People assume our channel order, voice, then SMS, then the website widget, was a technology roadmap: build the easy thing first and work up to the hard thing. We did roughly the reverse.
We built the hardest channel first on purpose, and the order after that followed from what businesses needed next. It's a decision that's easy to second-guess. A chat widget would probably have got us to market faster, and it would certainly have demoed better. We don't think it would have got us to a better product.
This matters beyond our own history, because every business setting up an AI receptionist faces a smaller version of the same question: which channel do I turn on first? The answer we reached for ourselves, start where the pain is sharpest, applies just as well to a clinic or a plumber.
If you're here for that practical question, the comparison table and the section on choosing your first channel are the parts to read; the rest is the story of how we got there.
What channel order means for an AI receptionist
An AI receptionist can talk to customers in several places. In Aiva Chat's case that's the phone line on a business's own Twilio number, text messages (where the AI can handle appointment texts once Booking by text is on, and a person can click Suggest a reply for anything else), and a chat window on the business's website that takes typed or spoken questions. The order we built them in is the subject of this piece.
Channel order is the sequence in which a product team builds these, or the sequence in which a business turns them on. It sounds like a scheduling detail. In practice it decides which problems get solved properly and which get a rushed first version.
For a product team, the first channel shapes the foundation. Whatever you build first becomes the engine everything else reuses. Build an easy channel first and later, harder channels have to fit a foundation that wasn't designed for them.
For a business, the first channel decides where you see results first. Turn on the channel that matches your biggest leak and the value is obvious within weeks. Turn on the wrong one and you may conclude AI doesn't help, when you were simply solving a smaller problem.
How we decided the order
Step 1: Find where businesses lose the most
Almost every small service business describes the same problem: the phone rings and either nobody's there, or the person who is there is with a patient, a customer, or mid-haircut. A missed call is the most visible form of lost revenue a small business has. You can count it in the call log.
Step 2: Pick the hardest problem to solve first
Voice is the hardest channel by a wide margin. It needs real-time speech, replies fast enough to feel like conversation, interruption handling, and natural speech in the caller's language, all while reasoning about a booking. Text can take a second or two to think without anyone noticing.
Step 3: Build an engine later channels can reuse
If we could solve real-time multilingual conversation under that kind of time pressure, everything after it would inherit the same assistant, knowledge, and rules and be comparatively easier. We were paying down the hardest debt first.
Step 4: Add the channel that closes voice's gaps
Voice handles the moment someone calls. It can't help the customer who won't call, or the one who needs a follow-up after hanging up. That pointed to SMS next.
Step 5: Give the widget dedicated attention
The widget was a different kind of problem, embedding rather than conversation, and it deserved focus instead of being squeezed in alongside voice.
The three launches
Voice first: the hardest and most universal
Our first product had to work in more than one language, including Hindi and Gujarati alongside English, almost from the start. A voice product that only spoke English wasn't solving the missed-call problem for many of the businesses we were building for.
That made voice even harder to ship first, and made everything afterward easier, because the language infrastructure was already carrying weight.
SMS second: the conversation after the call
SMS reached the customers voice couldn't: people who prefer not to call, and people who need a quick follow-up after a booking. It reused the same business knowledge, which made it a smaller lift than the widget.
SMS turned out to be voice's unfinished sentence: the part of the conversation that happens after the caller hangs up.
"Smaller" still meant real work. Handling long threads, opt-outs, photos, and voice notes was its own problem, as we describe in our look inside the SMS engine. We also made texting suggest-only at first: Aiva Chat drafted, a person sent. Booking by text came later, so the AI can now book, move, and cancel appointments by text once a business turns it on. Carriers treat business texting carefully, and Twilio's Programmable Messaging documentation shows how much sits behind a single business text.
The widget third, on purpose
We could have built the widget first. It's the obvious AI product and demos beautifully. We didn't, for two reasons.
First, clinics, salons, and local service businesses usually aren't limited by website traffic. A beautiful chat widget on a site with forty visits a week solves a smaller problem than a phone line that rings sixty times a day.
Second, the widget is an embedding problem. It has to render correctly on many kinds of sites without breaking a business's existing styles or scripts, and it should stay out of the way until it's useful. Getting it down to a single script tag took real iteration.
What hardest-first gave our customers
One assistant across every channel
Because later channels reused the voice foundation, they share the same knowledge, rules, and corrections. Update a price once and it's correct on calls, in chat, and in SMS drafts. We explain this in one brain, every channel.
Multilingual from the start
Language support was built into the first and hardest channel, so the widget and SMS didn't have to retrofit it. All three serve customers in the same 13 languages.
A widget that respects visitors
Giving the widget its own focus meant it stays closed as a bubble until clicked, with an optional auto-open delay the business chooses, instead of jumping in front of every visitor.
No channel sold separately
Because the channels share one foundation, every plan includes all three. A business doesn't pay extra to add SMS or the widget later.
