July 22, 2026
Glia Lets Banks Loosen AI Answers Where the Risk Is Low and Lock Them Where It Isn’t
Banking AI vendor Glia has released three response modes for Glia Banker, its voice and digital AI agent, letting financial institutions decide topic by topic how much freedom the AI has when it answers.
A question about branch opening hours can now be handled in natural, generated language, while a question about a wire transfer can stay word-for-word as compliance approved it. The modes are available only inside Glia Banker, which the company says is used by more than 700 banks and credit unions.
Why Banks Have Been Choosing Caution Over Conversation
Most AI deployments in regulated service environments have forced one decision across every conversation, and that decision has usually favoured caution. Scripted answers keep auditors comfortable but make routine interactions sound mechanical, which is the opposite of what community institutions compete on.
Glia’s argument is that the choice should be made per subject rather than per deployment.
“The risk profile of a wire transfer is fundamentally different from a branch location query,” said Justin DiPietro, Co-Founder and Chief Strategy Officer at Glia. He described institutions that want fully generative language when a customer asks about travel, down to a remark about the destination, while insisting on fully approved wording on loan inquiries.
How the Three Settings Work
Strict Mode is the behaviour Glia Banker customers already run, delivering answers exactly as they appear in an institution’s approved knowledge bank. Glia says responses in this mode carry its contractual guarantee against AI hallucinations and prompt injection attacks.
Rephrase Mode keeps the same approved answers but adapts the wording to the language and context the customer used. The conversation sounds less rigid, and according to Glia the AI cannot introduce anything absent from the original approved source.
Compose Mode allows the agent to generate answers in real time from content the institution nominates, which can include policy documents, product information and approved knowledge articles. Glia states that the mode draws only on that designated material and does not reach into the open internet or unrestricted large language models.
Institutions can combine the three across inbound service channels and outbound campaigns. Glia says the configuration is intended to handle up to 80 percent of inquiries, freeing frontline staff for the conversations where a human adds most value.
Dan Michaeli, Co-Founder and CEO of Glia, said institutions scaling AI across inbound, outbound and procedural workflows want granular controls over where it is applied. “Response modes give them that control,” he said.
The Reporting Problem
Running different settings on different subjects produces a measurement requirement that many institutions are not yet equipped for. Previous Glia research found that 74% of banking contact centres already use AI while only 27% measure its business impact, which leaves most teams unable to tell whether a looser setting on one topic improved resolution or quietly damaged trust.
Customer tolerance pushes in the opposite direction. Genesys research covered in our recent reporting found that 84% of consumers abandon AI support after roughly three failed attempts, so leaving every topic on the most restrictive setting carries its own cost in abandoned conversations and calls escalated to already stretched teams.
Glia has approached the naturalness question before, having built an assistant that learns from the conversations of top-performing human agents rather than from documentation alone.
What CX Leaders Should Take From This
Teams in regulated industries have spent two years arguing about whether generative AI belongs in customer service at all, and this release suggests the more useful question is which conversations it belongs in.
Before configuring anything, service leaders would do well to agree internally which topics tolerate variation in wording, which carry regulatory exposure, and how they will judge afterwards whether the looser settings earned their place.
