Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Customer Experience, Retention & Loyalty › Service and Support

Building a Library of Saved Replies

How to build saved replies for a support team: start from contact reasons, one reply per question, gaps to personalise, links to help articles and review dates.
Tutorialbeginner12 min readUpdated

What saved replies are for, and when they hurt

A saved reply, sometimes called a canned response or a macro, is a pre-written answer an agent inserts and then edits. It does three jobs. It keeps the facts consistent, so two customers asking the same question get the same answer. It carries the agreed tone, so a new agent sounds like the rest of the team. And it frees the agent's attention for the part of the reply only they can write: the customer's situation.

The brand voice and cadence guide puts the rule plainly for social inboxes, and it holds for every channel: saved replies are fine for the questions that arrive daily, provided they are written in the voice and edited for the specific case. A reply that could have been sent to anyone tells the reader that nobody read their message. On social accounts, the social media inbox triage SOP is the shift routine in which these replies are adapted for each person, while complaints are moved to private and passed to the service team.

Saved replies hurt in three situations. When they are sent unchanged to a customer whose question was slightly different, the customer has to write again. When they hold a policy that has since changed, they repeat the old policy with the authority of an official answer. And when they are used for complaints, refusals or anything sensitive, a templated reply reads as indifference at the moment the customer most needs to feel heard. The service and support hub treats tone and resolution as standards; a library that undermines either is worse than none. Public replies to reviews at a chain's branches carry the same risk in front of every future customer, which is why the guide to managing reviews across many locations gives branch managers a reply guideline rather than a set of fixed replies.

Before you start

  • Have the tone guidelines. Replies are written in the support tone, not invented reply by reply. The tutorial on writing customer service tone guidelines builds them, with the default settings, tone by situation, words to avoid and the shape of a good reply.
  • Have your contact reasons. You need a list of why customers get in touch and how often, from your own records. If contacts are not yet tagged by reason, the tutorial on tagging support contacts by reason builds the list and the habit of tagging at close. A month of tagged contacts is enough to start.
  • Know your help centre. List the help articles that exist, who owns each, and when each was last reviewed. The support channels and self-service guide treats the help centre as a product with owners and review dates; saved replies lean on it.
  • Name an owner for the library. One person in the support team decides what goes in, how it is named and when it is reviewed, with agents contributing drafts.
  • Check where replies live. Use whatever feature your support tool offers for stored replies, so agents insert them without leaving the conversation. A document of replies agents copy from is a fallback, not a library.

The steps

  1. List the questions from contact reasons

    Turn each contact reason into the questions customers actually ask under it. "Delivery" is a reason; "where is my order", "can I change the address" and "it says delivered but it has not arrived" are three questions with three different answers. Read a sample of real contacts under each reason and write the questions in the customers' words. Note which questions are simple and factual, which depend on the customer's account or order, and which are complaints or sensitive matters that should not be answered from a template at all. Questions about AI, such as whether the customer is talking to a bot or what happened to their data, are contact reasons too, and the guide to answering customer questions about AI shows how to write those replies from the business's AI use register.

  2. Choose which replies to write first

    Plot the questions on an impact-effort matrix. Impact is how often the question arrives in your own contact counts and how much an inconsistent answer costs; effort is how hard the answer is to write and keep accurate. Questions that arrive often, have a stable answer and are easy to write come first. Questions that arrive often but have a messy answer are a sign the policy or the help article needs fixing before a reply can be written. Leave rare questions to agents' own writing.

  3. Write one reply per question in the tone guidelines

    Each saved reply answers one question. Combined replies that try to cover several situations become long and force the agent to delete half of them, which is where mistakes creep in. Follow the reply shape in the tone guidelines: the answer or outcome first, then what the customer needs to do or what happens next, then any background. Nielsen Norman Group's research on how people read online, as the support channels guide and the writing for the web guide describe it, shows that most people scan for their answer, so put the answer in the opening sentence and keep paragraphs short.

  4. Mark the parts an agent must change

    Every reply has parts that are the same for everyone and parts that must be written for this customer. Mark the second kind with an obvious placeholder in brackets, such as the customer's name, the order number, the date that applies to them and a sentence that responds to what they actually said. Make the placeholders impossible to miss, so a reply sent with a bracket still in it is caught by the agent, not the customer. A reply with no part to personalise is usually a sign the answer belongs in the help centre instead.

