Use cases

Honour opt-outs however the customer words them

Prismlet reads each reply your keyword rule did not catch and answers whether it asks you to stop, however it is worded, and how far the request goes. Replies where the wish to stop is unclear are marked for a person to read.

Ask for an invite

Today

  • An exact-keyword rule catches STOP, UNSUBSCRIBE and their usual variants. Every other reply is read by a person, often well behind the message it answers.
  • Under the US rule on consent to automated calls and texts, a customer can revoke consent in any reasonable manner, not only with a fixed list of words, so "take me off your list" or "wrong number, don't text" can count without a keyword.
  • A reply that says the sender already asked, for example "third time asking", is easy to miss when replies sit in a shared inbox instead of a record tied to the contact.
  • A reply that says the sender never signed up, or that you reached the wrong person, needs a different step from a plain stop and is easy to file with the rest.

What changes

  • Every reply gets the same questions, answered from its text, after your keyword rule has already run.
  • A reply that asks to stop, however it is worded, is read for how far it goes: all messages, one channel, marketing only, or one series.
  • A reply where the wish to stop is unclear, for example a question about how to stop, a condition or sarcasm, is marked for a person to read instead of being read as a stop or not.
  • A reply that mentions a lawyer, a regulator or a spam report, or says the sender asked before or never signed up, is flagged for your compliance team to look at.
  • Each answer carries how sure it is, so the unsure ones can reach a person the same way the flagged ones do.

Try it on this example

Example · Stop the sale texts, keep the shipping updates

Channel: SMS

Our message they replied to: Harbor & Pine: 48 hours only! 30% off all outdoor furniture, today and tomorrow. Shop now: harborpine.example/s30 Reply STOP to end texts.

Reply

ok enough with the sale texts, I only want the shipping updates for my patio set order not this every day. already told you guys twice. also tracking has said out for delivery since Tuesday and nothing's come??
  1. Does the reply ask us to stop sending some or all messages?Yes94%
  2. If the reply asks us to stop, which messages does it ask us to stop?Marketing only93%
  3. Is it unclear from the reply whether the sender wants messages to stop?No79%
  4. What is the reply mainly about?Opt-out request94%
  5. Does the reply say we reached the wrong person?No96%
  6. Does the reply say the sender never signed up or never gave permission to be contacted?No96%
  7. Does the reply say the sender has asked us to stop before?Yes97%
  8. Does the reply mention a lawyer, a lawsuit, a regulator or a spam report?No92%
  9. Apart from any request to stop, does the reply raise a question or problem a person should answer?Yes98%
  10. Is the reply abusive or threatening toward the company or its staff?No93%

These are real answers stored from one run on this example.

Test it on your own data

The quickest way to decide is a file run on your own records. A file run in the app answers up to 10,000 rows and gives you the file back with the answers added.

What to send
A CSV or TSV file with column names in the first row, or a JSON array of objects, one item per row. You pick the column for each thing the prism reads: Channel; Our message they replied to; Reply. A row's fields together take up to about 100,000 characters.
What comes back
The same rows with the prism's columns added: revokes_consent, revokes_consent_probability, stop_scope, stop_scope_probability, stop_request_unclear, stop_request_unclear_probability, reply_kind, reply_kind_probability, wrong_recipient, wrong_recipient_probability, never_consented, never_consented_probability, says_asked_before, says_asked_before_probability, legal_or_regulator, legal_or_regulator_probability, needs_human_reply, needs_human_reply_probability, abusive, abusive_probability. Each answer says how sure it is, so the unsure rows are easy to find and give to a person.
Against your team's answers
Add columns with the answers your team gave, and the app reports how often Prismlet agreed, question by question and at each cutoff, split by a column such as language or site, with every miss listed. Your team's answers stay in your browser.

How it reaches your team

Works today

  • Call the API from your own system as each reply arrives
  • Run a file of up to 10,000 replies in the app, one reply per row, up to about 100,000 characters a row

Built with you

  • We work with you to send each reply from your SMS or email platform as it arrives
  • We work with you to put the unsure and flagged replies where your team already reviews them
  • We work with you to feed each answer into the consent records you keep, for a person to confirm the stop and how far it goes

The prism behind it

Honour opt-outs however the customer words them10 questions

Fields

  • Channel
  • Our message they replied to
  • Reply

Context

These are replies from customers to the messages a retailer sends by text and email: marketing offers, order and delivery updates, and account alerts. Each reply is read before the next message in the sequence goes out. Our exact-keyword rule (STOP, UNSUBSCRIBE) runs first; these replies are the ones it did not catch. Rules we follow: - United States, texts and automated calls: under the TCPA a consumer may revoke consent in any reasonable manner. Words such as stop, quit, end, revoke, opt out, cancel and unsubscribe always count, and other words that make the wish clear count too. A revocation must be honoured within ten business days. - United States, email: under CAN-SPAM an opt-out from commercial email must be honoured within ten business days. - UK and EU: under PECR, the ePrivacy rules and Article 21 of the GDPR, a person who objects to direct marketing must not be sent it again, however they word the objection. - Messages about an order, a delivery or an account are service messages, not marketing. Legal decides how widely a stop applies. These questions report only what the reply asks. Our code keeps the consent records, the deadlines and the audit log.

