Everyone has AI. Almost nobody has connected it.
98 percent of companies have deployed AI somewhere, but only 15 percent connect it to the systems around it. That gap decides who gets a return and who is left counting deflected conversations.
Author: Heyloha Team
Two numbers that sum up the whole problem
Research by Talkdesk presented this autumn found that 98 percent of companies have deployed AI somewhere along the customer journey. Only 15 percent combine that AI with orchestration across departments.
That is the real story of AI in customer contact in 2026. Not whether companies use AI, because nearly all of them do. The question is whether that AI is attached to anything.
The same research found that companies combining AI with that orchestration were four times more likely to report major gains in customer satisfaction. That result belongs to the 15 percent, not the 98.
What 'not connected' means in practice
An AI that is not connected can talk but cannot do.
Take a customer calling to move an appointment. A standalone AI understands the request perfectly, produces a polite reply, and says a colleague will be in touch. The customer now has a promise instead of an appointment, and you have an extra task.
A connected AI looks at the calendar, sees which slots are free, moves the appointment and sends the confirmation. The customer has a new appointment and no task was created.
The same goes for leads. An AI that qualifies a lead beautifully but leaves it in a chat log has solved nothing. An AI that writes that same lead into your CRM, answers attached, has taken work off your desk.
The difference is not the quality of the language model. It is the integrations.
Why deflection is the wrong measure
The most commonly used metric for AI in customer contact is deflection: how many conversations never reached a human?
The number is seductive because it always looks like good news. But a conversation can avoid a human for two very different reasons. Because it was resolved, or because the customer gave up.
Deflection adds those two together. That is why a mediocre AI often looks fine in a report while your customer satisfaction slides.
Measure whether something happened instead. How many appointments are in the calendar? How many leads are in your CRM? How many customers came back within a week with the same question? Those numbers do not lie.
What orchestration means when you are not a contact centre
Orchestration sounds like something for companies with a department dedicated to it. For a business of fifteen people it means something much simpler: can the AI reach the things that hold the answer, and the things the answer has to go into?
For most smaller businesses it comes down to three integrations, and together they cover the bulk of the gain.
Your calendar. Without it an AI can never finish an appointment, and every conversation ends in 'we will get back to you'. With it, a request for an appointment becomes an appointment.
Your CRM or email. A lead that lands nowhere is not a lead. This is the integration that pays back fastest, because you can weigh it against the value of a single lost customer.
Your own knowledge. Website, prices, terms, common questions. Without that source an AI invents plausible answers, which is worse than no answer.
Start at the end of the conversation
The usual order when setting up AI is: first make it talk well, then see what it can do. Reverse that.
Write down first what needs to have happened by the end of a good conversation. An appointment is booked. A lead is logged. A quote has gone out. A customer knows when their order arrives.
Then work backwards: which integration does that need? It is a short list, usually three or four things.
That is how you end up in the 15 percent without ever writing an orchestration strategy. You simply refused to settle for an AI that only talks.
Talking became the easy part
Five years ago an AI that sounded natural was the hard part. It is not any more. Every serious platform today delivers an agent that speaks well and understands a question.
So the distinction has moved. Not how well the AI talks, but how much it can finish.
In Heyloha that side is built in: calendar integration for appointments, leads passed on, warm transfer to a colleague with context attached, and answers drawn from your own website and documents rather than a general model.
Want to see what your agent can actually complete? Try Heyloha free for 14 days. Chat is live within about 30 minutes.
Frequently asked questions about AI orchestration
How many companies use AI in customer contact?
Nearly all of them. Research by Talkdesk found 98 percent of companies have deployed AI somewhere along the customer journey. Only 15 percent combine that AI with orchestration across departments.
What does orchestration mean in AI customer contact?
That the AI is connected to the systems holding the answer and receiving the result: your calendar, your CRM and your own knowledge base. Without those integrations an AI can understand and answer a question but cannot finish anything.
Why is deflection a misleading metric?
Because a conversation avoids a human for two reasons: it was resolved, or the customer gave up. Deflection counts both as success. Measure booked appointments, logged leads and repeat questions within a week instead.
Which integrations deliver most for a small business?
Three. Your calendar, so an appointment request becomes an appointment. Your CRM or email, so leads land somewhere. And your own website and documents as the knowledge source, so the AI gives answers that are correct instead of inventing plausible ones.
Does connecting AI to your systems actually pay off?
According to the same Talkdesk research, companies that combine AI with orchestration across departments were four times more likely to report major gains in customer satisfaction. The gain goes to companies that connected the AI to their systems, not to those that only dropped a standalone chatbot onto a page.