Customer Experience News is a weekly newsletter about the most important news and discussions for Customer Experience and Customer Support Leaders.
This is all the weekly news you need in around 5 minutes.
Today’s headlines are from:
Maven AGI Turns Support Agents Into Sellers
Maven AGI, an AI Agent platform for customer experience, announced Maven Commerce, an extension that lets an existing support Agent sell a product, subscription, or service inside the same conversation. A Charter scopes exactly what the Agent can offer, at what price, and to whom, so the sales motion runs inside the guardrails that already govern support responses.
Support and sales have historically lived in separate systems, which means a customer calling about a broken charger gets routed away from the person who could sell them a spare rather than stay in one conversation. Because Maven Commerce extends the Agent that already resolved the support issue, the same conversation can close a sale without a handoff, using account history the Agent already has, such as whether a customer already asked about billing or a plan upgrade.
Rolling this out means writing a Charter that defines precisely what the Agent is allowed to sell and to whom, since the tool intentionally will not let the Agent freelance a discount or invent a bundle. Teams should expect early setup time scoping that Charter and testing it against real support conversations before extending it to sales moments, and should plan to review Agent Observability logs to confirm every recommendation traces back to a clear signal rather than a guess.
Gorgias Lets AI Agent Act On SMS
Gorgias, an ecommerce customer support platform, announced that AI Agent can now verify a shopper’s identity over SMS using a one-time code, then take action on their behalf instead of only answering questions. Once verified, AI Agent can cancel orders, edit addresses, and resolve other requests that previously required a handoff to a Support Specialist.
SMS had been a read-only channel for AI Agent, meaning any request that touched an order or subscription landed in a Support Specialist’s queue regardless of how simple the fix was. Hard authentication removes that ceiling, so shoppers texting in for something like an address change get a resolution on the spot instead of waiting for a human reply.
This is live for all merchants today, with no configuration required to turn on identity verification for SMS. Gorgias says social channels are the next step for the same capability, so teams that rely heavily on Instagram or Facebook messaging for support should expect a similar authentication flow to arrive there before long.
Zipchat Adds Purchase Attribution And Story Replies
Zipchat, an AI Agent platform built for ecommerce, announced two updates: purchase attribution across offsite channels, and the ability for its Agent to read and respond to Instagram story replies and mentions. Purchase attribution credits a sale to the conversation and channel that produced it, whether that started on Facebook, Instagram, WhatsApp, or email, and the dashboard now breaks revenue down by channel.
Teams running an Agent across several channels have had no reliable way to know which conversation actually drove a sale once a shopper left the chat to check out. Crediting the sale back to its originating conversation and channel gives support and marketing teams a shared number to point to when deciding where to invest, rather than treating every channel as equally effective by default. On Instagram, the Agent reading story replies and mentions closes a gap where a customer asking whether an item in a story is in stock previously got no answer at all.
Purchase attribution works automatically with no setup needed, so teams should check the new channel breakdown on the dashboard for anything that looks off before trusting it for budget decisions. Instagram story handling does not extend to Facebook page stories, since Meta’s API does not identify which story a Messenger reply refers to, so those replies will still be treated as regular messages.
Ada Adds a Snapshot API for Agents
Ada, a customer service AI Agent platform, announced a Snapshot API that exports an Agent’s full configuration, including persona, Playbooks, Knowledge, API tools, and Handoffs, as a single archive file. That archive can then be imported back into any Agent, including the one it came from, whole or in parts.
Teams that wanted a restore point before a major configuration change, or that needed to promote a sandbox build into production, previously had to rebuild that configuration by hand. A snapshot removes that manual step, making it possible to save a checkpoint before a risky change, copy an Agent into a new region, or refresh a sandbox from what is live in production.
A snapshot never carries credentials, API keys, secret values, or Ada’s built-in PII redaction switches, so those still need to be set up separately on the receiving Agent. Teams should read Ada’s documentation on what a snapshot includes before relying on it as a full backup, since some configuration intentionally stays behind.
Linear Splits Inbox Into Priority And Other
Linear, an issue tracking platform, announced a new Priority tab in its inbox that separates notifications needing attention from ones that can wait. The same release includes a project composer that lets Linear Agent help draft a project before it goes live, document editing on mobile, and richer file previews inside code diffs.
Teams that route customer-reported bugs or feature requests through Linear generate a steady stream of notifications, and until now the inbox treated a review blocking a release the same as a routine comment. Priority selects what needs attention by default, with the option to customize by source or filter, so a Support Specialist or engineering lead watching for customer-impacting issues is less likely to miss one buried in routine noise.
Linear selects what appears in Priority automatically, so teams should check the inbox display options and adjust the filter if the default selection does not match what actually matters to their workflow. Teams can also keep using the inbox exactly as before if the new split does not fit how they currently triage.
GitBook Extends Git Sync To Full Sites
GitBook, a documentation platform, announced that Git Sync now works at the level of an entire docs site rather than space by space. A site now maps to a single Git repository, with the sync process built around one gitbook-docs.yaml file, and existing space-level Git Sync setups can migrate to site-wide sync from the site dashboard.
Teams already syncing individual spaces to Git had to manually create a new space and set up Git Sync each time they added content, which slowed down bringing new documentation into a docs-as-code workflow. Mapping a whole site to one repository removes that manual step and pairs with new site-level permissions, which matters more as teams let Agents make documentation edits alongside people.