  5. Link the help article instead of copying it

    Give the short answer in the reply and link the help article that holds the detail: the steps, the conditions, the exceptions. Copying the article into the reply creates a second version that someone must remember to update. The help centre and documentation SEO guide covers writing and structuring those articles so customers can find them; the saved reply is one more route to them. If no article exists for a frequent question, write one before the reply, so the reply has something to point to.

  6. Name and file replies so agents find them

    Agents search the library mid-conversation, so names must match how they think. Use a consistent pattern: the contact reason, then the question, then any variant, for example "Delivery: change address before dispatch". Group replies by reason in the same order as your tagging list. Avoid internal codes and names that only the author understands. Test the names by asking an agent who did not write them to find five replies quickly.

  7. Give each reply an owner and a review date

    Record against every reply who owns its facts and when it will next be reviewed. The owner is whoever owns the policy or product behind it: the delivery reply belongs with whoever sets the delivery promise, not with the support lead. The review date depends on how often that area changes. A reply whose review date has passed is flagged, and the library owner chases it, so stale replies are found on a schedule rather than by a customer.

  8. Tie replies to policy and product changes

    Most outdated replies come from a change elsewhere that nobody told support about. Add the saved reply library to the checklist for every policy, price and product change, so the person making the change lists the replies it affects. The tutorial on writing release notes and a changelog links each note to its help article; link the affected saved replies the same way. If the team uses AI to draft replies, outdated replies and articles matter twice over: how grounding and retrieval work explains why a system retrieving from outdated content repeats it, and the AI in customer service guide covers drafting, which this tutorial does not. Which questions are better answered from the library and which from a draft written for each message is weighed in the comparison of saved replies and AI reply drafts.

  9. Check how replies are used and edited

    Once a month, look at which replies were used, which were never used, and how much agents changed the ones they sent. A reply that agents always rewrite heavily is wrong or unclear; a reply sent with no edits at all may be going to customers it does not fit. Read a sample of sent replies against the conversations they answered. Watch for repeat contacts after a saved reply: a customer who writes back with the same question was not answered. Treat usage as a signal, never as a KPI to push up, because rewarding the use of templates rewards replies nobody finished.

  10. Retire and merge replies

    Delete replies for questions that no longer arrive, policies that have ended and products that are gone. Merge near duplicates that grew up when several agents wrote their own versions. Keep a short record of what was retired and why, so nobody recreates an old reply from memory. A smaller library that agents trust is used more carefully than a large one they have learned to doubt.

What each saved reply records: a table

Each entry in the library carries more than the reply text. These fields make it findable, accurate and accountable.

FieldWhat it holdsWhy it matters
NameContact reason, question and variant in a fixed patternAgents find it mid-conversation
Question it answersThe customer's question in their own wordsStops a reply being used for a different question
Reply textAnswer first, next step, short background, in the support toneConsistent facts and tone
PlaceholdersThe parts the agent must write: name, order details, a response to what the customer saidPrevents replies that could have gone to anyone
Help article linkThe article that holds the full detailOne version of the detail to maintain
Do not use forSituations where the reply must not be sent, such as complaints or a related but different questionKeeps templates out of sensitive conversations
OwnerThe person who owns the policy or product behind the factsThe right person checks the facts
Review dateWhen the facts are next checkedStale replies are found on a schedule
Linked changesThe policies, prices or products whose change affects this replyChanges elsewhere reach the library
Change logWhat changed, when and whyShows which version a customer received

Channels change how a reply reads. A reply written for email can be too long for chat and wrong for a public social thread, where personal details move to a private channel after the first reply. Keep channel variants as separate entries only where the difference is real. Where customers are answered in several languages, the guide to supporting customers in several languages and time zones covers translated replies and who reviews them; this tutorial covers the library in the team's own language. Every saved reply is a Touchpoint customers judge the brand by, so treat it with the same care as anything marketing publishes in the brand voice.

