Procurement's guide to selecting the right AI data partner for generative AI projects

Lou Salmen 06 Feb 2024 7 minute read

Procurement plays a pivotal role in generative AI projects, as it’s often involved throughout the AI data services provider selection process. In some cases, as procurement manager, you may function as the primary point of contact between your organization’s data science teams and potential data services providers. In other cases, you might manage the process at a higher level, with a focus on ensuring lowest possible cost and adherence to your organization’s standard buying process. 

Regardless of your company’s procurement style, there are typically four areas of focus for procurement:

  1. Ensuring rules are followed
  2. Coaching for compliance
  3. Controlling costs
  4. Coaching for results

Most companies’ procurement style is based on some combination of all four focus areas, with a greater emphasis one or two of them. Regardless of which way your procurement organization leans, here are a few considerations to keep in mind when selecting an AI data services provider for your generative AI project.

1. Obeying the rules

Organizations that primarily focus on ensuring procurement rules are followed tend to be risk adverse, with fear of litigation being their primary concern. The dangers of this focus include higher administrative costs, slower decision-making and decreased innovation in the supply chain process. 
 
Generative AI is a relatively new type of AI that’s evolving faster than anyone expected. It’s important to understand the typical procurement rules in your organization and how AI – especially generative AI – will challenge those rules. 
 
Generative AI projects emerge quickly, requiring rapid ramp-up and scaling in the midst of legal grey areas related to data use, concerns about the ethical treatment of workers and the risk of public backlash. These are all important considerations that your data services provider should be prepared to address. 
 
Companies with this primary procurement focus tend to use RFIs or RFPs to evaluate different data services vendors. You should work closely with your data science teams and other AI consultants to ensure you ask the right questions and get the answers you need to effectively evaluate potential data services providers. 
 
As a first step, take the time to understand how generative AI works, its potential uses and associated risks. Next, learn about the different components of the data services process, such as recruiting and sourcing, training and onboarding, task execution and QA, performance monitoring and reporting and more. With that understanding, you can anticipate how your buying process may need to be adjusted to account for the unique requirements of generative AI.

2. Coaching for compliance

Organizations that make obeying the rules a top priority for their procurement teams tend to be more procurement-led. Whereas organizations that primarily focus on coaching for compliance tend to be more stakeholder-led. Procurement in the latter type of organization doesn’t seek to control, but rather acts as a facilitator in the buying process.
 
In theory, this leads to a more flexible buying approach that better meets the needs of stakeholders, since they fully understand the specific requirements of their generative AI project. It also shifts responsibility away from procurement teams that are often less familiar with the specific attributes required to select the right AI data services provider.
 
Leaving buying up to stakeholders has its advantages, but with generative AI being relatively new, there are many different pricing approaches that vary by provider. As procurement manager, you should first seek to understand the best practices for buying data services, such as defining clear project objectives, ensuring data diversity and quality and verifying vendor scalability. 
 
Work with project stakeholders to establish a consistent set of buying criteria and guidelines so you can effectively compare proposals across providers. Include thought-provoking questions to help you better evaluate providers, including how they plan to recruit, train and onboard workers onto the project, address issues like bias as they arise and ensure the privacy and security of data in compliance with all applicable regulations. 

3. Controlling costs

Given the current economic landscape, more and more organizations are focused primarily on controlling costs. Similar to organizations that focus on obeying the rules, these organizations run the risk of slower decision-making, as well as selecting a vendor based on an incomplete set of criteria.
 
Organizations with this focus should consider several factors. First, although cost is important, ROI is even more important. Start by understanding how the generative AI model will be used and what cost savings or additional revenue it will deliver down the road. Furthermore, consider the negative implications of selecting a low-cost vendor that may provide low-quality training data, or worse yet, inaccurate data that will end up increasing costs in the long run. Understanding these factors will help you better anticipate the appropriate costs that should be devoted to the project to minimize potential risks and deliver a strong ROI.
 
This type of cost analysis can have major implications for training and fine-tuning generative AI. Although it might seem like a good idea to fight for the lowest cost possible, and there are providers who would be willing to lower their bid to earn your business, it’s important to consider which steps in the data services process might be sacrificed and what major risks could be introduced as a result. 
 
For example, let’s assume you have negotiated an hourly rate for English resources of $10/hour. Considering that the data services provider has their own margin objectives, at that rate, it is likely that the take-home pay for the people performing the work will be lower, likely below minimum wage for a US-based resource. 
 
Two risks have now been introduced: either US resources will be used but underpaid, exposing your organization to negative publicity and legal risks, or only resources from lower-wage regions will be used. If the latter approach is taken, biases may be introduced as a result of only using resources from a particular lower-wage country or region. Either way, the results can be detrimental to your organization’s generative AI project.

4. Coaching for results

Most organizations consider procurement a business driver that can help improve the company’s financial performance. In the highest performing organizations with this procurement focus, a procurement mindset is embedded into every employee and stakeholder. Procurement within these organizations is highly collaborative, with stakeholders coached on how to make the best possible buying decisions.
 
Our recommendation for organizations with this procurement style is the same as what we would recommend to any other – except this type of organization is most likely to implement it. 
 
Work towards building an internal procurement team solely focused on AI. Even if you start with just one person dedicated to the procurement of AI-related services, with the rapid pace at which AI continues to evolve, this function will most certainly grow in the future. As AI becomes increasingly complex, so do its various uses and buying considerations. Start now with a procurement expert that you can trust – someone with a focus on innovation, as well as some knowledge of AI and its many potential uses and benefits. 
 
Be sure to ask AI data services providers about their innovation and thought-leadership initiatives, and request proposal responses that provide different options and scenarios for pricing, processes and tools. Data service providers are often limited to submitting one solution for evaluation – the one that they think will work best based on limited project information. 
 
One way to get additional insights beyond this limitation is to ask providers to share any assumptions they made, as well as any potential issues they see, or share any additional information that was not requested but that should have been. This approach not only provides opportunities for AI data service providers to demonstrate their expertise, but also for both procurement and project stakeholders to learn.
 
In the rapidly evolving AI landscape, successful procurement of training data for generative AI projects requires a comprehensive and strategic approach. By setting clear objectives, controlling costs without compromising on quality and fostering a procurement-savvy culture, organizations can navigate the complexities of AI data services. Thoughtful engagement with AI data services providers that demonstrate innovation and expertise is key. The future of procurement lies in establishing a dedicated, knowledgeable, AI-focused procurement team that can adapt and grow as the world of AI continues to evolve.
 
Evaluating AI data vendors to train or fine-tune your generative AI? Download our checklist to evaluate AI data services providers and get your project off to the right start.
Lou Salmen
Author

Lou Salmen

Strategy and Development Manager, TrainAI
Lou is Strategy and Development Manager of RWS’s TrainAI data services practice, which delivers complex, cutting-edge AI training data solutions to global clients operating across a broad range of industries.  He works closely with the TrainAI team and clients to ensure their AI projects exceed expectations.
 
Lou has more than 15 years’ experience working in sales and business development roles in the AI, translation, localization, IT, and advertising sectors. He holds a bachelor’s degree in Entrepreneurship/Entrepreneurial Studies from University of St. Thomas in St. Paul, Minnesota.
 
All from Lou Salmen