Quick Answer: A 500-person company spending $40,000 a month on SaaS licenses can replace 80% of those data-entry seats with deterministic API endpoints costing under $400 a month. Klarna is proving this math works at scale by actively dismantling its Salesforce and Workday infrastructure. The driving force isn't advanced artificial intelligence, but raw unit economics: consumption-based LLM API calls cost a fraction of traditional enterprise software subscriptions.

The Cold Facts: Klarna Rips Out Core SaaS

A 500-person enterprise spending $40,000 a month on licenses can replace 80% of data-entry seats with deterministic API endpoints costing under $400 a month. That is the financial shockwave currently hitting the enterprise software market.

Klarna is proving this math works right now. They are actively shutting down their enterprise deployments of Salesforce and Workday.

According to CEO Sebastian Siemiatkowski, the Swedish fintech is replacing these foundational SaaS platforms with custom, in-house systems powered by artificial intelligence. This is an intentional architectural shift. It is not a cost-cutting rumor. It is not a pilot program.

The transition is already happening. Klarna is systematically migrating core CRM and HR functions off proprietary third-party platforms. They are building internal tools that use large language models (LLMs) to handle workflows previously managed within the Salesforce and Workday ecosystems.

Siemiatkowski stated that internal AI initiatives have advanced to the point where maintaining expensive, external SaaS subscriptions is no longer justifiable. For twenty years, the rule was: buy, don't build. Klarna just proved that rule is obsolete. They are betting their proprietary, AI-driven infrastructure can deliver the necessary functionality without the constraints of legacy software architectures.

The Contrarian Read: It's About Margins, Not Magic

The prevailing narrative suggests Klarna is building a hyper-intelligent, omniscient AI to run its HR and sales departments. They aren't. They are building cheap, deterministic API endpoints.

The vulnerability of enterprise SaaS isn't a lack of flashy AI features. The fatal flaw is the per-seat pricing model itself. This is especially true when the "seat" is occupied by an employee doing basic data entry.

Consider the current state of enterprise software. Salesforce recently pushed 9% renewal price hikes across the board. Meanwhile, enterprise IT forums on Reddit are flooded with complaints about Workday's convoluted user interface, highlighting it as a major source of daily employee friction. Companies are paying a premium for software their employees actively dislike using.

Klarna doesn't need Artificial General Intelligence to replace these systems. They need basic LLM routing. Most interactions with Salesforce or Workday involve retrieving text, updating a status, or logging a simple interaction. You don't need a $150-a-month license for a human to click through three screens to update a record. A deterministic API call, triggered by a natural language query via Slack or Teams, can execute that same database update instantly.

The Math: APIs vs. Per-Seat Pricing

The financial logic behind ripping out legacy SaaS is brutal and undeniable.

Per-seat pricing assumes a human needs a graphical user interface (GUI) to interact with the system. API pricing assumes a script or an LLM needs an endpoint.

Feature	Legacy SaaS (500 Seats)	Custom AI Routing
Monthly Cost	~$40,000+	<$400 (API Tokens)
Deployment Speed	Months (Implementation Partners)	Weeks (Internal Engineering)
Maintenance	High (Dedicated Admins)	Low (Automated Scripts)
User Interface	Proprietary, rigid GUI	Natural language (Slack/Teams)

The enterprise software industry has spent two decades convincing companies they need complex, proprietary interfaces to manage their data. Klarna's move exposes the reality: the interface is often the most expensive and least necessary part of the stack. When you decouple the database from the UI, the justification for per-seat pricing collapses.

You aren't replacing the underlying database; you are replacing the expensive UI layer. The data still lives in a secure repository—an owned data lake (like Highstory) or an internal warehouse. The difference is how that data is accessed and modified.

Instead of paying Salesforce for 400 sales development reps to log calls, you pay OpenAI or Anthropic fractions of a cent per interaction to parse a call transcript and update the database directly via API.

The losers in this shift are clear: companies reliant on per-seat revenue models for basic workflow management, including Salesforce, Workday, and Hubspot. The winners are the infrastructure providers selling raw compute and API access: AWS, Azure, OpenAI, and Anthropic. The margin that SaaS companies previously captured via per-seat licensing is being aggressively compressed by the raw unit economics of LLM APIs.

The Playbook

This isn't a theoretical exercise for next year's roadmap. The arbitrage opportunity exists right now.

Audit: Identify the Data-Entry Seats. Pull usage reports for your top three SaaS deployments. Identify the percentage of users whose primary interaction is logging simple data (status updates, call notes, time-off requests). If more than 50% of a tool's user base falls into this category, that platform is a prime target for replacement. Stop paying $100+ a month for a glorified data-entry form.

Pivot: Build the Weekend Prototype. Do not attempt a massive rip-and-replace on day one. Scope a micro-workflow. Take one specific task currently trapped in a high-cost SaaS UI—for example, updating a lead status after a discovery call. Build a prototype using a low-code tool like Retool connected to an LLM API. Prove that you can execute the database update via a natural language prompt in Slack for pennies, bypassing the SaaS interface entirely.

Moat: Decouple the Data. Begin extracting your core business data from proprietary SaaS databases. You cannot build custom AI routing if your data is hostage to a vendor's closed ecosystem. Start mirroring critical CRM and HR data into an owned data lake or warehouse. Prepare your architecture for an API-first interface model, where the underlying data source is agnostic to the tool used to access it.

FAQ

Why is Klarna dropping Salesforce? Klarna is dropping Salesforce because they can build custom, AI-driven internal tools that handle CRM workflows at a fraction of the cost. The company determined that paying per-seat licenses for legacy SaaS is no longer economically justifiable when LLM APIs can perform the same data entry and retrieval tasks significantly cheaper.

Is custom AI cheaper than enterprise SaaS? Yes, significantly. Enterprise SaaS relies on per-seat pricing, which often costs tens of thousands of dollars monthly for a mid-sized company. Custom AI leverages consumption-based API pricing (paying only for the tokens used), which can reduce costs for data-entry heavy workflows by orders of magnitude, often turning a $40,000 monthly expense into a $400 one.

Can LLMs really replace Workday? Yes, for the majority of users. You don't need a complex UI to log PTO or update an address. Deterministic API endpoints, triggered by simple Slack commands, can replace the clunky interface while executing the same backend database updates instantly.