As Black Friday marks the onset of peak customer service pressure, a new report from Typewise lays bare an uncomfortable reality: despite soaring expectations, AI and automation are still failing to relieve frontline strain. After analysing more than 10 million interactions across the US, UK and Europe, Typewise forecasts another sharp holiday surge, while efficiency drops and chatbots are pulled offline. In this TechBullion interview, Typewise’s CEO discusses why thinking time collapses, errors rise, and AI remains under-adopted, and why the gap between promise and performance is now costing organisations dearly, echoing Gartner’s warning on unmet AI ambitions.
Please tell us more about yourself and the solutions you provide at Typewise?
My name is David Eberle, and I’m a Swiss technologist and former strategy consultant (Booz & Co.) turned entrepreneur. Today I’m the Co-founder & CEO of Typewise, a Y Combinator-backed startup that delivers enterprise-grade AI agents for customer service teams to automate and re-imagine customer experience: from fetching data buried in legacy systems to resolving tickets end-to-end, it masters the repetitive and the complex alike.
What patterns in customer behaviour during the holiday season surprised you most in this year’s analysis?
Among the most compelling findings in Typewise’s Holiday Support Surge Report is the revelation that holiday support doesn’t just intensify existing pressures surrounding holiday shopping and customer decisions – it actually transforms the customer service environment entirely, indicating a much larger impact than the simple term of ‘support’ might imply.
Customer service representatives typically handle 22% more sessions per week during the peak shopping period, up from 160 to 195 interactions per agent. And while volume is part of the story, the actual nature of work shifts dramatically as agents lose the time usually spent contemplating responses and instead transition into rapid-fire writing just to meet demands.
How do you see retailers balancing customer expectations for instant support with the operational reality of limited agent capacity?
While inefficiencies are amplified during the holiday season, the repetitive nature of customer service queries is a challenge that AI is well-suited to address. In fact, Typewise customers already using AI-powered autocomplete and writing assistance are seeing 20-60% reductions in typing and response time, with retail and e-commerce agents saving 35% on average, which is the equivalent to more than a full workday per month!
Your recent report highlights a sharp drop in “thinking time” for agents during peak periods. Tell us more about the report and what it implies about the future skill set required in customer service roles?
Our report shows that during the holiday period, agents spend 13% less time thinking and 8% more time typing, with retail agents seeing an even sharper swing – thinking time down 17% and typing up 14%.
That shift tells us something important about the future of customer service: The bottleneck is no longer empathy or expertise – it’s cognitive load. Agents aren’t struggling with what to say; they’re struggling with the volume of repetitive tasks that crowd out the higher-value skills customers actually benefit from.
As automation handles more of the predictable, repetitive writing, and our data shows 46% of human-written responses are predictable by AI, the agent role evolves into one centered on judgment, contextual escalation, and relationship-building.
The future skill set is less about typing speed and more about supervising intelligent systems and stepping in only when human interaction with the customer is needed.
Many companies deactivate chatbots during the holidays. What technical or strategic changes are needed before retailers can trust automation in their busiest moments?
Retailers throttle chat from 43% of interactions down to just 7% once peak season hits. Now it’s important to note they’re doing this with human-led or AI-assisted chat, not fully agentic chat.
And they do this – not because they dislike automation – they do it because current staffing and tools can’t handle the complexity and urgency that comes with the channel.
Three things need to evolve before full AI agent-led support automation is reliable under pressure:
1) Multi-agent orchestration instead of single-task bots:
Most chatbots today operate as linear FAQ responders. When volume spikes, they break. Retailers need agentic systems that can retrieve data, interpret context, decide next steps, adapt to edge cases, and hand off intelligently. This is the shift the industry is now making.
2) Real-time decision intelligence:
Peak season issues aren’t static, but instead, they involve logistics, inventory, delays, and emotionally charged customers. Retailers need systems that understand urgency and act accordingly.
3) Proven operational resilience:
Retailers need evidence, not promises. Our platform is built on tens of millions of interactions and has already shown that AI can reduce the time required for human staff responses by 20–60%, giving teams breathing room when it matters most.
When automation becomes genuinely reliable during peak chaos – not just during quiet hours – retailers will stop turning it off.
Typewise works with brands such as Unilever, DPD and Beurer. What differences do you see in how global enterprises adopt AI compared with smaller retailers?
Enterprises like Unilever, DPD, and Beurer tend to adopt AI with a mindset of transformation, not tools. They’re thinking in terms of end-to-end journey redesign, multi-country rollout, and deep integrations into CRMs, ERPs, and legacy systems. That also means more stakeholders: security, legal, works councils, and operations all need to be aligned, so a lot of our early work is around governance, data residency, and change management rather than just “turning on” a feature.
Smaller retailers, by contrast, want the same outcomes, faster responses, happier customers, but need to get there with leaner teams and less IT capacity. They look for quick wins, out-of-the-box value, and a pilot that can go live in days, not months. So we adapt the same core technology to both realities: for a global enterprise, we might orchestrate a whole “team” of AI agents across channels and regions, while for a mid-market retailer, we focus on a single unified assistant that sits on top of their helpdesk and immediately takes cognitive load off the team. The ambition is similar; the path and constraints are very different.
