- Conversational SMS means someone, or something, actually reads and answers what subscribers text back. It pays off where a question stands between a shopper and an order: considered purchases, higher order values, fit and compatibility doubts, and repeat or subscription orders.
- Each conversation has a cost (agent time or AI fees, plus messages both ways), so judge it on contribution per conversation and test it against a holdout group, not on reply rate.
- Two-way texting adds compliance work: free-text opt-outs, clear consent for any marketing in replies, and honesty about bots. This covers practical basics; confirm your setup with counsel.
What is conversational SMS?
Most eCommerce texting is one-way. A campaign or a flow goes out, a share of people click, and replies land in an inbox nobody checks. Conversational SMS treats the reply as the point. Messages invite a response, and every response gets a useful answer within minutes.
There are three common ways to staff it:
- Human agents. Your support team, or an outsourced team trained on your products. Postscript, for example, offers a service with human agents for one-to-one sales conversations.
- AI agents. Software that reads the message, checks your catalog, policies and order data, and replies. Postscript’s Shopper is one example from an SMS platform, and Attentive announced AI features for its Concierge product in 2023.
- Hybrid. AI answers routine questions and hands anything sensitive, unclear or high value to a person.
A related format is text-to-buy, where a subscriber replies to a message to place an order. Attentive and Postscript have both launched text-to-buy features with Shop Pay, and subscription brands can let customers reorder or skip by reply through their subscription app or SMS integration.
When does two-way texting pay off?
Two-way texting earns its cost when an answer changes the outcome. If the shopper would have bought anyway, or would never buy, the conversation is pure cost.
| Moment | Example question or reply | Pays off when | Usually not worth it when |
|---|---|---|---|
| Pre-purchase questions | “Will this fit a 6-foot bed?” | Considered purchases, sizing, compatibility, higher order values | Low-price impulse items |
| Cart and checkout recovery | “Does the code work on sale items?” | The cart is large and the objection is answerable | The only objection is price |
| Product guidance | “Which serum for dry skin?” | Wide catalogs where choice causes hesitation | Small catalogs with one obvious pick |
| Order and delivery questions | “Where is my order?” | It replaces a costlier support channel | You can’t answer faster than the tracking page |
| Replenishment and subscriptions | “Skip next month” or “Send it again” | Repeat-purchase products with a known cycle | One-time purchases |
| VIP and win-back | “What’s new since my last order?” | High-value customers with a buying history | Low-value or one-time buyers |
The pattern is consistent. Conversations pay where average order value is high enough to cover the cost, where a real question blocks the order, and where the answer builds a habit of buying again. They rarely pay as a general broadcast channel.
Outbound prompts that invite replies
You don’t have to wait for questions. A campaign can ask one: “Tell us your skin type and we’ll pick your routine,” or “Reply with the size you wear and we’ll check stock.” The answers are zero-party data you can store on the profile and use in later segmentation, which makes them valuable even when the conversation doesn’t end in an order that day.
What does a conversation cost, and how do you know it’s worth it?
Count the full cost of a conversation, not just the outbound text:
- People or software. Agent time per conversation, or the AI product’s fee, which may be a platform fee, a per-conversation charge or a share of revenue.
- Messages both ways. Check how your platform bills inbound replies and multi-segment answers. Long, friendly answers can run to several segments each.
- Discounts. Conversations that end with a code to close the sale carry the code’s cost.
Worked example: break-even per conversation
All numbers are illustrative.
A furniture accessories store considers adding two-way SMS to its checkout abandonment flow. Average order value from these customers is $150 and contribution margin is 35%, so each extra order is worth $52.50 of contribution.
- Cost per conversation, including agent time and messages: $3.
- The store expects 1,000 conversations a month, so the program costs $3,000 a month.
- Break-even: $3,000 ÷ $52.50 = about 57 extra orders a month, or roughly 5.7% of conversations turning into an order that would not otherwise have happened.
The key word is “extra.” Many people who reply to a cart text were going to buy anyway. To know whether the program clears 5.7%, the store sends half of eligible shoppers into the conversational flow and the other half into the standard one-way flow, then compares order rates. That comparison, a simple holdout group, is the only number that settles the question. Our guide to incrementality testing explains how to size and read one.
Should replies come from people, AI or both?