Teams migrating from space-level to site-wide Git Sync manage that switch from the site dashboard and retain control over the configuration and content mapping, but should plan a review of how existing spaces map into the new structure before starting. New sites can lean on GitBook’s coding agent to help build the gitbook-docs.yaml file rather than writing it by hand.
Fin (Intercom) Tracks Ticket Time Across Every Team
Fin (Intercom), a customer service platform, announced a new Ticket assignments dataset that reports on every team and teammate a ticket passes through, not only the team it happens to be assigned to when a report runs. A new Sankey chart maps the full path a ticket takes, and a prebuilt Tickets lifecycle report puts the whole picture on one page under Reports.
Until now, a ticket’s time only counted against whichever team held it at the moment of reporting, so once a ticket changed hands the earlier team’s time on it simply disappeared from the numbers. Being able to see how long a ticket sat with each team, how many times it was handed off, and which team pairs pass work back and forth gives support leaders a much clearer picture of where a ticket actually loses time.
Teams that route tickets across multiple teams before resolution should pull the new Tickets lifecycle report soon after this ships to see what the handoff pattern actually looks like, since it may surface routing loops that were invisible in team-level reporting. This is a reporting change only, so it will not affect how tickets are currently routed or assigned, only what leaders can see about that routing after the fact.
Aha! Connects Discovery Clips And Builder Apps
Aha!, a product management platform, announced that Aha! Discovery can now use AI to capture clips from customer interviews and share them directly in Microsoft Teams and Slack. The same release lets Aha! Builder applications connect to outside tools such as Salesforce, Jira, and Slack through Elle, the platform’s AI assistant.
Customer interview insight often stays locked inside a single recording that only the person who ran the interview actually watches end to end. Sharing a clipped moment directly in Teams or Slack makes a specific piece of customer feedback easy to pass to a Support Specialist or product manager who was not on the call, without asking them to sit through the full interview.
This sits closer to product management than day to day support work, so CX and Support leaders using Aha! mainly for roadmap alignment should weigh whether interview clip sharing changes anything about how customer feedback currently reaches their team. Builder’s new tool connections depend on those third party integrations, Salesforce, Jira, and Slack among them, already being connected and permissioned.
DeepL Adds Two Languages To Voice API
DeepL, a translation platform, announced that its Voice API now generates translated speech output, not just transcription and translation, for Croatian and Tagalog, in beta. Both languages already supported transcription and translation; this update adds the synthesized audio output layer, delivered through external service partners.
Support teams offering live voice translation to Croatian or Tagalog speaking customers previously got translated text but no translated audio for those two languages, requiring a workaround for the speech portion of the call. Full speech-to-speech support for these languages, though in beta, closes that gap for teams already using DeepL’s Voice API for other languages.
Both languages carry a beta label, which is worth confirming against DeepL’s supported languages table before promising this capability to customers or building a workflow that depends on it. Teams should also confirm through the API’s language query with the beta flag whether their existing integration surfaces these two languages by default or requires an explicit opt-in.
Pylon Ships 118 Fixes
Pylon, a support platform built for B2B teams, announced 118 updates shipped in the past week, and nearly all of them are fixes and small improvements rather than new features. The batch touches account intelligence, the support system, product intelligence, and the core platform, with items like corrected SLA sorting, faster internal thread creation, and clearer error messages for failed email authentication.
A changelog full of new features looks impressive, but a support team living inside Pylon every day feels the fixes more: a knowledge base image rendering at the wrong size, a trigger firing when it should not, an export that silently drops columns. Committing a full week to that kind of maintenance, instead of a headline feature, is the difference between a tool that looks good in a demo and one that holds up under daily use.
None of these 118 changes require setup, since they are bug fixes and small usability improvements rather than new capabilities to configure. Teams that rely on Pylon heavily should still skim the full changelog for anything touching a workflow they depend on, since a fix to trigger logic or SLA calculation can quietly change behavior they had built around the old bug.
Dixa’s Mim Can Now Answer Calls
Dixa, a customer service platform built for ecommerce and retail brands, announced that Mim, its AI agent, can now answer the phone. Mim already resolves refunds, order tracking, and order changes across chat, email, and WhatsApp; voice is the newest channel added to that same set of capabilities.
Phone has stayed the hardest channel to automate, since a caller cannot be shown a menu of quick replies and has to be understood in real time. Extending Mim to voice means a customer calling about a routine order issue can get it resolved without waiting in a queue for a Support Specialist, while still landing with a person for anything Mim cannot handle.
Dixa has not published detailed documentation on this release beyond the announcement itself, so teams considering it should confirm exactly which call types Mim can resolve versus route to a human before turning it on for a live phone line. Voice carries higher stakes for a bad automated interaction than chat, so a phased rollout starting with a subset of call types is worth considering.
Mintlify Moves To Outcome-Based AI Pricing
Mintlify, a documentation platform, announced that every AI action now has a fixed credit price, and teams only pay when it delivers a result. An assistant answer that actually answers costs 25 credits, a documentation update from an automation run costs 250 credits, and an assistant that cannot answer or an automation that finds nothing to update costs nothing.
The previous meter charged by tokens used, so the same automation could cost a different amount every run and an assistant question that failed to answer still cost credits. Tying price to outcome means a team never pays for a run that produced nothing, and Mintlify says most teams should see a meaningful drop in credits spent on automations that check for changes and find none.
This applies automatically with nothing to configure, and existing plans and prepaid credits carry over unchanged. Editor and Slack agents move to unlimited usage on Pro and Enterprise plans at no charge, which is worth knowing if a team had been rationing use of either because of the old metered cost.