Given that AI can predict almost half of agent responses, what barriers still prevent teams from fully leveraging that capability?
Workflow friction is a big one. During peak weeks, agents are so overloaded that they stop using the very tools meant to help them. Snippet and automated form usage drops by 27% during the holidays, not because the tools don’t work, but because they require too much context switching. When volume spikes, agents revert to manual typing because it feels faster in the moment, even though it creates more errors and slows teams down.
Furthermore, legacy support tools aren’t designed for automation. Most support platforms weren’t built with native AI in mind. They bolt it on, which means agents still carry a massive cognitive load.
Finally, there is compliance anxiety. Even when AI can generate accurate responses, teams hesitate because they lack guardrails, auditability, and role-based controls.
How is Typewise approaching data privacy, model training, and regional compliance when working across the US, UK and Europe?
We start from a simple principle: your customer data belongs to you. Typewise acts as a data processor; we don’t sell data, and we don’t train shared, cross-customer foundation models on your conversations by default. Where customers want model fine-tuning, we do it on anonymised data, with clear contractual controls around retention and purpose limitation. Our platform is ISO 27001-certified and GDPR-compliant, with data encrypted in transit and at rest, and hosted in AWS or Azure regions selected per customer, including EU-only processing where required.
From a regulatory perspective, we design for the strictest regimes first. In Europe and the UK, that means privacy-by-design aligned with GDPR/UK GDPR and the emerging EU AI Act: data minimization, clearly defined legal bases, support for data subject rights, detailed audit trails, and AI assistants configured in the “limited-risk” category with human hand-off for high-impact decisions.
Practically, that means customers can decide where their data lives, which systems an AI agent may access, what it is allowed to do autonomously, and how long data is retained. Every automated action is logged and reviewable, and sensitive workflows (like large refunds or policy exceptions) can be kept human-in-the-loop by design, regardless of whether the end customer is in the US, the UK or the EU.
With customer service volumes forecast to rise again, where are retailers likely to see the strongest return on investment from AI over the next 12–24 months?
Over the next two years, the biggest ROI for retailers will come from AI that reduces the cognitive load on both agents and customers, not just the time it takes to resolve a ticket. We’re seeing a shift away from basic automation toward systems that actually structure agents’ work, especially during peak seasons.
One example is multi-agent orchestration. Instead of relying on a single chatbot to do everything, retailers are beginning to adopt AI “teams” in the background – one system pulling data, another interpreting context, another drafting, and another validating compliance or policy alignment. This division of labor yields much higher accuracy and dramatically reduces the back-and-forth that eats up time and frustrates customers.
Retailers will also see significant ROI from automating the first draft of repetitive messages. Our data shows that nearly half of what agents type could be generated automatically, and when that happens, efficiency gains of 20–40% per agent are realistic. Beyond that, intelligent triage and intent recognition will make a measurable dent in misrouted conversations, while proactive logistics communication will reduce the massive volume of “Where is my order?” inquiries that spike during the holidays.
The bottom line is that the strongest ROI will go to retailers who focus on eliminating friction across the customer journey. AI that clarifies, structures, and reduces effort, rather than simply speeding up typing, is where the next wave of value will be created.
Beyond customer service efficiency, how do you see AI reshaping the wider retail and e-commerce experience, from logistics to post-purchase engagement?
AI will reshape retail far beyond the service desk, because the real frontier is not a more innovative chatbot, it’s a unified, always-on concierge for the entire customer journey. Instead of separate experiences for “shopping”, “support”, and “order tracking”, you’ll interact with a single intelligent assistant that knows who you are, what you’re trying to achieve, and can move seamlessly between product discovery, problem-solving, and purchase in one continuous conversation. The line between sales and service will blur: a question about a delayed parcel might turn into a tailored upsell; a sizing concern might instantly surface the right product, policy and shipping option without ever changing channels. That’s what conversational commerce looks like when it finally matures.
Under the hood, this requires AI to stitch together logistics, fulfilment, and post-purchase engagement. In logistics, AI will coordinate with carriers to anticipate delays, trigger proactive notifications, and propose solutions, refunds, replacements, and alternative delivery options before the customer has to ask. In parallel, personalization will finally extend beyond marketing into operations: policies, messaging, and resolutions can be dynamically adapted based on context, history, and loyalty level.
A big unlock here is “unified retail memory.” Retailers sit on vast amounts of fragmented data, support transcripts, order history, inventory, product feedback, loyalty programmes, but very little of it is accessible in real time. AI can turn that into a living memory that every assistant can tap into, so each interaction makes the next one smarter.
So, beyond efficiency gains, AI is preparing to reorganize the retail experience into something that feels much more human: a single, consistent concierge that takes care of you before, during, and long after the transaction, regardless of which channel you start on.
