I gave a talk recently on conversational commerce — the intersection of chatbots, messaging platforms, and actual revenue generation. Not the “chatbots are the future!” hand-waving that’s been circulating for years. The concrete business case: where the money is, which platforms are winning, and what implementation patterns actually drive results.

The short version? Conversational commerce isn’t speculative anymore. The market was around $41B in 2020 and is projected to hit $290B by 2025. That’s roughly 7x growth in five years. When a market moves that fast, you either figure out your strategy or watch competitors figure out theirs.

The Business Case (With Actual Numbers) #

Let me start with what convinced me this space deserves attention.

Chatbot-driven cost savings hit $11B in 2022. That’s not a projection — that’s the current run rate. Businesses using chatbots for customer service see 30% reductions in support costs. Those deploying them for commerce — not just support, but actual sales — report 67% revenue increases through chatbot channels.

Here’s the stat that surprised me: 88% of customers had at least one conversation with a chatbot in 2022. Not “were offered a chatbot.” Had an actual conversation. Consumer behavior shifted faster than most companies’ technology stacks.

Abandoned cart recovery is where ROI gets specific. Chatbot-driven cart recovery — reaching out via WhatsApp or web chat when someone abandons a cart — can increase sales by up to 25%. Compare that to email recovery campaigns converting at 5-10%, and the channel advantage is clear.

Platform Ecosystem: Where to Build #

I covered the platform landscape in the talk because choosing the right foundation matters more than most teams realize.

WhatsApp Business API is the heavyweight outside North America. WhatsApp updated its pricing in February 2022 to a conversation-based structure — you pay per 24-hour conversation window rather than per message. This aligns incentives: businesses aren’t penalized for being responsive within a conversation. It also makes commerce flows (browse, ask questions, purchase) more economically viable.

For building conversational intelligence itself, the landscape breaks into a few categories:

Dialogflow (Google’s platform, and yes, I’m biased by proximity) offers both CX and ES variants. CX is the enterprise-grade option with visual flow builders, versioning, and environment management. ES is simpler, faster to prototype with, but hits walls at scale. The strength is the NLU engine and integration with Google’s broader cloud ecosystem.

Amazon Lex got a major upgrade with v2, and it’s the natural choice if you’re already on AWS. Integration with Connect for voice and the Lambda backend model is clean. Where it struggles is conversational design tooling — functional but not inspiring.

IBM Watson Assistant has been in this space longer than anyone and it shows, for better and worse. The conversational modeling is mature, but the platform feels heavy. Good for enterprises wanting a batteries-included solution; less ideal for teams that want to move fast.

Rasa is the open-source option and deserves more attention than it gets. If you need full control over your NLU pipeline, want to run on-premise, or have specific data residency requirements, Rasa is often the right call. The tradeoff? You own the infrastructure and the ML pipeline, which requires genuine ML engineering capacity.

Real-World Examples That Actually Worked #

I’m always skeptical of “success stories” in conference talks because they cherry-pick wins. So I tried to include examples where implementation details were public enough to verify.

American Eagle built a chatbot on their app and web presence that handles product recommendations and style advice. The interesting part isn’t the recommendation engine; it’s the conversational UX. Instead of presenting a grid of products (traditional e-commerce), they structured it as a dialogue: “What’s the occasion?” “What’s your style?” “What’s your budget?” Conversion rates on chatbot-assisted sessions were significantly higher than unassisted browsing.

Domino’s has been doing this longer than most and it shows. Their ordering bot works across multiple platforms (web, app, voice assistants) and handles the full purchase journey — customization, payment, tracking. The key insight from Domino’s implementation: they didn’t try to make the bot conversational about everything. It’s very good at ordering pizza. It doesn’t try to chat about the weather. That focus is what makes it work.

The common thread across successful implementations? Narrow scope, clear transactional value, and fallback to human agents when the bot hits limits. Chatbots that fail try to be everything — support agent, sales rep, brand ambassador, therapist — and end up bad at all of it.

Implementation Patterns That Drive Revenue #

From the talk, here are patterns I see working consistently:

Pre-purchase engagement. The chatbot initiates contact based on behavioral signals (time on page, scroll depth, cart contents) and offers help. Not a generic “How can I help you?” but a contextual prompt: “I see you’re looking at running shoes. Want help finding your size?” The specificity makes this work rather than feel intrusive.

In-conversation transactions. The entire purchase flow happens within the messaging thread. Product selection, customization, payment, confirmation — no redirects to a website, no app switches. WhatsApp Business API and Google Business Messages both support this pattern natively now. The friction reduction is measurable; every redirect you eliminate increases completion rates.

Post-purchase follow-up. Order updates, delivery tracking, and (here’s the revenue play) cross-sell recommendations based on what was purchased. “Your running shoes are arriving tomorrow. Want to add socks or insoles to your next order?” This works because the context is warm and the timing is relevant.

Abandoned cart recovery via messaging. I mentioned the 25% lift earlier. The implementation is straightforward: when a cart is abandoned, trigger a message (WhatsApp, SMS, web push) after a configurable delay. Include the cart contents, a direct link to complete the purchase, and optionally a time-limited incentive. The key is timing — too fast feels creepy, too slow and they’ve forgotten what they were buying.

What Most Companies Get Wrong #

The biggest mistake I see: treating the chatbot as a cost center instead of a revenue channel. Companies deploy chatbots to deflect support tickets (valid) but don’t invest in commerce capabilities. Support deflection saves money; commerce generates it. Both matter, but the ROI argument for commerce is stronger and easier to fund.

Second mistake: underinvesting in conversational design. The NLU engine and platform infrastructure are maybe 30% of the work. The other 70% is conversation design — the flows, the copy, the error handling, the personality. Most teams staff this with engineers who don’t have UX writing backgrounds. The result? A chatbot that feels like talking to a command line.

Third mistake: no measurement framework. If you can’t attribute revenue to the chatbot channel, you can’t justify continued investment. Track channel-specific conversion rates, average order values, customer satisfaction scores, and (critically) the handoff rate to human agents. That last metric tells you where your bot is failing and where to invest next.

Where Conversational Commerce Goes From Here #

The market is growing fast enough that even mediocre implementations generate returns. But the bar is rising. Consumers who’ve had good chatbot experiences now expect them everywhere. Consumers who’ve had bad ones are harder to re-engage.

I think the next wave will be driven by richer media within conversations — product carousels, video demos, interactive configurators, all within the messaging thread. The platforms are building support for this; the question is how quickly brands adopt it.

The businesses that win in conversational commerce won’t be the ones with the most sophisticated NLU. They’ll be the ones that understand their customers’ purchase journey well enough to know exactly where a conversational interface reduces friction — and where it doesn’t. Sometimes the best UX is still a well-designed webpage. The chatbot should augment the journey, not replace it entirely.