Talla is one of the most popular bots in the Slack App Directory, but it doesn’t want to be known as an intelligent assistant anymore, CEO Rob May told VentureBeat in a phone interview.

The company is entering a transitional stage after learning a series of lessons in beta testing this year. Talla is one of a few dozen startups — most less than a year old — that make enterprise bots for platforms like HipChat or Slack with the backing of major venture capital firms.

“It’s not that people want digital assistants; it’s that they want pieces of their information and workflows to show up in messaging platforms,” Talla said. “We’re moving away from an intelligent assistant company and more to a company in the space of what I would call ‘intelligent information delivery’.”

Intelligent information delivery is retrieval of information from datasets or CRM software like Salesforce, the delivery of information to a company (think routine tasks like employee onboarding), and event-based information delivery that alerts an employee to something that happened and advises them what to do about it.

The main difference between a company that makes an intelligent assistant and one that makes intelligent information, May said, is that bots for intelligent information use less conversation and more buttons.

“I think with an intelligent assistant, you’re trying to package up a series of workflows and give them to a user that is available primarily through chatting with a bot. That has a bunch of problems,” he said.

Talla has been in private beta with 10 companies this year and says the process has been instructive. One of the big lessons learned while trying to build bots for business customers is to scrap the intelligent assistant.

Another takeaway: Not everyone in the office uses Slack yet, so enterprise team services provided through bots must have a web presence and a presence on chat apps.

Howdy, another popular Slack bot maker, also came to the conclusion recently that it needs to provide web access for its users.

“You need multiple inputs, multiple ways to interact with it,” May said.

Talla has experimented with different use cases: recruitment, scheduling, a bot that teaches you how to be a Slack power user. With the release of its Slack bot in July, the Talla team found that people who didn’t use the bot often liked the natural language processing (NLP), and that power users actually wanted less NLP.

“The model to build and deploy a bot within a company is not necessarily to start with natural language, but to move to that as a method that lets the functionality of the bot be more pervasive within the company over time. [This] is also a good way to start, because building these NL pieces is very expensive with time, compared to getting other functionalities out the door,” he said.

The company also found that were too many instances in which a digital assistant failed to complete a task — like adding an event to a calendar with a graphic user interface.

Talla also learned that its product is best in the hands of companies with 30 or more employees.

“If you have eight people, much of the software that you would use at an 80-person company, you don’t need yet,” May said.

Talla has 14 employees and is based in Cambridge, Massachusetts. In late 2015, the company raised a $3.5 million seed round.