The right model depends on volume, question complexity and what a wrong answer costs you.
| Model | Strengths | Weaknesses | Fits |
|---|---|---|---|
| Human agents | Judgment, empathy, handles unusual cases | Cost per conversation, slower at night and in peaks | High-value products, VIP programs, low volume |
| AI agent | Instant replies at any volume, consistent policies | Can be confidently wrong, needs good source content | High volume of repetitive questions |
| Hybrid | Fast on routine questions, people on the hard ones | Needs clear handoff rules and someone on call | Most growing stores |
If you use AI, set guardrails before launch:
- Source content. The agent is only as accurate as the product data, FAQs and policies it reads. Fix gaps in your product pages first; they are the same gaps.
- Discount rules. Decide what the agent may offer, to whom and how often. An agent that hands a code to anyone who hesitates trains customers to hesitate.
- Escalation triggers. Complaints, damaged orders, refunds, health or safety questions, and any message the agent isn’t confident about go to a person.
- Tone and length. Short answers, one question at a time, and your brand name early in the thread.
What changes for compliance when texting is two-way?
Two-way texting raises questions that one-way campaigns don’t. These are practical basics, not legal advice, and you should confirm your setup with counsel.
- A reply is not marketing consent. Answering a question a customer texted you is different from adding them to marketing. If the conversation could lead into promotional messages, make sure the person has opted in to marketing texts first.
- Opt-outs can arrive in any wording. Under FCC rules in effect since April 2025, consumers can revoke consent by any reasonable means. In a conversation that means “please stop texting me” or “I’m done with this” mid-thread. Human agents and AI agents both need to recognize it and trigger the opt-out, not argue or keep selling. See our SMS compliance guide for the full rules.
- Quiet hours still matter. A shopper may text you late at night. Decide with counsel how you handle replies outside your marketing hours, and keep any promotional content inside them.
- Be honest about bots. Some states have bot-disclosure laws, California’s being the best known, and more rules on AI are being proposed. Whether they reach your SMS program is a question for counsel. The simple practice is to say plainly when a customer is talking to an automated assistant and to offer a person.
- Keep sensitive data out of texts. Never ask for full card numbers or passwords by SMS. Text-to-buy features work through stored payment methods for this reason.
- Match your carrier registration. If you registered your number for marketing, check that customer care conversations are covered by your registered use case, or update it through your platform.
How do you measure conversational SMS?
Reply rate and sentiment are easy to report and say little about profit. Use these instead:
- Response time. Median minutes to first reply, split by hours of the day.
- Resolution rate. Share of conversations closed without a handoff or a follow-up ticket.
- Conversion within a set window. Orders within, say, 7 days of the conversation, compared with the holdout group.
- Contribution per conversation. Incremental orders × contribution margin per order, minus cost per conversation.
- Opt-outs after a conversation. A rising rate means the tone or frequency is wrong.
- Questions by topic. A monthly count of what people ask. The top five belong on your product pages and in your FAQs.
Link conversation records to the customer profile so these effects show up in repeat behavior and customer lifetime value, not just first orders.
Checklist: launching two-way SMS
- [ ] One moment chosen to start (for example, checkout abandonment or pre-purchase questions), not the whole program.
- [ ] Cost per conversation estimated, including messages both ways.
- [ ] A holdout group set up to measure extra orders, not total orders.
- [ ] Reply staffing defined: human, AI or hybrid, with hours and handoff rules.
- [ ] AI source content reviewed: product data, FAQs, shipping and returns policies.
- [ ] Discount rules written down for agents and AI.
- [ ] Free-text opt-outs tested with several wordings, including mid-conversation.
- [ ] Bot disclosure and an easy path to a person in place.
- [ ] Carrier registration and marketing consent checked against the new use case.
- [ ] Monthly review of questions by topic feeding product page and FAQ updates.
For how two-way texting fits into the wider program, see our guide to SMS marketing for eCommerce.
FAQ
Is conversational SMS only for large brands?
No, but it has to be staffed. A small store can start with one flow, such as checkout abandonment, answered by the support team during business hours. What doesn’t work at any size is inviting replies nobody answers.
Can AI fully replace human agents for SMS conversations?
For routine questions such as order status, sizing from a chart or shipping times, AI can handle most of the volume if your product data and policies are accurate. Keep people on complaints, refunds and anything the AI isn’t sure about, and review a sample of AI conversations every week.
How do I know if two-way SMS is making money?
Compare a group that gets the conversational flow with a holdout group that gets the standard one-way flow. If the extra orders, multiplied by contribution margin per order, exceed the cost of the conversations, it pays. Platform-reported “conversation revenue” alone counts orders that would have happened anyway.