Relieving the phone: calls the chat answers first

A phone that rings constantly looks like demand. Look at why it rings and a large part of it turns out to be something else: people who need a single piece of information and have found no faster way to get it. When are you open? Where can I park? Is my item ready? Roughly what does something like this cost? Four questions nobody wants a conversation for, and which still occupy lines every working day and pull someone away from their work. This article describes how to survey the real reasons for calls once and properly, aim the chat precisely at them, measure the decline instead of claiming it - and why the phone number still stays clearly visible afterwards.

Enlarge graphic

Labels in the illustration

• 1. Survey the reasons
• Two full weeks of tally marks
• One mark per call, one reason
• Rough time, no names
• Count the unanswered ones too
• Date the list and keep it
• 2. The chat answers
• Opening hours
• with holidays and exceptions
• Directions and parking
• from a maintained source
• Progress of a case
• per case, only after proof
• Price range
• a range with a date, not a promise
• 3. The phone keeps
• Complaints and anger
• Goodwill, discount, exception
• Cases that carry weight
• Situations not yet sorted out
• Anyone asking for a person
• Technology can look things up,
• deciding stays with people
• 4. Evidence the decline, do not claim it
• Comparison period
• Four weeks before against four
• weeks after, matching weekdays
• Two numbers, not one
• Call volume and duration sit
• side by side, never alone
• Name the disturbances
• Holidays, advertising, outages
• belong in the very same table
• The number stays visible: the chat absorbs calls, it does not replace the route

Calls the chat answers first

Which calls are not really conversations at all

A call is expensive because it occupies two people at the same time. That is justified when something has to be discussed, assessed or decided. A large share of incoming calls does not meet that condition: nothing is discussed, something is looked up. The caller would have preferred to read the same information themselves, but did not find it, did not trust it, or wanted it briefly confirmed. In these cases the phone is not a channel but an emergency exit from a website that failed to give an answer.

The distinction matters because it sets the order of the work. Anyone who wants to reduce call volume is not looking for a way to turn callers away, but for the pieces of information that were missing beforehand. Each of them has a source that already exists in the business - a calendar, an address, a job system, a price list. The chat invents none of it. It makes it reachable, at the moment when someone would otherwise pick up the receiver.

Survey the reasons for calls once, instead of guessing them

Almost every business believes it knows why the phone rings, and almost every survey shifts that picture. The reason is simple: the exhausting call is the one you remember, the frequent one is the one you have to count. Two weeks are enough to turn an assumption into a basis, and it requires neither a project nor new technology. One sheet next to every phone, one mark per call, the reason beside it in two or three words. If you can read out your phone system, add the time of day and the duration - both will say more later than raw volume.

• Count two full weeks, not a three-day sample. Monday morning and Friday afternoon are different businesses.
• Exactly one reason per call, in your own words. Do not pre-sort; sorting comes at the end.
• Note the time roughly, half-hour steps are enough. Later this answers the question which calls would have arisen outside office hours.
• Count the calls nobody picked up. They are the part of the volume that memory reliably leaves out.
• At the end, group them and keep the four to six most frequent reasons. The long tail is not a case for the first stage.
• Date the list and keep it. Without it there is no comparison period later, and therefore no decline you can evidence.

This tally sheet is deliberately a piece of paper and not a call recording. As soon as you start noting names, phone numbers or the content of calls, you create processing of personal data that needs a purpose, a legal basis and a deletion deadline - for a question that anonymous tally marks answer just as well. The General Data Protection Regulation states this principle briefly.

Four answers that need no phone call

In almost every survey we have seen, the same four reasons sit near the top. They are not frequent because the website is poor, but because they share one property: the answer depends on a detail a general page cannot know. The public holiday, the direction of travel, the individual case, the particular scope. This is exactly where an assistant has the advantage, because it may ask back before it answers.

What all four share is that their value depends on the source, not on the assistant. A wrong holiday rule does not improve in a chat window, it is merely distributed faster. Before any of these answers goes into automation, it must be clear who maintains the source and how you can tell it has gone stale. The page on the knowledge base describes what that looks like in practice.

The price range: the answer with the greatest risk

Of the four answers, the price range is the only one that can bind a business. A wrong opening time is annoying; a wrong statement about price works against the company that operates the assistant. It does not take a court case: an indication that creates the impression of an order of magnitude which is not correct is already a commercial practice with consequences. German competition law sets this out as a general standard.

In practice that does not mean the assistant has to stay silent about price - silence simply sends the caller back to the phone. It means three things belong together in the same answer: the range, the date from which it applies, and the conditions under which it applies. That includes the sentence that the binding price comes from a quote, not from the chat. Phrases such as ‘that will cost you’ or ‘then there is nothing to pay’ do not belong in the free text of an assistant, because they remain unchanged in the transcript and are later read as a statement by the company. How tightly this line has to be drawn is shown by the ruling of the Higher Regional Court of Hamm on liability.

Aim the chat at the reasons you surveyed

An assistant that can do a little of everything takes little off your hands. The survey provides the order: whatever is at the top is entered completely first, with exceptions, special cases and the question the assistant has to ask before it may answer. Only once those four reasons are solid does the next one follow. The route from call to chat also runs through visibility: wherever the phone number appears, there also belongs the note that the same information is available directly in the chat.

