#12 - Did Klarna really manage to replace Workday with AI?

According to the news portal Seeking Alpha, Klarna CEO Sebastian Siemiatkowski announced that the company has shut down its Salesforce instance and plans to do the same with Workday soon. (?? https://seekingalpha.com/news/4144652-klarna-shuts-down-salesforce-as-service-provider-workday-to-meet-same-fate-amid-ai-initiatives)

The reason given? Major internal initiatives to leverage a combination of AI, standardization, and simplification. He further explained that AI solutions have allowed Klarna to build a significantly leaner tech stack. No additional details about these internal initiatives were provided.

Now, you're probably asking: Okay, how could Klarna have pulled this off?

That's exactly the question I'll give you my take on. Not because I have any insider knowledge about Klarna's case—I don't—but because I have a general idea of how you could replace a large HRM system like Workday with the help of AI, standardization, and simplification.

Step 1: The Basics

SaaS products like Workday or other comprehensive "core HR tools" are extremely useful in many companies. In my view, their main strength lies in their focus on process and organizational structure. All features and reports essentially revolve around a strong framework of "Organizations" (e.g., Supervisory Organizations to represent the hierarchy) and "Business Processes" (e.g., Job Requisition for requesting a new role). This framework is essentially a type of database structure that tells the software, "Hey, if someone requests a new role, it follows this reporting line up."

Within each process and structural element, there are certain mandatory data fields that must be filled out for the software to function properly. If an organizational unit doesn't have a designated leader, the approval flow for a job requisition along the hierarchy won't work. Or if there isn't a clean hierarchy of organizational units, your org chart in Workday will be wrong.

From this core structure arises the responsibility of various individuals for different HR processes, which are then mapped into flexible workflows during implementation. For example, you decide whether every job requisition must be approved all the way up to the executive board or not.

Step 2: The Database

To put it simply, the backbone of such SaaS tools is an editable collection of information about the "affiliation" of every person, position, cost center, team unit, and legal entity within the company. From this comes an access rights concept, typically managed through a role-based permission system.

Again, to simplify: SaaS tools for HR administration and process digitization are basically "just" user-friendly frontends for extremely powerful and versatile HR databases in the background.

Once you understand this, you'll start to see how Klarna might have managed to replace Workday using a combination of standardization, simplification, and AI.

Step 3: The (Likely) Role of AI

I've implemented Workday myself at one of my former employers. So I know that the first big hurdle is clearly defining the key HR processes. I imagine Klarna built upon the process maps already defined in their Workday project for each HR process. These diagrams provide a transparent description of who needs to do what at each step of the process. Essentially, they're a guide for entering the relevant data at each step.

An example: For a job requisition, you need to input information like the cost center, position in the org chart, job title, location, job family, and vacancy start date. Typically, the HR managers responsible for the role handle this and pass it along for approval to the respective hiring managers. After that, it goes to recruiters, who post the job and fill the role. This example shows that, strictly speaking, it's just an orchestrated collection of data across defined process steps.

Thanks to a successful Workday rollout, you already have detailed process maps for all HR processes, which can be fed into AI-based tools. The AI then receives a simple Word document with the information or even has a chat or voice interface. This way, the AI can prompt you for the necessary information. With the process diagrams, the AI knows whom to ask for information and when. With products like Microsoft's Azure OpenAI Services, such queries could be handled directly within Microsoft Teams. All information is simultaneously transferred into a structured database in the background. Such databases can be established using programming languages like SQL or cloud-based database solutions built on top of that.

With this setup, you theoretically wouldn't need any additional frontend to manage your HR processes beyond a Microsoft Teams chat. Wild, right?

However, in the backend, you would need some interfaces where knowledgeable data engineers or data scientists can handle database maintenance.

You'd go through this process step by step for all your HR processes, potentially replacing your entire HR admin tool with the described combination of front and backend.

Step 4: Simplification and Standardization

For this approach to work, Klarna must keep its process diversity and the number of "special cases" extremely low. AI-based tools struggle when process flows are vague or ambiguous. Too many process steps involving different data sources or tools also aren't helpful. I'll say it again:

Transparent and interconnected HR processes create integrated and validated HR data streams, which then serve as valuable learning material for AI.

Klarna CEO Siemiatkowski provides various insights in an interview with Sequoia Capital on how Klarna might be proceeding. (?? https://www.youtube.com/watch?v=m3niSE-8ZvE&t=1327s)

A major internal initiative seems focused on how internal knowledge is integrated and shared. What's particularly interesting is that using vendor systems led to silos where knowledge was stored. It seems they’ve tackled this with four key approaches.

1. They use a Knowledge Graph technology to integrate this knowledge into a kind of intelligent wiki. This is essentially a knowledge-focused version of my database proposal from Step 3.

2. He also mentions an internal chatbot named KiKi, which can (likely with generative AI and Retrieval Augmented Augmentation) access this knowledge. Yes, all knowledge—not just HR knowledge.

3. They appear to have built some of their microservices (like generating an employment certificate, for example) around their systems using Slack workflows. This is essentially an alternative to my proposed combination of Microsoft Teams and Power Automate in Step 3.

4. Instead of Workday, they plan to use Deel—but only for payroll processes. This makes a lot of sense, as correct payroll processing can't be handled by (generative) AI alone.

What Are the (Likely) Challenges?

The main issue I see is in information security and data protection. Solutions like Workday come with a multitude of auditable admin functions and reports to ensure that everything stays in check during a data privacy audit. Such functions are extremely valuable, and it's advisable to ensure their accuracy through contracts and service level agreements. It's definitely not easy to build such a governance structure internally with basic tools. For a cloud-based software company like Klarna, it's probably more feasible. After all, their products in the payment processing space are under strict regulatory oversight, which means they have the necessary expertise internally or among their service providers. Additionally, the complexity of internal HR processes in a digital company with around 2,000 employees is likely manageable. The diversity of job groups probably isn't directly comparable to a mid-sized manufacturing company. And the regulatory environment is probably different from your industry.

However, if you're leading HR in a highly digitized company with up to a four-digit number of employees, you should seriously consider the approach I've outlined here. Given the current license fees of typical HR SaaS providers, it’s not out of the question that you could achieve significant efficiencies and create an impressive reference project.

And if you know others in your network who could benefit from ideas like these via email, please recommend they subscribe to my newsletter: [https://www.hr-datenliebe.de/news](https://www.hr-datenliebe.de/news).

Peace out,

Your Data Dan