Chat, Email and Messaging Quality
Text channels remove the transcription problem entirely and introduce their own. Most programmes apply a voice scorecard to them, which measures the wrong things.
Non-voice interactions are a growing share of contact centre volume and they are usually assessed with a form designed for calls.
What changes
No transcription. The text is the text. Every accuracy problem described elsewhere on this site disappears, which makes text analytics substantially more reliable than speech analytics.
No acoustic measures. Silence, talk-over and pace do not exist.
Concurrency. An agent handles several chats at once, which changes what response time means and what a long gap indicates.
Asynchrony. Email and messaging conversations span hours or days, and "the interaction" has no clean boundary.
A permanent artefact. The customer keeps the transcript, which raises the stakes on what is written.
Templates and macros, which are the dominant quality issue and have no voice equivalent.
What a text scorecard should assess
Accuracy of the information, as with voice.
Whether the customer's actual question was answered, which is the most common failure in text and is easy to check because the question is written down.
Appropriate use of templates. A macro sent without editing, answering a question the customer did not ask, is the characteristic text failure.
Response latency, in context of concurrency.
Readability. Sentence length, jargon, structure. Measurable directly, unlike anything in voice.
Closure. Whether the interaction ended with the customer knowing what happens next.
Not: tone items imported from a voice scorecard, which assess something that does not transfer cleanly.
The template problem
Macros are necessary at volume and they are the main source of poor text quality.
Sending an unedited template that addresses a different question is the failure customers complain about most.
It is detectable automatically. Compare the template text to the sent message: an edit distance of zero on a substantive macro is a flag worth reviewing.
Track template usage by macro. A macro that is always sent unedited and generates repeat contact is a bad macro, not a bad agent, and this is a process finding available with almost no effort.
What analytics does well on text
Everything it does on voice, more reliably, because the input is exact.
Categorisation with far higher precision.
Detecting unanswered questions by comparing the customer's questions to the agent's response.
Readability scoring, which is straightforward and genuinely useful.
Template detection and edit measurement.
Sentiment, with the same conceptual caveats but without transcription error compounding them.
Concurrency and its effects
An agent handling four chats writes worse than one handling two. This is measurable and it is a staffing decision presented as a quality problem.
Response gaps during concurrency are not inattention.
Score against the concurrency level, or the measure ranks agents by how busy they were.
Track quality against concurrency to find where the trade-off sits for your operation. Most have never measured it and set concurrency limits by intuition.
Email specifically
Longer form, higher stakes, and permanent.
The failure mode is not answering the question, frequently because the email contained three and one was addressed.
Count the questions asked and the questions answered. This is mechanical, it is the single most useful automated check on email quality, and almost nobody does it.
Thread quality matters more than message quality. A programme scoring individual messages misses the interaction.
The practical arrangement
A separate scorecard per channel, sharing the compliance section and differing on the behavioural items.
Analytics applied more aggressively to text, because it is more reliable there.
Concurrency recorded with every text evaluation.
Template performance tracked as a process measure rather than an agent one.
The unanswered-question check
A mechanical check on email and chat quality that most operations have never run.
Extract the questions from the customer's messages — sentences ending in a question mark, plus common interrogative patterns.
Check whether each is addressed in the agent's response.
Count the unaddressed ones.
This is straightforward on text and it is the single most common failure in email handling: three questions asked, one answered.
Report it per queue and per template. A macro that answers one of the three common questions in its topic is a process finding.
Sample by hand to validate the automated matching, which will be imperfect but is usable at aggregate level.
Track it as a quality measure alongside the scorecard. It is objective, it relates directly to repeat contact, and it cannot be gamed except by answering the questions.