Design your conversation before you begin creating it. As you develop the design, consider these tips and considerations:

• Five questions — Try to limit the conversation to no more than 5 questions. Especially for SMS channels, keep the survey short. Conversation abandonment rate typically increases sharply after five questions.

• Wording matters — Test your dialogs to ensure you get the responses you require. Vague or ambiguous questions result in low-quality responses. If the respondent is confused, they might abandon the survey. For example, asking "Was it a good experience?" might return "Not really", which might not be useful. Instead, ask "Please rate your experience from 0 to 5, where 5 is Great."

• Don't chase perfection — If the respondent's text response is not what you expected, don't keep asking the respondent for a better response. Retrying too many times increases abandonment rates, os it is better to skip the question than to lose the respondent.

  Tip: Retry questions only when they are for key metrics. Don't risk losing a respondent to abandonment.

• Avoid complex response parsing rules — Overly complex response parsing rules are an indication the question is vague. Don't try to catch every possible answer; instead, prompt respondent how to answer. For example, "Was the sushi fresh (Y/N)?", and then look for 'yes', 'y', 'no', and 'n' in the responses. Similarly, response rules are good for catching common misspellings, but again, don't chase perfection.

  Tip: Start simple, track parsing rates and work on refining parsing rules if the rates are too low. Otherwise a lot of effort can be sunk in that rabbit hole for very little value in the end.

• Key questions first — Place the most important questions first to ensure the answers gets collected. Respondents typically begin to abandon the survey after a few questions, so make sure you ask the important ones first.

  Important: Medallia Experience Cloud evaluates alert conditions after every Medallia Conversations question. Because there is a limit of one alert per survey, place key-metric questions first to catch important alerts. For example, if identifying a detractor or more important than identifying a long wait-time, ask the likelihood-to-recommend question first.

• Avoid segmentation and demographic questions — Unless absolutely necessary.

• Comment questions last — Ask open-ended comment questions last. Respondents often abandon them.

• Comments are less rich — Unlike a traditional web survey, conversation comments tend to be shorter and contain less rich responses, which can impact text analytics.

• Include a thank-you message — Include a short "thank-you" message; they help indicate the conversation is finished.

• Conversation versus web survey completion — Traditional web surveys are only considered "complete" after the respondent finishes the last required question and submits the survey. However, a conversation survey is considered "complete" either

  • after all Mandatory questions have been resolved, or
  • (when no mandatory questions) after the first valid response/answer, and again after each subsequent question.

  As such, and because of the real-time nature of the system, responses can start appearing in Experience Cloud reports before the respondent has answered all the questions. And, alerts are evaluated with each "complete" submission, which can be after every question. 

• Limit SMS message size — Try to limit the size of SMS messages to no more than 160 characters. Many SMS systems break up longer messages into smaller 160-character chunks, and then send each chunk separately. Depending on the device, the original message may then appear as multiple messages. (In some cases, the chunks can even arrive out-of-order!)

  Restriction: The 160 character limit applies to single-character languages. For multibyte languages, the limit can be half (80 characters), and for some complex languages, some symbols may require 3 characters to transimit, in which case the limit might be as low as 53 character/symbols/

• Be timely but respect sending windows — Response rates usually fall if surveys are not received on the same day. But don't send invitations when the respondent is no likely to respond. Similarly, night-time transactions can be sent out the next morning, but the response rates may not be as high. Note, availability windows are based on the respondent's timezone, which must be included in the invitation feed. Invitations are queued to only be sent when during the respondent's availability window.

• Respect a quarantine — Feed-based invitations come through the Medallia Experience Cloud sampling system, which provides a quarantine mechanism to avoid annoying respondents by sending them too many invitations.

  Tip: Generally, 30 days is a good quarantine period. Though there are exceptions like for 1-question surveys sent at each touch-point in a respondent journey, such as during an airline trip.

• Survey expiration — Research shows most respondents respond to invitations within 24 hours, with a significant drop-off after 48 hours. There is generally little value in keeping conversation surveys available for more than 72 hours.

• Tell respondents how to opt out — Best-practice is to tell respondents how to opt-out in the introduction message. Experience has shown the very few respondents actually opt-out immediately after getting the message. It is also good practice to setup a "Help" conversation (triggered by the "help" keyword) that includes information on how to opt-out.

• Use region-specific opt-out phrases — "STOP" is the most common phrase respondents type to opt-out in the U.S. Other countries have region-specific terms, such as "STOPP" in Norway. Learn the terms for the regions the conversation covers, and add them to the channel.

• Scales as text — For channels that show options as text, such as SMS (as opposed to others that can show buttons), be explicit about what the scale and the options are. Whenever possible it is good practice to spell out all the options available.