Last week, Mesh-AI’s GenAI Symposium brought together over 150 enterprise leaders to explore how Generative AI is transforming businesses. The consensus was clear: the era of experimentation is over. Organisations in highly regulated industries are now moving beyond pilots to deploy GenAI at scale, navigating regulation and risk to achieve tangible returns.
Opening the event, Mesh-AI set the tone, describing this as a defining moment in technology. The critical question for enterprises is no longer if they should adopt GenAI, but how far and how fast they can go. The challenge is to think boldly—beyond the tech and tools - and focus on how AI can amplify human potential to create entirely new value.
In his keynote, Stuart McMeechan from OpenAI emphasised the unprecedented speed of GenAI adoption, labelling it the fastest tech adoption in history. With exponential improvements in models and APIs, the race for competitive advantage is fully underway.
He noted that financial institutions are leading the charge, primarily using AI to boost employee productivity. A key trend is the rise of AI agents, which are becoming a central focus for many organisations. The advice for getting started is to begin with a simple chat interface and scale from there.
For best practices, Stuart highlighted the critical role of evaluations (evals). Organisations should start by measuring a model’s performance against gold-standard benchmarks for accuracy and tone, then progress to more complex model-graded and human-graded evals. Prioritising use cases is essential; the most successful customers combine quick, high-impact wins with a few strategic "big bets" that have the potential to redefine their market.
Crucially, he stressed that GenAI initiatives work best when they have buy-in from the very top of the organisation.
Lisa Harrison-Chan and Christopher Hughes from LSEG presented a compelling case study on using AI to transform risk intelligence. Their journey underscores a fundamental principle: start with a clear business problem, not just a technology in search of one.
LSEG’s AI Content Curation platform tackles the historically manual process of sourcing risk data. The goal was to accelerate this journey while improving data quality and maintaining the trust their clients rely on. To achieve this, Sergei Batischev from Mesh-AI explained how the team built a robust, repeatable pipeline for a ‘human-in-the-loop’ workflow, ensuring consistent and accurate data extraction from unstructured text.
A key learning from LSEG was managing a cutting-edge AI project, which is inherently different from a traditional project with a fixed scope. Lisa emphasised the importance of starting with a strong, agile team and a solid foundation that can stabilise and scale. This involves managing stakeholder expectations for an iterative process where requirements evolve.
Chris highlighted that accuracy is non-negotiable in risk data. The human-in-the-loop model orchestrates the workflow between the GenAI and LSEG researchers, ensuring quality and building trust in the system. He advised that you need a business process that anticipates where the LLM could err and inserts a human to address it.
Their advice for others is to start with small, manageable pieces. Prove you can ingest and process data first, and don't worry about the front end initially. Build iteratively and, most importantly, listen to your users - the people who live and breathe these problems have the essential insights needed for success.
Adam Marchant from Mace shared how they aligned a GenAI project directly with a core business goal: achieving 30% efficiency gains. The specific challenge was reducing the immense cost and effort of bid generation, with over 1,000 bids created manually each year.
Finding no suitable off-the-shelf tool, Mace collaborated with Mesh-AI to build a bespoke application called pAIge. The app gives technical experts a first draft for any bid, drawing exclusively from pre-approved content to ensure accuracy and lower the margin for error.
A critical success factor was ownership. Mace broke from tradition by making the sales team- not the IT department - the product owner. This ensured the tool was built for the end-user's needs. The journey from proof-of-concept to rollout took just 16 weeks, aided by detailed workshops that ensured the solution worked with Mace’s existing processes.
Securing senior leadership buy-in early was vital, as was gathering continuous feedback from end-users. The pAIge app is now used by over 500 people globally and is fully maintained internally. While hard cost-saving metrics are still being gathered, the project has already delivered significant value, with one user reporting it saved them from working every Sunday night. More importantly, it has built internal appetite for further innovation.
A panel featuring experts from AWS, National Grid, and Databricks tackled the crucial balance between innovation and responsibility. The key takeaway: succeeding with GenAI isn’t about choosing between speed and safety. Governance is not a barrier but the essential framework that allows for confident scaling.
The panel acknowledged that while many pilots are moving into production, governance frameworks have sometimes struggled to keep pace. However, organisations are proving they can overcome these policy gaps. Pete Heywood from Databricks stressed the importance of transparency and engaging the wider business, keeping subject matter experts involved to validate AI outputs and ensure relevance.
Gavin Goodland from National Grid addressed the buzz around AI agents, noting they cause real concern in large enterprises due to data flow issues. He pointed out that in many organisations, foundational data is in a mess, and rapid progress on this foundational piece is required before agents can be fully exploited.
Mark Keating from AWS advised against being overly prescriptive with AI. A shift in mindset is needed: instead of telling the AI how to do a task, focus on telling it what you want it to produce.
The Mesh-AI GenAI Symposium painted a clear picture of an industry at an inflection point. To avoid being left behind, organisations must act now. The path forward involves:
1) Aligning AI with Business Value: Think beyond the technology. Ask what new value you can create and ensure every project is tightly coupled to broader strategic goals.
2) Securing Buy-In at All Levels: Foster top-down support from senior leadership to secure funding and bottom-up engagement from end-users to ensure adoption.
3) Starting with the Problem, Not the Tech: Identify clear business challenges and leverage GenAI champions within your teams to drive projects forward.
4) Building Iteratively and Inclusively: Begin with small, core components and scale out. Continuously listen to user feedback, as their insights are critical for nuanced success.
5) Prioritising Foundational Data: Address data quality issues to remove the primary barrier to advanced AI applications like agents.
6) Embedding Governance from the Start: Build accountability frameworks for data security, compliance, and trust. This isn’t about slowing down, but about building the confidence to scale fast and safely.
The message from the symposium was unified: Generative AI is delivering real-world impact now. By following these strategic principles, enterprises can navigate the complexities and unlock transformative value.
