On this page
Ask anyone on a city front desk which questions they answer most and they will tell you without pausing. When is my bin collected. How do I get a parking permit. Is the office open on Monday. Where do I pay this. Has my application been looked at yet. The same questions, in the same words, several times a day, every week of the year.
AI is an obvious candidate for absorbing that volume, and it can. But answering a question faster is only the second-best outcome. The best one is that the question stops arriving — and deciding which of the two you are aiming at, question by question, is what separates a useful programme from an expensive one.
Which questions are actually repetitive?
Fewer than most teams expect, and not always the ones they would name. In most agencies a small number of questions account for the large majority of contacts, and they cluster around three things: whether I am eligible, when is the deadline, and what is happening with my thing.
The distinction between those three matters, because they are not solved the same way. Eligibility is a content problem. Deadlines are usually a content and publishing problem. Status is neither — it is a systems problem, and no amount of rewriting will answer it.
Find the real list before automating anything
Staff recollection is a reasonable starting point and a poor finishing one. It over-weights the memorable and the irritating, and under-weights the question so routine that nobody notices answering it.
- Have the front desk tally questions for one ordinary week.Marks on a sheet are enough. The point is to replace recollection with a count.
- Export your site search queries for the same period.These are questions residents asked without speaking to anyone, in their own words.
- Read the last hundred emails to your general inbox.Written enquiries show the full question, including the context the phone call leaves out.
- Check the three pages with the highest exits.Somebody leaving from a service page often phones next. That is where a repeat question is born.
- Ask what gets asked after hours.Voicemails and overnight emails are a different list from the daytime one, and the one an assistant covers best.
Three ways to remove a repeated question
Most teams reach for the second of these and skip the other two, which is why volume so often returns after the first year.
- Fix the page
- The answer exists but is unfindable, buried in a document, or written in language nobody uses. The cheapest fix available, and the one that also helps residents who never ask.
- Automate the answer
- The content is good and the question is still asked, usually because it arrives out of hours or needs several pages read together. This is where an assistant earns its place.
- Remove the need
- The question exists because of how the process works. People ask about status because there is no way to see status. They ask whether the form arrived because nothing confirmed it. Fix the process and the question disappears permanently.
The order to work in
- Count for a week. Stop guessing which questions are frequent.
- Take the top five and ask, for each, which of the three routes applies.
- Rewrite the pages that are simply unfindable or unclear. This needs no technology and often removes a surprising share of the volume.
- Publish the answers that only exist in documents or in somebody’s head.
- Then automate what remains — the out-of-hours questions, and the ones requiring several pages read together.
- Take the status questions to whoever owns the system, because content will never answer them.
- Re-count after a season, and compare the lists rather than the totals.
What to measure
The usual dashboard counts conversations, which tells you the assistant is being used and nothing about whether anyone was helped.
- Which questions left the front desk tally entirely. This is the real result.
- Which questions the assistant could not answer. This is next season’s content plan.
- How often it handed off, and on what. A rising handoff rate on one topic names a page that needs work.
- What residents asked outside office hours, which is volume you were previously losing rather than handling.
- Whether the same questions return after a seasonal change. They usually do, and that tells you which page goes stale.
What not to automate
Some repeated questions should stay with a person however often they are asked.
- The status of an individual case, payment or application, unless a system of record can actually be read.
- Anything requiring identity or involving personal data.
- Anything with legal consequence — enforcement, appeals, permits with conditions.
- Complaints, which are not questions and should be captured and routed rather than answered.
- Anything a staff member would have asked a follow-up question about before answering.
- Emergencies, which need the shortest possible path to a person and should never sit behind a conversation.
Where SmartBot fits
SmartBot answers from your approved website content, documents and knowledge sources, which places it squarely in the middle route above: it handles the questions whose answers you have already published, including the ones arriving long after the office closed.
Its reporting on common questions and content gaps is what feeds the other two routes. The questions it cannot answer are the pages to write; the questions it answers constantly are the ones worth asking why they are still being asked.
Frequently asked questions
How much volume can we realistically expect to remove?
It depends almost entirely on how much of your repeat volume is already answered somewhere on your site. Count a week before estimating, because the honest answer for a given agency is in that tally and not in a vendor’s average.
Will this let us reduce front desk staffing?
Treat that as a separate decision and a risky one. What reliably changes is the mix: fewer routine questions, so the same people spend longer on the enquiries that genuinely needed them. Agencies that plan for a headcount saving usually find they have moved the work rather than removed it.
Our top question is "has my application been processed". Can AI handle that?
Not as a content question, and that is the most useful thing to understand. Status questions need a way for a resident to see status. Until that exists, an assistant can explain the usual timescale — which manages the question without answering it.
Should we write an FAQ page instead?
Write the answers, but put them where the question arises rather than collecting them on one page. FAQ pages are where good content goes to become unfindable, and a resident on a permit page should not have to navigate elsewhere to learn whether they qualify.
How often should we revisit the list?
Once a season for anything with a calendar, and once a year otherwise. The list changes when your services change, and a question that vanished after a page rewrite can return quietly when that page goes stale.
What if staff do not trust the answers?
Show them the sources and give them a fast route to correct a wrong one. Staff trust an assistant that is visibly answering from pages they wrote, and distrust one that produces answers from nowhere — which is the correct instinct in both cases.
Where to go next
Put a tally sheet on the front desk for one ordinary week. Not a project, not a survey — marks on paper. At the end of it you will have the only list that matters, and the top five lines on it will tell you more about what to build, rewrite or redesign than any amount of planning beforehand.