• For each reason for calling, name a source that is maintained, and a person who maintains it.
• Define the question that is needed before the answer: location, direction of travel, case number, scope of service.
• Settle how identity is proven before the assistant names any case status - a name typed into a chat window is not proof.
• For each of the four answers, store a sentence the assistant says when the source currently yields nothing.
• Offer the chat where the phone number sits today: contact page, footer, location page, confirmation email.
• Set up the handover to a human from the start, not as an afterthought - described in the article on the handover to staff.

The reason these answers stand any chance against the phone at all is waiting time. A call starts with dialling, ringing and a recorded message; a chat answer starts with the first character. In our own measurement series the time to the first character of the answer, median was 1.3 s; this was measured on 64 questions to our own assistant, not on customer projects. The method is documented under measurements, and the traps behind it in the article on measuring response time.

Evidence the decline, do not claim it

After the launch, someone regularly says that things have got quieter. That is an impression, not a result. Things also get quieter during the summer holidays, after a campaign has run out, or when an extension was quietly diverted. The decline only becomes evidence once a comparison period is fixed, several figures sit side by side, and the confounding factors are written down with them. That is exactly what the dated tally sheet from the start was for.

The most telling of these figures is not call volume but average call duration. When the short lookup calls fall away, it rises - the remaining conversations are the longer, substantively harder ones. A lower number of calls together with a higher average duration is therefore a far better indication than a lower number of calls alone. How far an assistant gets without a human can only be measured: in our own measurement series 43 out of 48 questions stayed without a handover to a person; this was measured on a single assistant on our own knowledge base, not on customer projects, and a figure from live operation is not yet available.

The opposite direction matters just as much. A drop in calls that coincides with a rise in unanswered chats is not a success but a relocated queue. And a drop that starts exactly when the phone system failed is measuring the failure. Anyone serious about the analysis writes the known influences of the period into the same table: holidays, public holidays, advertising, price changes, outages. Whatever is not written there turns up later as an explanation nobody remembers precisely. The evaluation of the conversations themselves is described in the article on conversation analytics.

Why the phone number stays visible

The obvious shortcut after a successful launch is to set the phone number smaller or move it one level down. It is wrong for three reasons. First, it does not remove the reason for calling, only the fast route to it - a question becomes a search and then an irritated call that takes longer than the one avoided. Second, there are people for whom a chat window is no alternative: a poor connection, operation via assistive technology, uncertainty about writing, an urgent case. Third, openness about the nature of the counterpart is legally mapped out; the AI Act requires exactly that for systems in direct contact with people.

Anyone who discloses that a machine is answering must, consistently, also say how to reach a human. Both belong in the same view, not on two different pages. Accessibility comes on top: a contact route that runs only through a text field excludes people who cannot communicate well that way. German accessibility law requires many services in electronic commerce to be designed accessibly; a visible second route is the easiest part of that. What this means for the chat window itself is in the article on the accessible assistant.

• The phone number stays where it is: header, contact page, footer, location page.
• The assistant gives the number on its own initiative as soon as it cannot proceed, or when someone asks for it.
• It says when the line is staffed and what happens outside those hours, instead of implying an availability that does not exist.
• It states at the start of the conversation that it is an assistant, not only when asked.
• For complaints, urgency or an explicit wish to speak to a person, the shortest route leads to the phone, not to another follow-up question.

What is better kept on the phone

The dividing line does not run along difficulty but along the question of whether something is looked up or decided. Technology can look things up. Deciding, assessing, reassuring and negotiating stay with people - not out of caution, but because the outcome of those conversations binds the company. An assistant that recognises this and hands over early relieves the phone more than one that wants to finish every case itself and creates cases someone has to put right afterwards.

• Complaints and anything that already starts in anger - there the tone counts, not the information.
• Goodwill, discounts, exceptions: any decision that departs from the stored rule.
• Cases with medical, legal or financial weight, where an assessment is expected.
• Situations the caller has not yet sorted out themselves and that only take shape during the conversation.
• Anything where someone explicitly wants a person - that wish is not an escalation stage, it is the beginning of the conversation.

Honest limits and a checklist

A chat absorbs calls, it does not organise a business. Where opening hours are internally inconsistent, automation makes that inconsistency visible faster, but not smaller. Where nobody maintains the calendar, the assistant distributes the outdated entry more reliably than the website does. And where a business has so far managed phone availability by calling across the room, it has to agree beforehand what actually applies. The effort does not disappear, it moves forward in time - and in total stays smaller than answering the same four questions every day. How the support assistant answers such lookup questions from your knowledge base and hands the rest over to your team is shown on the product page; we are happy to work out with you via the contact page which four questions they are in your case.

• The reasons for calls have been counted over two full weeks, dated and kept.
• The four most frequent reasons each have a named source and a person who maintains it.
• The assistant asks the necessary question before answering and names no case status without proof of identity.
• The price range is stated as a range with a date and a condition, and the binding price comes from a quote.
• The phone number is visible unchanged, and the assistant offers it on its own initiative when it cannot proceed.
• Comparison period, call volume, call duration and the confounding factors of the period sit in one table.
• A date is fixed for repeating the analysis - otherwise the first figure is also the last.

Frequent questions