Navigating the Generative AI Landscape

Dataiku and Databricks surveyed more than 400 senior AI professionals in large companies around the world in June 2023. What we found is that 64% of organizations will “Likely” or “Very Likely” use Generative AI technology over the next year. Plus, 45% of respondents said they are already experimenting with it.

This urgency means now is the time to start building a Generative AI strategy. One that will both turn applications of Generative AI use cases into reality (quickly) as well as safeguard against risk. Plus, as with any investment, your Generative AI strategy should be future proof for further developments that are sure to come.

However, there is a complex landscape of new and existing technology providers as well as consulting and systems integration (SI) offerings. This means choosing the right strategy is not easy. As with traditional artificial intelligence (AI) and machine learning, executives will need to weigh the tradeoffs between:

Short-term time-to-value
Long-term return on investment (ROI), including technical debt
Up-front versus ongoing costs
Available resources, including existing skills within the organization
Current technology assets and medium-term roadmap
Budget

In this blog post, we’ll walk through the four main pathways available to scaling Generative AI. This includes unpacking the pros and drawbacks of each approach. Plus, get our recommendation for the most logical approach given today’s Generative AI landscape.

abstract image of marbles rolling through a maze symbolizing the complex Generative AI landscape

The Services Approach to Generative AI

A services approach means outsourcing the development and deployment of all Generative AI capabilities to a consulting or SI provider. These providers increasingly offer services to build Generative AI-powered capabilities for their enterprise customers.

There are some upsides to this approach, to be sure. For example, highly customized solutions to the specifics of your technology and operations. Also, service providers can provide scarce expertise that might be absent in your organization. This is especially the case for companies in more traditional industries that have struggled to hire and retain data talent.

Of course, established relationships with consultancies and SIs means work can start quickly. This can be a serious plus given the urgency with which companies are diving into Generative AI today.

On the other hand, when it comes to services, developing new applications means an ongoing relationship is all but required. If you have plans for Generative AI to become an integral part of your overall AI or even business strategy, you risk creating a dependency on an external organization.

In other words, if you’re not developing the knowledge required to build and leverage Generative AI within your company, you’re building an ongoing reliance on an external provider for what will likely become a core strategic initiative.

In addition, total costs can be high, and those high costs might bias project selection toward complex use cases. The problem, of course, is that complex, big-bet use cases can have a higher likelihood of failure.

The Point Solution Path to Generative AI

As the attention on Generative AI increases, ever more startups will develop AI-powered solutions solving specific problems in the organization. For example, AI-powered email generation for sales development representatives, AI-powered contract review for purchasing, etc. The Generative AI application landscape will surely continue to grow in the coming months and years.

One Generative AI strategy could be purchasing dedicated, AI-powered point solutions to augment individual operations and processes throughout the organization.

Many point solutions boast models with very high performance in their specific area. In many cases, they can also provide rapid time to value, as they are nearly ready to use.

The drawbacks, however, are significant:

Point solutions, by definition, are not scalable. They augment one process, but do not provide any benefit to adjacent processes.
Technical debt accumulates as more point solutions become part of critical business processes, creating dependencies on external vendors.
Buying off-the-shelf solutions does not help develop core skills necessary to develop Generative AI-powered capabilities throughout the organization.
The beneficiaries in the business may not trust the results of the AI system, especially if it's a black-box system.

The Do-It-Yourself Route to Generative AI

Ah, the old build vs. buy dilemma. This approach is about developing the internal AI and software development capabilities to build custom Generative AI solutions throughout the organization.

Technology teams in companies of all types have become increasingly sophisticated as they have faced successive waves of innovation. They may be able to build Generative AI-powered solutions, combining open-source software with components provided by cloud computing partners.

These solutions will be well-integrated into the company’s IT stack and will often be highly customized to the needs of the business. Bonus: Developing solutions in-house will preserve maximum IP in the company.

However, the skills required to develop Generative AI-powered solutions are scarce and expensive. Many traditional businesses face challenges recruiting these profiles who are in demand at technology companies.

Plus, developing more Generative AI-powered capabilities internally will gradually add additional maintenance requirements. You may eventually reach a point where you have little time to spend on creating new capabilities.

The skills needed to develop and maintain robust systems is not the core expertise of many traditional businesses. Not every company needs to be a software company. But every company can start building their own Generative AI-powered capabilities if they use an AI Platform.

The AI Platform Strategy

Over the last decade, software platforms have emerged that allow enterprises to build machine learning, natural language processing (NLP), and other AI capabilities into their business. Dataiku is the established leader in this category.

Investing in an AI development platform, like Dataiku, empowers teams to build AI into their operations throughout the organization. This, of course, includes Generative AI and large language model (LLM) capabilities.

For nearly all organizations, investing in an AI platform will be the best choice. It allows you to take the best aspects of the other three possible paths, all while also addressing their shortcomings.

With a platform, service providers can still provide staff augmentation. That is, they can work in the platform to ensure full-time staff are leveraging and maintaining deliverables after the end of their engagement.

In addition, you can still integrate point solutions when they provide an incremental benefit for a particular application. This while maintaining an overall governance structure across all AI initiatives.

Here are some more specific advantages as well as realistic drawbacks to this approach:

Advantages to a Platform Strategy for Generative AI

Modern AI platforms lower the technical barriers to entry. This means a far wider range of employees participate in the development of AI, bringing domain expertise deeper into the process.
A common platform used across the organization fosters the reuse of data and AI assets. This ensures that teams aren’t constantly reinventing the wheel when starting a new project.
A common platform allows for the establishment and enforcement of an AI governance policy. This reduces the risk associated with the broad scaling of AI throughout the organization.

Drawbacks to an AI Platform Strategy

A platform that enables collaboration between different parts of the organization requires a willingness for organizational change.
The platform features that accelerate common tasks may limit the flexibility to adapt to edge cases. Selecting a highly extensible platform like Dataiku mitigates this risk. For example, Dataiku enables users to perform less common tasks using common coding languages.
Some platforms may create the risk of vendor lock-in. However, again, best-in-breed options like Dataiku mitigate this risk through the use of open standards and exportability.