Voice vs. SMS vs. website widget
| Voice | SMS | Website widget | |
|---|---|---|---|
| Best for | Calls you can't answer, urgent questions | Follow-ups, customers who prefer texting | Website visitors researching before they call |
| Answers on its own | Yes | Appointment texts, once Booking by text is on; drafts other replies for a person to send | Yes |
| Books appointments | Yes, from Aiva Chat's appointment book | Yes, once Booking by text is on | Yes; text chat checks Google Calendar free/busy |
| Handles photos and voice notes | N/A | Reads photos, transcribes voice notes | No |
| Getting a person involved | Live transfer to your team | Handed to your team in the inbox; staff can take over any thread | Optional pre-chat form captures contact details |
| Hardest part to build | Real-time speech and response speed | Threads, opt-outs, carrier rules | Embedding safely on any website |
Voice carries the most weight for most service businesses, because the phone is still where urgent and high-value conversations happen. It's also the only channel with live transfer to staff.
SMS is deliberately the most conservative channel. The AI only answers texts once you turn on Booking by text, and texts it can't handle go to a person, which suits confirmations, follow-ups, and anything where a written record matters.
The widget is strongest where your website already gets traffic. For businesses whose enquiries mostly start online, it can be the most valuable of the three, which is exactly the trade-off our ordering asked them to wait for.
How to choose which channel to turn on first
Step 1: Count where enquiries come from
Look at a month of call logs, missed calls, website form submissions, and texts. The biggest number, or the biggest leak, is your starting point.
Step 2: Start with voice if the phone is the leak
If missed calls are your problem, connect your Twilio number and set up the assistant. Our guide to setting up your first call flow walks through it.
Step 3: Start with the widget if your website is busy
If visitors arrive but don't call, paste the widget snippet into your site and let text and voice chat answer questions and book.
Step 4: Add SMS for follow-ups
Once bookings flow, use SMS for follow-ups and for customers who text instead of calling: turn on Booking by text so the AI handles appointment texts, and use drafts for the rest. Register for A2P 10DLC if you're texting US numbers.
Step 5: Keep one set of business information
Teach the assistant once, in the AI Training Center. Every channel uses it.
Step 6: Review results per channel
Check bookings, topics, and handoffs by channel after a few weeks, then decide what to expand.
Mistakes in channel strategy
Choosing the channel that demos best
A widget looks impressive in a five-minute pitch. If your customers phone you, it solves a smaller problem. Choose by where the business is lost, not by what looks most like AI.
Turning everything on at once without a plan
All three channels are included, but launching them together means testing and reviewing three things at once. Start with the biggest leak, get it right, then add the next.
Expecting SMS to work like voice
In Aiva Chat, the AI answers texts only once you turn on Booking by text, and even then texts it can't handle go to your team. If you expect every text to be answered automatically, you'll be surprised. Plan who on your team reviews and sends drafts.
Maintaining separate information per channel
Separate tools with separate FAQs drift apart within weeks. One shared knowledge base avoids customers hearing one price on the phone and another in chat.
Lessons from the order
- Order by pain. Start where business is visibly lost, even if it's the harder build.
- Solve the hardest problem first. Everything after it gets easier.
- Give different problems dedicated focus. Embedding and conversation engineering need different skills.
- Be honest about who waits. Our order meant website-first businesses waited for the widget.
- Reuse the foundation. One assistant and one knowledge base across channels.
- Match your order to your own bottleneck. Your first channel may differ from ours.
Why the order wasn't obvious at the time
None of this was as clean in the moment as it reads now. The argument kept coming back, usually whenever a competitor announced a chat product. Each time, the answer held: the phone was still carrying the most volume and the most visible pain.
The order stopped being a live debate once SMS shipped and it became clear how much of its value came from reusing voice's foundation. That was when we could point to "hardest first" as a strategy that had paid off.
Conclusion
We'd make the same call again. Hardest-first cost us time to market and made website-first businesses wait, but it meant every later channel was built on an engine that had already survived real-time, multilingual phone calls. Our voice platform is where that foundation started.
For your business, the decision is smaller and more practical. Pull a month of call logs, missed calls, and form submissions, and turn on whichever channel matches the biggest number. The one caveat worth planning for is SMS: even with Booking by text on, texts the AI can't handle wait for a person, so decide who owns the inbox before switching it on.
Ready to start with your biggest channel?
The channel we built first is the easiest to try: call +1 (970) 972-0974 and ask it something your customers ask you.
Whichever channel you start with, the other two are already in your plan for later, Starter at $49 included; the plan limits show what changes between tiers. Calls, texts, and AI run on your own Twilio and OpenAI accounts.
Turn on the channel that's leaking customers with a 14-day trial, no card needed, or book a demo to decide which one that is.
Posts from the people building Aiva Chat in Rajkot — how the product works, what we have learned building it, and guides for getting the most out of an AI receptionist.