Conversational AI for retail: selling by phone, text and chat
In retail, the phone line carries the gift orders, the big baskets and the holiday rush, and it’s usually the last channel to get AI. What a retail agent has to do on each channel, the systems it needs, and how Omaha Steaks runs it at peak.
In this guide
Conversational AI in retail is software that talks with shoppers in their own words and then does something about it. It finds the right product, answers the question about sizing or delivery, places the order or moves the delivery date. It powers conversational commerce, and it now works on every channel shoppers use, the phone included.
Most guides on the topic are about website chatbots. For brands that sell by phone, though, the phone carries the gift orders, the larger baskets and the customers who would rather talk than tap, and it’s the first channel to back up when volume spikes.
Chatbots, assistants and agents
| What it does | Where it stops | |
|---|---|---|
| Scripted chatbot | Answers FAQs from a decision tree on the website | Anything off-script, and anything that needs an action |
| Shopping assistant | Helps shoppers search and compare products in chat | Usually the website only, and handing back to the cart |
| Conversational AI agent | Holds the conversation on any channel and acts in your systems: orders, changes, payments | At the rules you set, with a handoff to your team |
The last column is the one to read. A bot that answers “where’s my order?” with a tracking link has deflected the question. An agent that sees the delay, offers a new date and updates the order has resolved it. If it can also recommend and take an order, the same conversation can end in a sale.
Where it pays in retail
Start with phone orders if you take them. A shopper calls to send a gift box; the agent finds the products, builds the order from your catalog, applies the promo code and takes payment before hanging up. Pre-purchase questions are close behind. “Will it arrive by Friday?” or “what’s the difference between these two?” is often all that stands between a browser and a buyer, and the answer has to come from your catalog and policies, not the model’s guess.
Order changes are the third. Delivery dates, addresses and damaged boxes get handled in the order system while the shopper is on the line, with a confirmation text after. If a shopper texts a photo of a crushed box, the agent can read it and send the replacement.
Recommendations work best one at a time. A single suggestion tied to what the shopper just said (“he can’t have sugar, so skip dessert”) beats a carousel of best-sellers. Then there’s follow-up, with callbacks and texts about abandoned carts, delayed shipments and reorders to customers who agreed to hear from you.
All of these depend on one customer record across channels. Someone who calls about a mattress on Tuesday evening, gets the comparison by email, and asks about Saturday delivery in chat on Wednesday morning is having one conversation. An agent with one profile and memory across every channel treats it as one, which is what personalization comes down to.
Peak season is the test
Retail volume isn’t steady. Holidays, launches and promotions bring the year’s most valuable calls at once, which is when hold times climb and callers give up. An agent answers every call immediately, however many come in.
Omaha Steaks sees about 12× its normal call volume in the holidays. With Simple it handles 78% of its sales calls and 72% of all calls, cut call abandonment from 16% to 2%, and has a 30% upsell rate against 22% for seasonal agents. It went live in 8 days, and seasonal hiring is planned down from 5,000 to 692. The Omaha Steaks story has the details.
What it needs from your systems
An agent can only be as useful as what it can see and change. Connect these before launch:
- the catalog, with products, SKUs, prices and approved product claims, so it never invents a price or a promise;
- inventory and delivery dates, so “will it arrive by Friday?” gets a true answer;
- the order system, with permission to place and change orders as well as read them;
- the customer record, with past orders, preferences and earlier conversations, through your CRM or commerce platform;
- promotions and policies, including which codes apply, discount limits and return rules;
- payments, so the card is collected securely in the conversation, masked in transcripts and passed to your payment system;
- your contact center, so a shopper who needs a person reaches one with the conversation on screen.
Why voice is the hardest channel
In chat, a two-second pause goes unnoticed. On a call it’s awkward, and a voice agent that talks over the shopper or stops at every “mm-hmm” sounds broken. Voice needs low latency, careful turn-taking and a model of when the caller has finished speaking. We built our Tango model for this. It recognizes a real interruption about 100 ms after the caller starts talking, and it cut interruptions by 64% against our previous agent in production. The guide to backchanneling has the details.
Shoppers should also get the same service in the language they prefer. Simple speaks 80 languages and dialects.
How to measure it
| Metric | What it shows |
|---|---|
| Revenue per conversation | Whether the agent sells, on every channel |
| Conversion rate | The share of sales conversations that end in an order |
| Upsell rate and order value | Whether recommendations land |
| Containment on sales calls | How many sales the agent finishes without a transfer. Read it next to revenue, never alone |
| Abandonment at peak | Whether shoppers still reach someone on the busiest day |
| Sentiment | How shoppers felt about the conversation, call by call |
Rolling it out before peak
- Start with the calls that carry the most revenue, usually phone and gift orders.
- Connect the catalog and the order system first, since everything else builds on them.
- Test with simulated calls based on last season’s conversations, then with test calls on the live line.
- Go live weeks before the rush, so the agent is tuned on normal volume when peak arrives.
- Score every conversation against your own checklist and review the results with the vendor weekly.
Simple for retail
Simple is one agent on calls, texts, web chat and email, with one profile of every customer. Answer picks up on the first ring through the holiday rush, on the phone system you already own. Close places the order, makes the add-on that fits and collects the payment securely. Book follows up on carts and reorders, and Develop scores every conversation. The retail page shows how the four fit together.