Glossary
›
Scripted bot

Scripted bot

Glossary
Also known as:
Rule-based bot, FAQ bot, answer bot

What is a scripted bot?

A scripted bot is a software program that holds a text conversation with a person, usually in a website or app chat window, to answer questions or route requests using pre-written flows, a fixed set of trained intents or answers retrieved from a knowledge base. Unlike an AI agent, it is not designed to carry a task through to a completed outcome.

The line between the two is blurring as bots gain language models and some ability to act. The working test is simple: when the customer wants something done, does the conversation end with it done, or with a link, a form or a transfer? A scripted bot answers; an AI agent finishes.

How a scripted bot works

Scripted bots have gone through three broad generations, and all three are still in use.

TypeHow it decides what to sayWhere it struggles
Rule-basedButtons and decision trees written in advanceAnything the customer types that the tree did not foresee
Intent-basedSorts each message into a trained intent and returns the matching responsePhrasings and requests it was never trained on
Language-model basedGenerates a reply, often from a knowledge base searched at answer timeAccuracy on specifics, and acting rather than answering

All three are built around the reply. Even a fluent language-model bot typically stops where the work begins: it explains how to apply, links to the form or passes the conversation to a person. An AI agent adds a goal, access to the systems where the task is completed, and a record of progress that carries across steps.

Why the difference matters

Automated chat is already mainstream in some industries. The Consumer Financial Protection Bureau reported that over 98 million users, about 37% of the US population, engaged with a bank's automated chat in 2022, and that all of the 10 largest commercial banks used it. The same report described complaints from customers stuck in "doom loops," unable to resolve an issue with the bot or to reach a person.

Use is not the same as success. In a Gartner survey of 497 customers, only 8% had used a bot in their most recent service interaction, and just 25% of those said they would use that bot again.

The gap comes from scope. Customers often arrive wanting something done: an application finished, a payment arranged, a booking changed. A tool designed to answer questions meets that customer with information, and the task waits for another channel.

Where scripted bots are used

For a defined set of simple questions, a scripted bot is often the right tool: inexpensive to build, quick to update and easy to govern.

  • Common questions. Fees, opening hours, eligibility basics and policy explanations.
  • Status checks. Where an order, claim or application stands.
  • Triage. Working out what a customer needs and sending them to the right team or page.
  • Information gathering. Collecting details before a person takes over, so the customer does not repeat them.
  • Internal help desks. Answering staff questions about policies and systems.

The fit ends where a task needs several steps, judgment or changes in other systems, such as completing an application, negotiating a payment plan or rebooking a group reservation. Those are the cases where organizations extend or replace a scripted bot with an AI agent.

What a good scripted bot looks like

  • Stated scope. It tells customers what it can and cannot help with.
  • An easy exit. A person is reachable without repeating the same request several times.
  • Recognition of complaints and disputes. These are identified and routed, not answered with help articles.
  • Grounded answers. Responses come from approved, current content.
  • Context passed on. A handover carries the conversation so far.
  • No pretending. It identifies itself as automated.

How to measure a scripted bot

Containment, the share of chats that never reach a person, is the usual headline number. It also rises when customers give up, so it is best read alongside these:

  • Confirmed resolution. Conversations where the customer confirms the question was answered, rather than simply leaving.
  • Repeat contact. Customers who come back about the same issue within a week, on any channel.
  • Clean handovers. Escalations where the person receiving them did not need to ask again.
  • Complaints about the bot. Tracked over time and by topic.
  • Completed tasks, where it is meant to act. A bot expected to complete tasks should be measured like an agent: on outcomes, not contained conversations.

Rules for automated chat

A scripted bot speaks for the business, and the rules that govern customer communications apply to what it says. The US points below apply broadly; banking, insurance and other regulated sectors layer more on top.

  • Recognizing disputes. The CFPB's review of automated chat in banking noted that federal consumer financial laws give customers rights when they dispute an error or raise a concern, and that a bot which fails to recognize one can leave those obligations unmet.
  • Accuracy of answers. Wrong answers about fees, terms or rights are the institution's own statements. The CFPB also warned that bots built on language models can produce fabricated information.
  • Saying it is a bot. Maine makes it an unfair trade practice to use an AI bot in trade in a way that could mislead a reasonable consumer into believing they are dealing with a human, unless the consumer is clearly told otherwise. California and Utah have their own disclosure rules.
  • Personal information. Details typed into a chat are personal data, protected under the Gramm-Leach-Bliley Act for financial institutions and under state privacy laws more broadly.

Frequently asked questions

What is the difference between a scripted bot and an AI agent?

A scripted bot is built to answer and route; an AI agent is built to complete a task. The agent has a goal, access to the systems where the work is done and a record of progress, so the conversation can end with an application submitted or a payment arranged rather than with a link.

Are scripted bots still worth using?

Yes, for high-volume simple questions with stable answers. They are inexpensive and easy to control. The mistake is asking them to do work they were not designed for.

Can a scripted bot be upgraded into an AI agent?

Partly. Adding a language model improves understanding, but an agent also needs tool access, a defined process, state across steps and rules for escalation. Many organizations keep the scripted bot for questions and add an agent for tasks.

Why do customers get stuck with scripted bots?

Usually because the request falls outside what the bot can handle and there is no clear route to a person. Stuck conversations are a design choice that escalation rules can fix.

Does a scripted bot have to say it is a bot?

In several US states, yes, at least in commercial interactions where a customer could be misled. Disclosing it everywhere is the simplest policy.

Past the answer, to the outcome.

Encore AI's agents do not stop at answers.

Scale your best to be the best.

Book a Demo