Perspectives
22 October 2024
2025 is expected to be a watershed year for Generative AI (GenAI) as the fog of hype clears and its true potential becomes better understood through analysis of its deployment across multiple use cases in a diverse range of industries.
Building off the adoption of AI tools over several decades, enterprises and investors are recognising that identifying targeted use cases where GenAI technology can optimise specific business functions will be key to maximising strategic impact and growth.
“It’s now a common assumption that organisations that delay the adoption of GenAI will lag behind those that have begun embedding it within their workflows,” says Christopher Hieb, Managing Director in Macquarie Capital’s Software & Services team. “When deployed optimally, GenAI technology can increase the speed of decision making and enhance the efficiency and productivity of individuals, freeing up teams to focus on more creative work – such as generating new ideas and innovation.”
Like many technologies, GenAI cannot sustain a linear upward trajectory on the ‘Gartner Hype Cycle’, which describes the typical phases that innovations go through – inflated expectation, disillusionment, enlightenment and an eventual plateau of productivity – and depicts their path to productivity as being more cyclical.
There is a belief, however, supported by Hieb, that although GenAI will follow the same progression, it will enjoy a faster transition through the typical hype cycle.
“GenAI is evolving and being deployed rapidly across several markets – this means it could pace through the hype cycle faster than other technologies. A big part of that transition is likely to happen in 2025,” says Hieb.
A key element of this accelerated transition, he says, will be the increasing ability to measure the impact of GenAI; until now, return on investment has been hard to quantify. In a 2024 Gartner survey, 49 per cent of respondents identified difficulty in estimating and demonstrating the value of AI projects as the main obstacle to AI adoption.1
“Since the release of OpenAI’s ChatGPT in 2022, markets have been able to collect and analyse performance data on practical GenAI applications, including measuring error reduction rates, compliance improvements, customer satisfaction, and cost reductions and revenue improvement.”
With up to two years’ worth of data now available, we are sharpening our understanding of GenAI’s impact on business performance to emerge. 2025 is likely to become the first year its true impact begins to be measured at enterprise levels.”
Christopher Hieb
Managing Director, Software & Services
Macquarie Capital
In that time, Hieb has seen a swathe of successful GenAI deployments, particularly across functions in the customer interactions space, due to the relative ease of deployment in these areas and straightforward application to improve existing business functions.
When coupled with automated speech recognition (ASR) technology, foundational models, or other digital communications like messaging, GenAI can be an extremely powerful tool used to understand a conversation in human terms. For example, it can be used to measure satisfaction, sentiment, trust and loyalty, among many other key traits and reactions.
“This massive evolution of technology has revolutionised how enterprises view contact with their customers – from the very first interaction to every communication touchpoint along the customer journey, including human-to-human, via SMS or through a virtual agent, or any other digital channel,” Hieb says.
To demonstrate this, Hieb points to the debt collection sector, where conversational AI is used to match operators to clients based on behaviours and personality traits to increase the likelihood of a positive interaction. This has the potential to make every customer touchpoint more meaningful, productive and lead to better outcomes for both parties.
GenAI has also been instrumental in helping enterprises collect more accurate and robust data on customer satisfaction, often in real time. “Historically, there were very few options when trying to assess the quality of a customer interaction,” Hieb says. “Human intensive call monitoring and recording and surveys were heavily relied upon – with surveys being the most prevalent and with many drawbacks. While bimodal survey data often overweighs on dissatisfaction and occasionally strong satisfaction, GenAI technology largely addresses the challenges associated with classic surveys by understanding conversations and facilitating human-like responses, resulting in dramatic improvements to customer satisfaction without the need for a survey.”
Hieb explains that while these use cases may not seem revolutionary, the significance becomes clearer when you consider that many industries are now saturated with competition.
Customer satisfaction is the holy grail of retention and the most critical KPI for many enterprises. If GenAI can be deployed to meaningfully improve those metrics, it could provide that marginal gain for a given enterprise over its nearest competitor.”
Christopher Hieb
Managing Director, Software & Services
Macquarie Capital
In Hieb’s opinion, GenAI investment does not have to be bold and expensive to transform certain business practices. Many enterprises are discovering that large language models (LLMs) are not always appropriate for certain aims and that small language models (SLMs) with fewer parameters can be more cost effective, more flexible and better suited to solving discrete problems. They are also less likely to run into regulatory hurdles in a climate where global regulations on AI implementation are set to rapidly evolve.
“Small language models are generally less expensive to train and run, and easier to modify, making them better for regulatory compliance and many other targeted applications than large language models. Bigger is not always better,” says Hieb.
Although they have been used for more than a decade, small language models are less discussed and perhaps less exciting, but Hieb believes they can be extremely effective and useful when considering the resources required for training and deployment. They also use far less computing and energy resources which is a challenge not yet addressed by the continued expansion of many large language model approaches.
Software technology trends over the past several decades could be a good indicator for how markets will evaluate and likely move toward small language models in the future. For example, the software as a service (SaaS) business model (using single instance, multi-tenant platforms) has all but obsoleted the license/maintenance model for several reasons – increased efficiency and cost savings associated with software deployments and updates being major drivers.
Similarly, vertical software applications have shown to be highly desirable in the investment community, in large part due to their targeted industry solutions, which are easier to implement (due to workflow and process alignment), and their ability to offer specialised solutions, minimal need for customisation, intuitive user experience, reduced learning curve and overall effectiveness. In general, CTOs will move toward fewer layers in the software stack and fewer applications with the simplest, most cost-effective tool to solve their organisation’s need.
“While large language models are used for very complex decision-making systems, this can be overkill for specific use cases,” Hieb says. “If history is an indicator, we believe smaller models will become increasingly popular with use cases across education, base code development, recommendation agents, chat bots, and conversational agents, among others.” He adds that applying, and subsequently updating, regulations is far easier when using a small language model due to the smaller datasets and can be very useful for ‘on device’ applications where data processing capabilities are limited.
“Companies like Google, Grammarly, HubSpot, Zoom and Zendesk all use small language models today for various applications and products like Gmail-suggestion assists, CRM automation tools, marketing automation tools, and chatbot automated responses, among others,” Hieb says. “We expect applications for the technology to continue to gain traction and believe it holds a lot of promise.”
Opportunities for investment in the AI space more broadly are immense – considering AI software spending will grow to almost $US300 billion by 2027,2 creating a massive incentive for new market entrants across many vertical industries.
Global AI-related M&A continues to dominate the markets, comprising approximately 60 per cent of total AI-related M&A since 2018 with most of these transactions focused on software opportunities. According to Hieb, this is reflective of the technology advancements and the capital being deployed in areas of GenAI, among others.
“AI has become a critical consideration in investment and acquisition decisions more broadly, with many businesses and investors analysing the potential headwinds or tailwinds created by AI. Investors are rightly concerned about how the technology revolution is going to impact their investments and whether the uptick in AI deployment might put a company out ahead of others or even fall behind and out of business in the next three to five years,” says Hieb.
“With so many businesses looking to develop AI products or simply to position themselves as AI enterprises, investment decisions are complex. Early signs are that the outlook will become much clearer in the year ahead,” he adds.
Learn more
