Listen & Follow Along — Article Summary
Article Summary
Revolut, a financial technology company, has significantly reduced its costs for running AI code. In less than a year, they managed to slash these costs by about 90 percent. They did this by using their own AI models instead of always having to rely on expensive services from other companies. When they started using AI tools for about 2,000 engineers, they saw a 60 percent improvement in coding time. Also, the time it took to complete projects, known as lead time, got 70 percent better. Revolut was initially skeptical about the high cost of AI and worried about the return on investment (ROI), so they started with strict budget limits.
Because of these budget limits, engineers only had a small number of AI processing units, called 'tokens', to use. This made Revolut think about self-hosting their own models. They found that using their own GLM 5.1 model cost much less, about a tenth of the price of commercial options. When tokens became cheap and almost unlimited, engineers felt free to use AI for many tasks. They used it not just for writing code but also for other important jobs. For example, they used AI for debugging (finding and fixing errors) and for dealing with help tickets.
Revolut also learned that the biggest improvement did not come from just using more powerful computers. Instead, they found better results by making technical adjustments to their AI models. They also realized that having a very large memory for the AI (a context window) was not always the most efficient. A smaller context window, around 256,000 tokens, worked best for their engineers. This allowed the AI to start responding in less than 3 seconds. Revolut now makes using open-source AI models a priority because it helps them control costs and avoid unexpected high bills, known as bill shock.
Key Vocabulary
slash
/slæʃ/
Click to reveal
rely on
/rɪˈlaɪ ɒn/
Click to reveal
lead time
/liːd taɪm/
Click to reveal
skeptical
/ˈskɛptɪkl/
Click to reveal
return on investment (ROI)
/rɪˈtɜːrn ɒn ɪnˈvɛstmənt/
Click to reveal
self-hosting
/sɛlf ˈhoʊstɪŋ/
Click to reveal
debugging
/diːˈbʌɡɪŋ/
Click to reveal
context window
/ˈkɒntɛkst ˈwɪndoʊ/
Click to reveal
priority
/praɪˈɒrɪti/
Click to reveal
bill shock
/ˈbɪl ˌʃɑk/
Click to reveal
Comprehension Questions
1. What was Revolut's main strategy to lower AI coding costs?
- They stopped using AI tools completely.
- They relied more on external commercial AI services.
- They started self-hosting their own AI models.
- They hired more engineers to write code manually.
2. What improvements did Revolut see after using AI coding tools?
- Only coding time improved by 10%.
- Coding time improved by 60% and lead time by 70%.
- Lead time decreased by 90% and costs increased.
- There were no noticeable improvements in efficiency.
3. Why did Revolut decide to investigate self-hosting AI models?
- They wanted to offer AI services to other companies.
- Their engineers had unlimited access to expensive tokens.
- They were unhappy with the performance of their human engineers.
- Engineers had limited access to tokens due to strict budget limits and high commercial costs.
4. What does Revolut's experience suggest about the most important aspect of AI for business tasks?
- Using the most powerful hardware is always the best solution.
- Having a very large AI memory (context window) is always necessary.
- The greatest benefits often come from using AI for support tasks like debugging and small fixes.
- The speed of generating tokens is more important than how quickly the AI starts responding.
5. Do you think prioritizing open-source AI models is a good long-term strategy for all companies, based on Revolut's experience?
- Yes, because it always guarantees better security.
- No, because commercial models are always cheaper.
- It depends, as it can save costs and avoid bill shock, but might have limitations for certain tasks.
- Yes, as it means companies will never need to make technical adjustments.
Discussion Prompts
1. How does your company currently manage technology costs? Have you explored new ways to save money?
2. Can you think of any daily tasks in your job that AI could help with, besides writing code?
3. What are some advantages or disadvantages of a company building its own technology instead of buying it from others?
Live Session Prep & Cheat Sheet
🎯 Speaking Targets (Vocabulary)
Try to use these target terms in your speaking turns:
- slash
- rely on
- lead time
- skeptical
- return on investment (ROI)
⚙️ Grammar Target Formula
Gerunds after Verbs and Prepositions: Verb + Gerund (-ing) OR Preposition + Gerund (-ing)
💬 Discussion Openers
Use these phrases to open or structure your arguments:
- In my opinion...
- What do you think about...?
- From my experience...
- I agree/disagree because...
Teacher Notes
This lesson focuses on Revolut's journey to reduce AI costs. Encourage students to connect the concepts of cost-saving and technology adoption to their own professional experiences. Pay attention to their use of gerunds and prompt them to use the target vocabulary.
Speaking Class Facilitation Guide (Tutors/Moderators Only)
🎭 Role-Play Scenario
Situation: Your company is looking for ways to reduce operational costs. You believe that investing in a new, self-hosted technology could save a lot of money in the long run, but your manager is skeptical due to high initial costs.
Employee (You): You are enthusiastic about the new technology. You need to explain its benefits, especially how it can slash costs and improve lead time over time. Prepare to address concerns about initial investment and ROI.
Manager: You are cautious about spending money. You are skeptical about new technologies and worry about the immediate costs and the return on investment. You need clear evidence and a strong plan before agreeing.
Goal: The employee must convince the manager to consider running a small trial of the new technology.
⚖️ Debate Prompt
Question: Should companies always prioritize self-hosting technology to save costs, even if it means more internal effort?
Side A:
- Self-hosting reduces long-term costs and avoids 'bill shock'.
- It gives companies more control over their data and security.
- Internal teams gain valuable knowledge and skills.
- It can lead to custom solutions tailored to specific company needs.
Side B:
- Commercial solutions are often easier to implement and maintain initially.
- Self-hosting requires significant internal resources, staff, and expertise.
- External providers handle updates, security, and technical issues.
- It can be a big initial investment with an uncertain ROI.
💡 Discussion Facilitation Tips
Encourage students to use phrases for expressing opinions and agreement/disagreement. Prompt students to use the target vocabulary, especially 'slash,' 'rely on,' 'ROI,' and 'bill shock.' Correct common errors related to gerunds, like using an infinitive instead of a gerund after a preposition.