Scaling AI and its obstacles
DataIQ members have mentioned on numerous occasions that they experience difficulties when it comes to integrating their data into AI models. Some of these issues can be quality based, but others cite a lack of business interest in bringing the right data to the models.
When it comes to tools like AI models that are designed to be central to future operations, it is pivotal that the training data and subsequent data is clean, effective, and trustworthy, otherwise the risks of creating inaccurate results or misleading conclusions grows exponentially.
Perhaps more importantly, businesses need to trust that mission critical, often sensitive and confidential data can be used safely for genAI within their own governed AI platform.
The most common ways that inefficiencies arise tend to be from a lack of:
- Standardised tools
- Governance processes
- Automation for pipeline generation
- Version control
- Data access
- Guardrails
“The promise of genAI for business will be realised when companies trust that they can use internal data safely with the power of Large Language Models,” said Ian Spatz, Partner, Data Reply. “AIOps is the way to achieve this trusted genAI platform.”
One of the major selling points of AI technology is that it is supposed to expedite time to market and reduce costs, but these goals will never be fully achieved if the groundwork at the experimentation and scaling phases are not carefully and meticulously examined. The standing of the data office within the organisation is also at risk of being damaged if investments have been made on promises and expectations that are not materialising.
A recurring theme for DataIQ members is the apparent disconnect from decision makers between investment in new technologies, but then not receiving the same enthusiasm and support when it comes to investing in training to utilise the new technologies. For example, genAI requires a very specific set of specialised skills to as well as the upkeep and development of LLMs, and it is in the best interests of the organisation to ensure that data team members are provided the correct support and training to extract the most from the tools.
AIOps best practices
Businesses that bring AIOps into their day-to-day processes are far more likely to identify and address infrastructure and resource issues. They will be able to rectify any bottlenecks swiftly and can effectively implement solutions that can be futureproofed and suitable for scaling.
Dominic Rehn, Head of Digital Data Science, TUI, worked on their programmes with Data Reply, stating: “They worked side by side with us to build an MLOps approach that is right for us, and we will be expanding on that throughout the company.”
Data Reply explained to the Summit audience that different areas of the business can be positively impacted by sufficient AIOps. One example provided was about the development of a genAI newsletter generator and product description generator for a marketing specialist, which resulted in personalised B2B newsletters and market trend reports in under 30 seconds. The time savings and effort from old-style manual tasks had essentially been eliminated allowing the skills of the team to be directed to other projects.
Sohaib Ahmed, Founder of &facts, a marketing insights company that utilised Data Reply’s experience, said, “Data Reply’s expertise in model selection and cost optimisation was invaluable. Their collaborative approach accelerated the development of POCs, and now we are moving from POCs to production, enhancing our consumer insights with real-time market data.”
Data leaders have a difficult job when it comes to delivering production and value with new tools and tech, but a robust set of AIOps are the best way to achieve goals in the shortest timeframe. Organisations naturally want to grow and scale up, so it is key that data leaders are able to meet the demands of scaling in whatever form it takes, and this is possible through AIOps.
Organisations can contact Data Reply about their AIOps Maturity Assessment Framework, to identify gaps in AI capabilities.
