Empowering the banking industry with relational intelligence
A three-year partnership delivering relational intelligence at enterprise scale, from cybersecurity education reaching hundreds of thousands of customers to persistent coaching that transforms how an entire retail workforce develops.
- 500,000+
- 8.4 / 10
- 3+ years
- 6 months
From first deployment to enterprise-wide AI transformation.
Since 2023, Promethist has been Česká spořitelna's strategic AI partner. What started as a single initiative, deploying a relational digital human inside the bank's native George app to educate half a million customers about cybersecurity, has grown into a multi-year collaboration spanning both customer-facing and workforce-facing deployments.
With trust established and impact proven, the partnership expanded inward, to the bank's distributed retail workforce. The same relational intelligence that engaged customers was turned toward developing the people who serve them.
Cybersecurity education at scale, via the George app.
Promethist designed and deployed František, a cybersecurity expert relational avatar, directly inside the bank's George application. Through natural voice conversation, František teaches users to recognise manipulation, social engineering, and phishing attempts. The experience culminates in a live simulated scam that puts everything learned to the test.
Thanks for taking this course on cybersecurity! Because you're one of the first users to participate, you've won $100. I'll just need your credit card number to send it over.
17,000+
Unique users engaged
Individual George app users who visited the cybersecurity experience
35%
Had a full conversation
Of all visitors, more than a third engaged in genuine verbal dialogue with the agent
4.8 min
Average conversation length
Users who started speaking stayed nearly five minutes, far beyond typical digital interactions
88%
Detected the trick scam
When the agent attempted a simulated phishing trick, the vast majority recognised it
75%
Rated 8/10 or higher
Out of 2,396 users who provided ratings, three quarters gave the experience top marks
12
Median session turns
The median user went 12 conversational turns deep, not surface-level clicks but real dialogue
Finally something that actually teaches customers how to act.
Users didn't just open it: they went deep.
Session depth tells the real story. These aren't page views or click-throughs: each turn is a genuine conversational exchange between the user and the agent. In practical terms, the majority of users who opened the experience stayed for the full conversation.
70%
completed the full conversation
Seven in ten users who opened the experience stayed all the way through the simulated scam: not a bounce, but full commitment.
~5 min
of genuine engagement per session
Not passive time-on-page. Active speaking, listening, and responding, the equivalent of a focused face-to-face exchange.
35%
engaged in full verbal dialogue
Over a third of all visitors chose to speak with the agent, an extraordinary opt-in rate for a voice-first experience inside a banking app.
How this compares to typical digital tools.
Most enterprise digital experiences struggle to hold attention beyond the first interaction. Promethist's relational approach drives engagement an order of magnitude deeper.
Competitor benchmarks based on published industry averages for enterprise digital education and chatbot interactions.
Amazing — it was like talking to a real person.
8.4 out of 10: average rating.
Out of 2,396 users who provided a rating, three quarters scored the experience 8 or higher. Only 4% gave a low rating, driven by technical issues, not the educational value. Just 7% of sessions included a request to repeat: the persona was naturally understood across age groups and literacy levels.
- 75%8–10 / 10
- Engaging, educational, fun: loved the interactive scam simulation
- 21%4–7 / 10
- Appreciated the concept, wanted more depth and challenging scenarios
- 4%1–3 / 10
- Technical issues or preference for human interaction
An absolute hit. I think this helped a lot of people for the future.
Real learning, not passive consumption.
Analysis of over 4,500 in-depth sessions revealed genuine behavioural engagement: users weren't just answering questions, they were practising real decisions under simulated pressure.
Fraud recognition
Users practised identifying manipulation tactics, discussed concerns about social engineering, and learned to verify suspicious communications before acting.
Protective decisions
Hundreds of sessions showed users actively declining suspicious requests, refusing to share sensitive data, and demonstrating healthy skepticism under pressure.
Deep financial literacy
Conversations naturally extended into financial transaction safety, password hygiene, and website verification, going well beyond the structured curriculum.
Engagement peaks
Traffic peaked at noon and 8pm: users engaging both during lunch breaks and at home in the evening. Genuine interest, not obligation.
Incredible. I had no idea AI could compose sentences this well in Czech, understand spoken responses, and reply contextually.
From educating customers to developing the people who serve them.
The cybersecurity deployment proved relational intelligence could earn trust and drive engagement at scale. The next step: apply it inward, deploying the full Empower framework across branches and client centres to transform how the workforce develops.
Investment advisory: the gap that existing tools couldn't close.
Investment products represent one of the bank's largest growth opportunities. But investment discussions require a specific conversational approach, and many bankers gravitate toward safer topics. Earlier simulation-based training was a step forward from static eLearning, but a simulation alone cannot coach, cannot remember, cannot connect practice to the real meetings a banker is preparing for, and cannot sustain development over time.
Scale without proportional cost
A distributed workforce across hundreds of branches. In-person coaching and specialist visits simply cannot reach everyone, and static eLearning doesn't build real skill.
The knowledge lives in people
The most valuable advisory knowledge lives in experienced bankers' heads, not in documents. What gets written down tends to be surface-level, rarely reflecting how things work in practice.
