Why we don't ship a chatbot

Tom Johansson
•

Why we don't ship a chatbot

Tom Johansson
•

Why we don't ship a chatbot

Tom Johansson
•

When we started Maya, the same question came up in the first three sales calls: is it like a chatbot for finance?
No. And the difference matters more than it sounds.
Chatbots wait for prompts. Agents do work.
A chatbot is reactive. It needs you to type a question. The best ones can summarize a vendor's payment history or draft a reminder email. That's useful. It's also not what's clogging your team's calendar.
An agent is the opposite. It's a process running against your real systems — reading inboxes, matching invoices, posting entries. You don't open an app to use it. You notice it's done.
The work is in the doing, not the asking
The bottleneck in finance isn't "I have a question and need a fast answer." The bottleneck is "I have 800 invoices and not enough hours."
A chatbot helps the first problem. It doesn't touch the second. We built Maya for the second.
What this means in practice
There's no chat window in Maya. There's a queue, a log, and an exception list. The queue clears overnight. The log shows you exactly what happened. The exception list is short, judgment-heavy — the work that needed a human anyway.
If you want a chatbot for finance, there are good ones. If you want the work done, that's a different product.
When we started Maya, the same question came up in the first three sales calls: is it like a chatbot for finance?
No. And the difference matters more than it sounds.
Chatbots wait for prompts. Agents do work.
A chatbot is reactive. It needs you to type a question. The best ones can summarize a vendor's payment history or draft a reminder email. That's useful. It's also not what's clogging your team's calendar.
An agent is the opposite. It's a process running against your real systems — reading inboxes, matching invoices, posting entries. You don't open an app to use it. You notice it's done.
The work is in the doing, not the asking
The bottleneck in finance isn't "I have a question and need a fast answer." The bottleneck is "I have 800 invoices and not enough hours."
A chatbot helps the first problem. It doesn't touch the second. We built Maya for the second.
What this means in practice
There's no chat window in Maya. There's a queue, a log, and an exception list. The queue clears overnight. The log shows you exactly what happened. The exception list is short, judgment-heavy — the work that needed a human anyway.
If you want a chatbot for finance, there are good ones. If you want the work done, that's a different product.
When we started Maya, the same question came up in the first three sales calls: is it like a chatbot for finance?
No. And the difference matters more than it sounds.
Chatbots wait for prompts. Agents do work.
A chatbot is reactive. It needs you to type a question. The best ones can summarize a vendor's payment history or draft a reminder email. That's useful. It's also not what's clogging your team's calendar.
An agent is the opposite. It's a process running against your real systems — reading inboxes, matching invoices, posting entries. You don't open an app to use it. You notice it's done.
The work is in the doing, not the asking
The bottleneck in finance isn't "I have a question and need a fast answer." The bottleneck is "I have 800 invoices and not enough hours."
A chatbot helps the first problem. It doesn't touch the second. We built Maya for the second.
What this means in practice
There's no chat window in Maya. There's a queue, a log, and an exception list. The queue clears overnight. The log shows you exactly what happened. The exception list is short, judgment-heavy — the work that needed a human anyway.
If you want a chatbot for finance, there are good ones. If you want the work done, that's a different product.

