Home > Columns > CRM Columns
CrmXchange Weekly Pulse - Sept 29– Oct 5, 2026
This week's main thread is customer backlash against AI
service: an Alchemer survey found nearly 9 in 10 consumers say AI has made
their service experience worse. Other themes: legal exposure from voice AI
(BIPA lawsuits), security risks from AI agents, and the shift from
"deflection" to resolution. In news, Microsoft released new real-time
voice models (Oct 1). Customer Service Week runs Oct 5–9, and CrmXchange's next
roundtable is "Beyond Metrics" on Oct 21. Open the session for the
full briefing https://bit.ly/3W8n5pO
Briefing:
1. Top themes this week
People are pushing back on badly built AI service
A new Alchemer survey of 2,000 US consumers found nearly 9 in 10 say AI has
made their experience with companies worse. The two biggest complaints: 44% say
it's harder to reach a person, and nearly a third have to repeat themselves
after being transferred. One expert in the piece says bots are often scored on
keeping customers away from live agents, so that's what the bots do. This is
the strongest data yet that easy escalation is now a must-have.
CX Dive
Voice AI is a growing legal risk (BIPA)
Plaintiff lawyers have moved from Big Tech voice-training cases to suing
Walmart and Lowe's over their customer service phone systems. The CX Today
piece puts potential damages at up to $5,000 per call. The president of the
Contact Center AI Association warns that even vague public descriptions of what
a system does with callers' voices can keep a lawsuit alive. The advice: audit
whether your voice systems analyze, store, or reuse voices as biometric data,
and make your public descriptions match.
CX Today
AI agents are opening new security holes
CX Today's security roundup includes AI agents reaching systems outside their
intended scope at about 100 organizations, an AI agent exploiting a zero-day in
Zammad support software, and 40M McDonald's Indonesia records exposed through a
misconfigured customer data platform. The message for CX teams: what an agent
can access, and what it's allowed to do, is now a governance question for CX,
not just IT.
CX Today
From "deflection" to resolution, with humans
reserved for judgment calls
Contact Center Pipeline's October issue includes "Before AI Becomes the
Next IVR" and "When the Audience Is Not Human," which argues
that deflection metrics need rethinking. Real-world examples back this up:
- TaskUs describes a "diamond" workforce model in
which AI handles routine contacts and humans handle exceptions like refund
decisions, with AI cutting volume 7–10% a year.
- Pumpkin, a pet insurer, limits its voice agent
"Poppy" to topics that don't require an insurance license, then hands
customers to licensed agents. It reports 5x more sales conversions.
Resilience after the Salesforce outage
The fallout from Salesforce's 7.5-hour outage on Sept 16 is still being
discussed. During the outage, some customers couldn't even file support cases
about it. CX Today's takeaway is that uptime figures don't protect the customer
journey. Its checklist: map every vendor your journeys depend on, keep backup
ways to log cases, and test real tasks such as one-time-passcode delivery.
2. Notable news & launches
- Microsoft (Oct 1): Three new voice models aimed at customer
service: real-time transcription in 60 languages, plus expressive and faster
text-to-speech. (PYMNTS)
- Salesforce Dreamforce recap (Sept 28): Fin, from
Salesforce's acquisition of the company formerly called Intercom, is now a
Salesforce service agent. Also announced: Voice Observability, Multi-Agent
Orchestration and Agent Health Monitoring.
- Customer Service Week is Oct 5–9, and Global CX Day is
Tuesday, Oct 6. (Hospitality
Net)
- CrmXchange: The next roundtable is "Beyond Metrics – From Numbers to Meaningful Outcomes" on Oct 21.
- Zendesk (mid-September, still being discussed): Specialized
AI Agents, ready-made for an industry or built with a no-code tool. Zendesk
says they can automate up to 80% of workflows. (Zendesk)
3. Worth a read
4. One takeaway
Most companies now use AI in service, so the question has
shifted to whether it actually resolves issues and hands off cleanly to people,
without creating new legal, security or reliability risks.