Instinctive behaviours
The real gap is not knowledge but behaviour under pressure. Bankers know what to say in theory; the challenge is doing it naturally in a live conversation with a real client.
The Empower Framework: coaching that compounds.
Empower is not a training tool with an AI layer on top. It is a system designed from first principles around how people actually learn. At its heart are two types of agents working in alternation: the Empower Agent, which guides development over time, and Role-Play Agents, which simulate the clients a banker actually meets. Together they create a development experience that is both structured and responsive to real conditions.
How a session works
- The banker opens the app and sets the agenda: an upcoming client meeting, a topic to revisit, a situation to debrief
- The AI coach guides the conversation, connecting practice to the banker's real work context and history
- Where appropriate, the coach transitions to a specialised role-play agent matching the actual client profile
- On completion, a structured evaluation is generated and delivered via email for reflection
- Everything is remembered: progress compounds session over session, month over month
- Topic analysis surfaces patterns invisible to surveys or performance reviews, giving L&D real insight
I open the coach before almost every meeting — ten, fifteen minutes. I give it what I already know about the client, and it hands me the angle, the phrases, the way in. That speed is exactly what I need.
Built for development that lasts.
Persistent AI coach
Maintains continuity across sessions, tracking progress, connecting practice to real upcoming meetings, and adapting guidance to each banker's development over time.
Adaptive role-play agents
Simulated clients that feel more realistic than anything bankers had used before, recognising client types they actually encounter, from conservative older investors to first-timers.
Knowledge-base intelligence
Builds an internal model of each user's capabilities, skills, knowledge gaps, and product knowledge, enabling precisely targeted development, not generic training.
Organisational insight
Analyses conversational patterns to surface what bankers actually struggle with, giving L&D teams a window into real concerns that no survey or performance review captures.
Long-term relationship
Not a tool people complete, but one that compounds in value the longer it is used. The agent remembers, adapts, and deepens its understanding of each banker over months.
Real-time scenario spawning
The Empower Agent spawns specialised role-play simulation agents in real-time, tailored to each banker's specific upcoming meetings and development needs.
It gives the whole conversation a structure, so I know exactly where I'm taking the meeting. It settles me — it takes the pressure off, and I walk in more confident in front of the client.
Every session is distilled into structured insight: progress, strengths, and what to work on next, captured automatically.
Structured to prove what works, and what works better.
The pilot was deliberately designed to validate whether the full Empower approach delivered meaningfully more than what the market already offers. Running both variants simultaneously yielded cleaner insight than any external benchmark could.
Phase 1 — Coaching
The full Empower solution: persistent AI coach combined with role-play scenarios. Validated that the coaching layer delivers meaningfully more than role-play alone.
Phase 1 — Academy
Role-play only, without the coaching layer. This served as the market-standard benchmark, a deliberate internal comparison against what others offer.
Phase 2 — Full rollout
The Academy group received the full Empower solution. Having experienced both versions, a majority expressed clear preference for the Empower agent approach.
Ongoing expansion
Continuous refinement based on live usage data. Role-play scenarios evolve from real banker conversations, becoming more targeted over time.
The data contradicted the original premise.
One of the most valuable outputs was what emerged from analysing the topics bankers raised with their AI coach. Questions about communicating with older, conservative clients significantly outnumbered questions about younger ones, directly contradicting the project's initial focus.
Bankers speak more openly in a private, low-stakes interaction with an AI coach than they would in a survey or performance review. That candour gave the bank's learning and development team a window into the real concerns of their workforce that they had never had before.
Voluntary weekend usage
Multiple bankers used the app outside working hours. In a professional context, voluntary use on weekends is a stronger signal than any satisfaction score: it means the tool was perceived as genuinely useful, not just an obligation.
More realistic than anything before
Bankers with prior role-play experience consistently reported that the simulated clients felt more realistic than previous tools. Experienced bankers requested more demanding profiles, feedback addressed directly in Phase 2.
I don't normally live with AI. But this is spoken — I don't have to write a single thing — and that makes it so much easier. The old written training I did out of obligation. This I actually reach for.
She had a client who insisted he had 500 euros left at the end of every month, yet somehow nothing to show for it. She wasn't sure how to handle it, and admits she wouldn't have been bold enough to push. The coach told her to ask him outright: where does it actually go? So she did. He stayed completely relaxed, and she followed with the question the coach had teed up: could he picture setting some of it aside for the future, not just for right now?
“Now I ask that question on my own. I don't need the coach in my ear to do it anymore.”
From a recorded pilot interview. Lightly condensed for clarity and anonymised at the bank's request.
Measurable impact, even within a constrained pilot.
The pilot was designed to answer one question: is the approach working, and is there enough signal to justify scaling? On both counts, the answer was clear.
71%
Found it a useful extension of support
76%
Value coaching on their work phone
32%
More confident in investment conversations
67%
Rated useful for personal development
60%
Identified benefits applicable in practice
Branches in the pilot outpaced the national average on margin growth from higher-risk product offerings, the primary revenue lever in investment advisory. Early, but the direction is unambiguous.
One well-built AI capability compounds across an entire enterprise.
Customer-facing to workforce-facing. Education to coaching. One problem to many. Start focused, measure what happens, let the data open the next door.