Questions

  1. Does the reply ask us to stop sending some or all messages? Yes / No

    Read the reply in light of the message it answers. Count any wording that makes the wish to stop clear, such as "stop", "remove me", "no more of these", "wrong number, don't text", "leave me alone", "take me off your list", or insults that demand no contact. Do not count a complaint about the content alone, or a stop word used in another sense, such as "can't stop buying these". Yes: The reply asks us to stop some or all messages. No: The reply does not ask us to stop anything.

  2. If the reply asks us to stop, which messages does it ask us to stop? Choice

    Pick the scope the reply asks for. When the reply asks to stop but is not clear how far the stop goes, choose All messages. Choose No stop request when the reply does not ask us to stop anything.

    • All messages Stop everything from us, on every channel.
    • This channel Stop texts, or stop emails, on the channel the reply came from, with no objection to other channels.
    • Marketing only Stop offers and promotions, but keep order, delivery, appointment or account messages.
    • One series or topic Stop one named series, product, store or event, and nothing wider.
    • No stop request The reply does not ask us to stop any messages.

  3. Is it unclear from the reply whether the sender wants messages to stop? Yes / No

    Look for sarcasm, a question about how to stop, a condition such as "stop unless you can match the price", mixed signals, or text in a language you cannot read with confidence. Yes: A person would need to read the reply to know whether the sender wants messages to stop. No: The reply clearly asks to stop, or clearly does not.

  4. What is the reply mainly about? Choice

    Pick the one option that best describes the reply. If the reply asks to stop any messages, choose Opt-out request, even when it also asks or complains about something else.

    • Opt-out request The reply asks us to stop some or all messages.
    • Question or request The reply asks something or asks for something that needs an answer from us, and does not ask to stop.
    • Interest The reply wants to buy, book or learn more.
    • Complaint The reply is unhappy with us, an order or the messages, without asking to stop.
    • Wrong person The reply says we reached someone other than the intended person, without asking to stop.
    • Automated reply An out-of-office, an auto-responder, or carrier or bounce text.
    • Acknowledgement Thanks, ok, an emoji, or nothing else to act on.

  5. Does the reply say we reached the wrong person? Yes / No

    Count a sender who says they are not the person named, or that the intended person has moved, died, left, or no longer uses this number or address. Yes: The reply says our message reached someone other than the intended person. No: Nothing in the reply says we reached the wrong person.

  6. Does the reply say the sender never signed up or never gave permission to be contacted? Yes / No

    Count "I never signed up", "I never gave you this number", "who is this?" said as a denial, and "I don't know who you are". Yes: The reply says the sender never agreed to our messages or does not know us. No: The reply does not deny signing up.

  7. Does the reply say the sender has asked us to stop before? Yes / No

    Count "third time asking", "I already unsubscribed", "I told you last week", "how many times do I have to say this". Yes: The reply says an earlier request to stop was made. No: The reply does not mention an earlier request to stop.

  8. Does the reply mention a lawyer, a lawsuit, a regulator or a spam report? Yes / No

    Count a lawyer, suing, the TCPA, the FCC, the FTC, a state attorney general, the ICO or another data protection authority, or reporting us as spam. Yes: The reply mentions legal action, a regulator or a spam report. No: None of these is mentioned.

  9. Apart from any request to stop, does the reply raise a question or problem a person should answer? Yes / No

    Count a question about an order, a delivery, a charge, an account or a booking, or a problem the sender wants fixed. Yes: The reply contains a question or problem a person should answer. No: Beyond any request to stop, there is nothing to answer.

  10. Is the reply abusive or threatening toward the company or its staff? Yes / No

    Swearing out of frustration alone is not abuse. Count insults aimed at staff, threats of harm, or hateful language. Yes: The reply is abusive or threatening. No: The reply is not abusive, even if it is blunt or annoyed.

Lens columns

revokes_consent, revokes_consent_probability, stop_scope, stop_scope_probability, stop_request_unclear, stop_request_unclear_probability, reply_kind, reply_kind_probability, wrong_recipient, wrong_recipient_probability, never_consented, never_consented_probability, says_asked_before, says_asked_before_probability, legal_or_regulator, legal_or_regulator_probability, needs_human_reply, needs_human_reply_probability, abusive, abusive_probability

This prism on its own page

See it on your own records

Prismlet is invite only for now. Tell us what you would send it and we will set you up.

Ask for an invite