work, not to be assembled afterwards to describe it.
The records officer’s question is where contact records live and how long they last. The answer is: in the system you already use, on your retention schedule, exportable by you. A vendor-held record with a vendor-held clock is a public-records problem waiting to be discovered.
Counsel’s question is what is being delegated. The answer is a written grant, short enough to read completely, covering only the band with a published answer, revocable immediately by the person who granted it and without a contract action.
Accessibility is documented for any surface reaching a member of the public, and language access is scoped explicitly rather than left to whoever is on shift.
And the person who answers for the queue needs the number that matters: what share of contacts fell into each band, and what happened to the pile nobody could classify. That number is produced by the observation phase before anything is delegated.
One queue — a single phone line, one web form, or one shared inbox — observed for a full cycle including at least one known surge, with no authority to answer anything.
The observation phase produces three counts and nothing else: contacts with a published answer, contacts requiring judgement, and contacts no written rule covers. Most offices have never had these three numbers and the third one is usually the finding that matters.
That is a legitimate place to stop. If the published-answer band turns out to be small, the honest conclusion is that this office does not have the problem this page describes, and you should not buy anything. If the unclassifiable band is large, the work is a policy question inside your agency and no vendor can help with it.
If it continues, the first grant covers one band of one queue, in one language, with escalation defined before anything answers a single contact.
For anything consequential, agreed, and that is the arrangement — every determination stays with your staff. The question is whether a person who waits forty minutes to be told which office handles their issue was well served by that principle. Moving the published-answer band is what makes the person available for the contact that actually needs them. If a caller asks for a person at any point, they get one, and that is a property of the design rather than a setting.
That is the common case and it is why the observation phase counts the unclassifiable band separately. If a large share of contacts is being answered from institutional memory rather than a published source, that is worth knowing on its own — it means your policy is currently being set at the point of contact by whoever answers, and it will not survive their retirement. The finding is yours whether or not anything is ever delegated.
The design point is that it can only answer from your published source and must attach the citation, so a wrong answer is a wrong source and it is traceable to the document that said it. That is a materially different failure from a wrong answer with no origin. When the source does not settle a question, the contact stops rather than being resolved on a plausible guess, and the record shows it stopped — which is the behaviour you want when the alternative is confident invention.
Then the record has to be better than what exists today, not merely equivalent. Every contact carries what was said, which published source it came from, under which written grant, at what time. When an elected office calls about a specific person, that is answerable in minutes from the record rather than by asking staff what they remember. Start with observation only, so the first thing you get is the measurement, and nothing has been delegated when you decide.
Then the useful comparison is not cost per contact, it is the training cliff. Seasonal staff reach competence roughly when the surge ends, and the published-answer band is exactly the part of the work they take longest to get consistent. Moving that band leaves the temporary staff on the work where a human is actually required, and leaves your permanent staff free during the weeks they are currently underwater.