ContactBabel's Inner Circle Guide to Voice AI, supported by Geomant: why 36% of AI projects get rolled back, and how the right focus and expertise unlock ROI.
Geomant has sponsored ContactBabel's new Inner Circle Guide to Voice AI (UK edition). It is based on surveys of UK contact centres and 1,000 consumers, and it shows why many AI projects stall and what the successful ones do differently. The full 72-page report is at the end of this article.
What the research found
36%
of UK contact centres have stopped, paused or rolled back an AI solution
16%
of calls are fully handled by voice self-service today (median just 10%)
67%
expect to be running voicebots within two years, up from 14% today
£987
modelled saving per 1,000 calls from doubling containment from 16% to 32%
AI rarely fails because of the AI. It fails at the joins.
32%
abandoned AI because accuracy or reliability wasn't good enough
23%
found integration too difficult
21%
saw no measurable benefit
86%
of stopped projects ended at pilot or proof-of-concept stage
Only 11% of rollbacks were due to customers reacting badly. The common causes are accuracy, integration and unclear benefits, and all three can be fixed with the right design, data and systems expertise.
Customers will use AI, as long as it's for the right jobs.
62%
of consumers would prefer AI in some circumstances
43%
prefer AI for a quick answer to a simple question
69%
rate AI worse than a human at understanding their issue
78%
worry about being unable to reach a human
Successful programmes automate simple, high-volume contacts and keep a clear, context-rich route to a person. Only 30% of organisations can say customers never have to repeat themselves after escalation.
Containment is the biggest prize. Agent assist adds more on top.
£6.17
average cost of an inbound call that a resolved bot interaction avoids
19%
of enquiries handled end to end by voicebots, against a 10% median for IVR
71%
say agents' post-call workload needs to be reduced
63%
report a positive impact on handling time from real-time agent assist
ContactBabel suggests 25–40% containment is a realistic target. Reaching it means integrating with business systems: only 29% of bots can complete tasks there today.
Many organisations are deploying AI without the controls to prove it works.
16%
have no formal controls over what their bots can do
18%
do no quality assurance of AI-handled interactions at all
42–54%
say it's 'too early to say' whether AI met expectations
74%
rank data security & compliance as the top selection criterion
No respondent expects their AI budget to fall over the next two years. To turn that spending into results, organisations need clear success measures, assurance and governance from day one.
From the report: a Geomant case study
4–5 → 2 min
email handling time with AI-drafted answers
92%
of AI drafts approved by agents unchanged
35% → 54%
bot efficiency once accuracy was proven, before voice was added
The rollout was phased: email first to prove value, then automation, then a focused voicebot. It was built on Geomant's Graia platform around the customer's existing systems.
Hear your own voicebot before you commit
Geomant VoiceShift is a structured discovery engagement. It covers an IVR audit, an automation opportunity map, a migration roadmap and business case, and a live voicebot prototype built on your real call flows. It draws on 25 years of contact centre expertise, and we're vendor-agnostic and business-first.
Read the full report
The Inner Circle Guide to Voice AI (UK edition), 72 pages, by ContactBabel, supported by Geomant.
Open the full report (PDF, 72 pages)Book a VoiceShift Discovery Session
Talk to a Geomant specialist about where voice AI pays back first in your operation.
Email usSource: ContactBabel, The Inner Circle Guide to Voice AI (UK edition), September 2026. Savings figures are ContactBabel worked examples.
