Delivering on the promise of AI in Paytech

Barry Levett

CEO

AI can and should be a meaningful contributor for many paytechs, but implementation requires careful consideration. This article explores where AI is already creating value and identifies areas where progress will develop more gradually.

AI is going to make the difference but not everywhere for everyone

When discussing major technological advances, we typically reference the transistor, integrated circuits, personal computers, networking, cloud computing and open source programming. Some transformed IT fundamentally; others dramatically reduced software development costs.

Whilst AI itself is not new, ChatGPT 3.5's November 2022 launch brought large language models (LLMs) into public consciousness and created widespread expectations of immediate, universal change. Many now question whether LLMs can deliver on this promise, overlooking the AI developments already embedded in everyday tools.

AI will reshape the world, yet success depends on the specific problem being addressed and the AI model's quality. Several paytech-adjacent areas demonstrate tangible potential for AI-driven transformation.

Hype cycle doesn't apply to AI

Major IT breakthroughs typically follow a predictable pattern: initial over-hype, followed by disappointment as promises fail to materialise, creating the famous "trough of disillusionment."

AI differs fundamentally. Much foundational work was completed during machine learning's (ML) development. AI didn't emerge suddenly in late 2022—it matured. ML systems' effectiveness depends on problem relevance, training data quality and available computing resources. As resources expanded and internet connectivity improved, ML systems advanced rapidly.

Today, pre-trained AI models are freely available through providers like Ollama.com or affordable subscriptions from Microsoft and GitHub.

The action is not in the LLMs but in other AI models solving real world problems. 6 key drivers are at work:

1. Most real challenges reduce to classification, prediction, optimisation or decision-making problems—areas where ML excels using proven models.
2. Recent explosion of high-quality datasets for training models.
3. ML tools like TensorFlow are free, user-friendly and integrated into development environments.
4. Abundant, inexpensive computing resources for model training and large-scale deployment.
5. Rapid return on investment through straightforward implementation using free pre-trained models and AI company services.
6. The ratchet effect: AI only improves; it never deteriorates.

Whilst AI research began in the 1950s, prerequisites for widespread adoption only recently became universally available—this time proves genuinely different. No single approach suits all situations; outcomes depend primarily on use case specifics.

AI helping developers to build better code faster

AI can address IT challenges rapidly and affordably. At Mypinpad, we observe rapid growth in AI applications within software development. Our developers use tools including GitHub Copilot and Cursor to construct, review and quality-test code substantially faster than previously possible.

These tools accelerate code writing, identify probable software defects, flag problematic sections and suggest alternative code lines for engineer approval and automatic implementation. Engineers can even train this technology to apply agreed modifications across extensive code sections.

This supports the Shift-Left movement, integrating quality assurance (QA), security measures and testing earlier in development cycles. Previously, developers wrote code before separate QA testing, then incorporated feedback through iterations. Now much occurs at initial drafting—frequently completed by the same engineer. The outcome: higher-quality software produced faster. This DevOps productivity advancement already manifests in Mypinpad's accelerated development cycle speeds.

Which applications demonstrate particular promise for AI-driven improvements in solution quality and deployment velocity? One area stands out for rapid advancement through AI-driven algorithms, alongside two representing slower adoption within our sector. Market-specific challenges will determine variation.

Use Case #1: Fraud Protection (Rapid Improvements due to AI)

Identifying fraudulent transactions requires analysing associated data: location, transaction value, timing, available account balance and purchase category. AI models assess and risk-score these factors instantaneously.

One executive described how purchasing two petrol tanks followed by trainers might trigger card blocking in the US. Whether this represents effective fraud identification aside, manually identifying and implementing patterns remains error-prone and inefficient.

AI classification models track significantly more variables than human analysis permits. They identify normally invisible patterns applicable immediately without disrupting existing payment processes. Implementation is straightforward.

Use Case #2: User Behavioural Analysis in Cyber Security (Rapid improvements due to AI)

Cyber security increasingly deploys AI to identify threats using multiple parameters linked to individual employee behaviour, determining whether online activity exceeds predefined risk thresholds. A platform might flag staff using potentially risky cloud file-sharing applications like WeTransfer externally.

Should historical records indicate a marketing team member legitimately uses WeTransfer sending branded brochures to external designers—containing no personal company data—that individual's usage receives approval whilst another employee's might not. This approach combines user behaviour analysis with other risk parameters to establish whether prescribed thresholds are exceeded.

Use Case #3 – Customer Service (Slower Burn)

AI captured substantial technology sector attention following ChatGPT 4.0's introduction, with firms contemplating, prototyping or launching chatbots for customer support—ourselves included.

However, experience reveals these generative AI chatbots perform only as effectively as supporting documentation. Chatbots function as enhanced automatic "Frequently Asked Questions" systems. Asking about pension policies or recent transactions should yield correct answers.

Success isn't about criticising AI systems but emphasising documentation quality and volume. It's fundamental GIGO (garbage in, garbage out)—superior content produces superior bot performance. Once adequate documentation exists, AI-powered customer service excellence follows.

Use Case #4 – New Products & Services (Slow Burn)

Innovations like Google Maps use AI and sophisticated algorithms, incorporating live traffic information and accident reports. Similar opportunities exist applying AI models to financial transactions—issuing unsecured loans or renewing motor insurance based on multiple-parameter heuristics enabling rapid decisions. These will emerge progressively, often with users unaware of underlying AI involvement.

Summary

AI is currently enjoying widespread adoption within technology communities for enhancing code quality—making generation cheaper and testing/finalisation faster. Resulting efficiencies prove substantial. Applying many AI models costs approximately nothing. TensorFlow, Azure AI and Ollama are all free. The focus now involves developing these models for particular use cases within specific markets.

Remember: the ratchet mechanism applies—AI models improve continuously. In certain areas, delivery of rapid improvements is inevitable. Retained developers will write more reliable code and construct solutions faster provided they possess requisite AI competencies, algorithmic knowledge and familiarity with relevant tools: GitHub Copilot, Cursor and Ollama.