DataEQ's SA Insurance Sentiment Index: Governance Edition found that the majority of insurers were not making use of AI within their customer service journey, despite 92% of public insurance conversation on X being suitable for AI automation.
The AI-addressable share of conversation ranged from 12% to 95% across the index, an 83-percentage-point spread between the brands with the biggest AI opportunity and those with the least.
1Life Insurance recorded the highest AI-addressable conversation share in the index, at 95%. Its public conversation was dominated by competition entries, broker congratulations and informational content, exactly the structured, predictable interactions a well-configured AI agent could resolve at scale. Yet the brand responded to just 5% of high-priority interactions during the measurement period, the worst in the index. The brand with the most to gain from AI also had the least response to priority conversation of any kind, human or automated.
Dialdirect showed a similarly high AI-addressable share, at 95%, built on campaign and hashtag content, informational mentions and low complaint volume. Unlike 1Life, its service outcomes back that up: Dialdirect was one of the fastest brands in the index to respond to customers, averaging under 20 minutes.
Naked Insurance, as a digital-first insurer, is one of a handful of insurers the index identified as deploying customer-facing AI. The brand also benefited from talk around the launch of their ChatGPT-based insurance app in May 2026 that offers a 90-second, zero-human process for binding car insurance quotes. In the run-up to that launch, conversation about it generated 4% of all industry purchase-intent conversation .Even though the app itself wasn't available during the analysed period, it represents the clearest public example in the index of customer-facing AI deployment.
OUTsurance sat at the opposite end of the spectrum, with 88% of its conversation classified as priority, the highest ratio in the index. Claim denials, premium deductions and cancellation difficulties were prevalent in its social presence. For OUTsurance, and brands with a similar profile, AI can support an interaction, but when it comes to high-value or high-risk conversations, human involvement is generally still needed.
Pineapple Insurance shows both sides of that same choice. Its AI-first model, a photo-to-quote process using computer vision, and an AI-powered claims journey that removes the call centre entirely, has been core to a genuinely distinctive digital proposition and helped it raise R400 million in funding. But that same design is also its most visible liability: many customers described limited routes to human resolution, where the AI-mediated channel often produced looping outcomes instead of solving the issue. Without a human fallback, product failures risk can become unresolvable conduct issues under the FSCA's Treating Customers Fairly complaints standards.
Across the high-risk categories in the index, claims disputes and delays made up 62% of the high-risk conversation, followed by vulnerability signals at 24%, cancellation threats at 9% and regulatory escalation at 5%. These categories often make up a small share of total conversation, but they carry a disproportionate amount of business and regulatory risk. They can easily get drowned out by everyday social chatter and campaign conversation if a brand isn't actively separating the two, a common problem for social servicing teams.
On the opposite end of the spectrum, there was a huge opportunity for AI automation in large-volume, low-risk conversation types: brand awareness hashtags (52%), incentive and prize campaigns (27%), FAQ-style queries (11%) and product and support queries (10%) which are all primed for service efficiency improvements.
But it is not a case of one size fits all. Cancellation is a high-risk category, but not entirely off-limits to AI: it can explain policy terms, answer process questions and gather information upfront, while retention itself, the conversation that actually decides whether the customer stays or goes, generally still needs a person, or at least some level of human oversight.
Not all AI is created equal, and the industry's own conversation data makes that case bluntly. Net Sentiment on bot conversation sat at -70.2% across the industry, one of the most negative categories in the index, with 85% of bot mentions running negative. The complaints weren't about automation itself: customers described getting trapped in rigid systems that couldn't understand context, generic replies that didn't move a complaint forward, and reference numbers that led nowhere. Social media consistently became the last resort after automated channels had already failed to resolve the issue.
This is the distinction agentic AI is built to close. Rather than forcing every interaction through a fixed decision tree, a well-configured agent can understand context, respond more naturally, and recognise when a conversation needs to be handed to a person instead of looping the customer back to the start. The opportunity isn't to replace human agents, but to redirect them: every insurer in the index carries a different ratio of AI-addressable to priority conversation, which means a single, blanket strategy will not work effectively for most of them. Naked showed what acting early can look like. Pineapple showed the harder side of that same choice: also an early adopter, but now carrying the operational and conduct complexity that comes with high-risk conversation when the human fallback isn't built in from the start.
“Agentic AI isn't a single on-off decision, it's a spectrum. A well-configured agent can resolve a straightforward policy query with no human involved at all. But the moment a conversation touches a complaint, a vulnerable customer, or anything with regulatory weight, that same technology needs a person reviewing every response. Where insurers keep going wrong is applying one setting to both kinds of conversation.”— Sarah Lamb, Managing Director, DataEQ