Blog

Insights and guides on AI agents and conversation automation.

Top models get it wrong 22% to 94% of the time: why your AI makes things up

An AI that invents a price for your customer is not a problem with the model's brand. It is a problem with where the answer comes from, and that part you control.

Charged per conversation or per message? The fine print that changes your bill

When you compare two quotes for handling WhatsApp, you are almost never comparing the same thing. One counts conversations, another counts messages, another counts AI replies. How to read each unit.

WhatsApp Business App vs API: the hour of the day the free app stops being enough

The WhatsApp Business app is free and covers more ground than most people assume. What breaks it is not message volume. It is the number of people answering the same number at the same time.

"My business is very specific": how to make the AI answer from your own documents

Every business has rules that are not on the internet: your returns policy, your lead times, your exceptions. That is exactly what you load into an AI so it stops improvising.

Do you have to tell customers they are talking to an AI?

The legal answer and the practical answer rarely line up. In Argentina, where we looked at it closely, no rule in force requires the disclosure. There is still a good reason to make it.

Customer service KPIs: the five numbers to measure before you automate

A global survey of 321 support leaders found that 91% are under pressure to implement AI in 2026. The problem with rushing is that afterwards nobody can say whether anything improved. These are the five numbers worth having before you touch anything, and how to get them with no tooling at all.

Retail automation: what to automate in customer service and what to leave alone

Retail automation is not one thing: it is four separate fronts with four different tools. This piece separates them, digs into the only one that handles conversations, and marks where automation should stop and a person should appear.

Customer service for online stores: who answers the question that comes before the purchase

US retail e-commerce reached $326.7 billion in the first quarter of 2026 alone, 16.9% of everything sold at retail, growing at 9.8% against 3.9% for retail overall. Behind almost every one of those orders there was an earlier question: size, shipping, stock, warranty. That enquiry decides the sale, and it is the one that usually goes unanswered.

Customer service for retail stores: the growth is in the channel you answer by typing

In the first quarter of 2026, US retail e-commerce grew 9.8% against the same quarter a year earlier. Total retail grew 3.9%. Online is now 16.9% of everything sold at retail, and it is pulling away at more than twice the speed. The customer did not go anywhere: the growth moved to the channel you handle by typing.

Who handled that customer? The case nobody owns is the one you lose

When three people share the same channel, a problem shows up that has nothing to do with speed or tone: nobody knows who owns each case. The result is not a weak reply, it is no reply. How assignment and a thread you can read end to end fix it.

How many support requests can AI actually resolve? What self-service promises and what one real account delivered

A global survey of 5,728 customers found that only 14% of support issues are fully resolved in self-service. In one account we run, with 312 conversations a month, AI resolves around 50% and hands off 40% to a person. The difference is not the software: it is what you asked it to resolve.

How long it takes to get your WhatsApp handled by AI: instantly if it exists, two weeks if it is custom

The question that comes up most before signing is how long it takes. The answer has two branches: if what you need already exists, testing starts right away; if it takes custom development, the typical turnaround is around two weeks. What almost nobody tells you is that the slowest clock is not ours, it is Meta's.