An online candle shop builds its first reply library (illustrative)

Illustrative example — not a real result

Plokwyn Candles builds a saved reply library from a month of contacts

Plokwyn Candles, an invented online candle shop, has 4 people answering email and chat. Over one month they tag 860 contacts by reason. Delivery accounts for 310, product questions such as burn time and scent strength for 190, returns and breakages for 140, account and payment questions for 120 and everything else for 100.

The support lead and 2 agents split the reasons into 31 distinct questions and plot them by frequency and effort. They write 14 replies first, all for questions with a stable answer and a help article, such as "Delivery: where is my order" and "Product: why my candle is tunnelling". Breakages stay out of the library because each one is a complaint the agent handles personally. Each reply opens with the answer, marks the customer's name, order number and dispatch date in brackets, and links the help article instead of repeating it. Every reply gets an owner and a review date: the 5 delivery replies belong to the person who manages couriers.

Two months later the shop changes its delivery policy, moving its free delivery threshold and dropping a weekend option. Because the delivery replies list the delivery policy as a linked change, the owner rewrites 4 and retires 1 on the day the policy changes. The monthly check shows that agents rewrote almost all of one product reply, about mixing scents in a gift box; it answered the wrong question, and is rewritten from the customers' wording. Repeat contacts on delivery questions, counted in the shop's own records, fall from 58 in the first month to 37 in the third.

Verify it worked

  • Pick a contact reason from your tagged list. Each frequent question under it has its own reply, named by reason, question and variant.
  • Ask an agent who did not write the names to find five replies quickly. They find them without help.
  • Open any reply. The answer comes first, personal parts sit in obvious brackets and the help article link works.
  • Read the library records. Every reply shows an owner and a review date, and any reply past its date is flagged.
  • Pick a recent policy change. The replies it affects are listed as linked changes and already updated.

If it goes wrong

  • Agents rewrite a reply almost entirely. It probably answers the wrong question. Reread the contacts under that reason and rewrite it in the customers' words.
  • A reply goes out with a bracket still in it. The placeholder was too easy to miss. Make it obvious, so the agent catches it first.
  • Agents cannot find a reply. The names follow internal codes or the author's thinking. Rename to the fixed pattern, regroup by reason and test again.
  • A reply repeats an old policy. Nobody told support about the change. Fix the reply, then add the library to the checklist for every policy, price and product change.
  • Customers write back with the same question. The reply answered something slightly different. Add a "do not use for" note, or split the reply into one per question.

Next steps

Common mistakes

  • Writing replies from memory instead of contact reasons. The library fills with answers to questions the team thinks customers ask, and misses the ones they do.
  • One reply for several questions. Agents delete the wrong half, or send all of it.
  • Copying help articles into replies. Two versions of the same detail, and the reply is the one that goes stale.
  • No placeholders. Replies go out with no sign that anyone read the customer's message.
  • Templates for complaints. A complaint needs a reply written for the person; the service recovery and complaints guide sets out the sequence.
  • No owner or review date. Policies change and the library does not, until a customer quotes it back.
  • Counting usage as success. High use of templates can mean replies nobody finished.
  • Never deleting anything. A library too big to search is a library agents stop trusting.

Frequently asked questions

Do saved replies make customer service sound robotic?

Only when they are sent unchanged. A good saved reply is a starting point: it holds the accurate answer, the right links and the agreed tone, and it marks the parts the agent must write for this customer, such as their name, their order and what they actually asked. Customers notice a reply that could have been sent to anyone; they rarely notice one that was built from a template and finished properly.

How many saved replies does a support team need?

As many as there are questions that recur often enough to be worth writing once, and no more. Start with the questions your own contact records show most often, and add replies when a new question starts to repeat. A library that grows without retiring anything becomes hard to search, and agents stop using it.

Should a saved reply contain the full answer or link to a help article?

Usually both in a light form: the direct answer to the customer's question in a sentence or two, then a link to the help article that holds the detail. Copying the full article into the reply means two versions to keep current, and the copy in the reply is the one that tends to go out of date first.

Terms used on this page

Frameworks

Where this page is used

Also related

Sources

  1. How Users Read on the Web — Nielsen Norman Group

Related