Editors Note: This is Part Three of a special feature is brought to you courtesy Industrial Alliance Securities and analyst Blair Abernethy. It will run in multiple parts over the next two weeks in Cantech Letter. For part one and to read disclaimers click here. For part two click here.
What are the Challenges to Creating AI-enhanced Applications?
At this point in time, the construction and deployment of a new AI application, whether internal custom-built to a specific organization or within an enterprise software company product portfolio, faces numerous technical and implementation hurdles, many of which have not been dealt with before. Similar to Data Warehouse/BI/analytics and more recent Big Data projects, assembling, preparing and ingesting massive disparate data sets to be used in an AI project absorbs a significant amount of the overall time and resources of a project. Successful AI projects must overcome numerous challenges, including:
• The need to access large data stores (and data ownership rights) is an important determinant of the ultimate effectiveness of the AI solution;
• Data scientists skilled at building AI models are difficult to find and expensive to hire, although they are not always required for straightforward, simple AI projects;
• Human subject matter experts are typically needed to properly train the AI model and validate results;
• Many AI applications can have negative implications for staffing levels as productivity enhancements can be significant. Adoption of AI by businesses may require re-training, process and corporate cultural shifts, which takes time and money;
• For software vendors, AI pricing models have yet to be standardized. For many SaaS companies, pricing approaches will have to change as “price per seat” charges may no longer apply;
• Machine learning and deep learning algorithms often require significant amounts of computing power, which is expensive and must be considered as part of a solution;
• The AI solution must be integrated and co-exist with legacy, often mission critical, enterprise software applications and workflows;
• AI Algorithms must be monitored, tracked, refreshed and archived as part of the corporate record;
• As AI projects are still a new area for most companies, determining an accurate project cost-benefit equation can prove to be difficult;
• Ethical and regulatory considerations (e.g., European General Data Protection Regulation (“GPDR”) must be taken into account, and project leaders are often in uncharted territory. We expect such considerations to grow in importance in coming years as the disruptive power of AI becomes clearer in organizations;
•Data anonymization approaches, such as removing Personally Identifiable Information (e.g., name, address, and social insurance numbers) and Protected Health Information, have been often proven to be ineffective. New approaches, such as Differential Privacy, need to be considered when building in certain applications.
Overall, we believe that the significant value proposition, whether it is improved accuracy, decision automation, efficiency or labour saving, that many AI automation projects can deliver will easily justify and far outweigh the costs and challenges of implementing the technology.
How Might AI Impact the Incumbent Enterprise Software Industry?
We believe that AI technologies are going to have a significant impact on many aspects of traditional enterprise software, SaaS applications, and consumer oriented applications (such as social media and apps) over the next five years. Outside of pure-play AI start-up ventures and BI/analytics companies, most software companies are still in the relatively early stages of assessing and evaluating areas of their businesses that could potentially exploit these technologies. In what follows, we highlight what we believe are some of the most likely near-term positive and negative impacts of AI on existing software vendors.
We see a number of potential positive impacts for enterprise software vendors that incorporate AI into their offerings, including:
1. Existing product line feature and function enhancement. We believe that AI can provide an entirely new vector (pardon the pun!) for innovation in both enterprise and consumer software applications and related services, particularly existing applications that generate significant amounts of operational data;
2. Creation of new, high-value recurring revenue streams from AI models, applications, and data scientist driven strategic consulting service revenue. We note that most enterprise customers have little in the way of in-house AI expertise at this point;
3. New and enhanced software value propositions. AI can provide meaningful speed-ups in business insight discovery, analysis, and automated decisioning. Enterprise customer labour productivity and workforce reduction opportunities can be significant with AI;
4. Increased application stickiness as AI algorithms are largely dependent on customer proprietary data and can evolve and improve with time;
5. AI can help create new market development and increased TAM opportunities for software vendors;
6. Leaders in the incorporation of AI into their software platforms can drive additional competitive differentiation;
7. Internally at software vendors, AI technology will eventually be useful for software program testing and quality control, which should help lower R&D costs.
On the other hand, the incorporation of AI technologies presents a number of challenges and potentially negative impacts upon the current business models of traditional on-premise enterprise software, SaaS and IT services vendors, including:
1. SaaS applications (and many on-premise software applications) have traditionally been billed out on a per user or per seat basis. However, the implementation of new AI applications very often replaces human workers, even skilled labour. Therefore, customer adoption of high value-added AI solutions could reduce the revenue of a SaaS vendor’s existing applications;
2. For the next few years, we expect AI R&D efforts will likely be relatively more expensive than traditional software application development as the demand for experienced data scientists and AI engineers will far outstrip supply, thus keeping costs high;
3. Software companies that don’t embrace AI product development could face the risk of faster-than-expected application obsolescence. We expect that AI-enabled enterprise applications will be competitively advantaged;
4. As AI applications can significantly reduce headcount at customer sites, SaaS vendors could have seat-counts reduced by non-competitive new third-party AI solution providers. For example, if an AI-enabled call centre required fewer live operators, then fewer seats of all Customer Service Representative (CSR) applications will be needed by the customer;
5. AI technology is likely to run into regulatory roadblocks that vendors will need to be aware of and address in coming years. For example, the European Union (“EU”) is in the process of developing regulations on algorithmic decision-making and a “right to explanation” for consumers. The EU’s General Data Protection Regulation, Article 22, is slated to take effect in 2018. Such regulations could increase AI product development costs and/or lower the anticipated value proposition.
Market hype over new AI-based solutions may turn to disappointment if end-markets fail to adopt the solutions and/or development and model maintenance costs prove to be excessive.
In conclusion, we think it is clear that AI’s rapid shift from the relative obscurity of university research labs into the daylight of the IT industry represents in important, long-term secular technology cycle. AI application development and subsequent adoption by consumers and businesses, given the global IT and communications infrastructure that is currently in place, could be quite rapid. The combination of several different types of AI technologies to create even more innovative and productive enterprise software solutions has barely begun.
AI presents both new product and revenue opportunities for the enterprise software industry, but also a number of threats and challenges. We intend to follow developments closely and expect significant advances in the technology to continue for many years. In the next section of this report, we look at the current and future potential state of AI development in several of our covered companies.
Stay tuned for Part Four of this report, in which Abernethy talks about how Canadian techs in his coverage universe are utilizing AI.