Should we throw away packaged software for AI development? Another perspective to consider

I wanted to add to my colleague Jon Reed’s article. Klarna’s plans to phase out Workday and Salesforce software When they build their own applications through AI. Jon posed questions to Workday executives about this and added his own analysis.

For the background, Inc. magazine noted:

Fintech company Klarna is cutting ties with two of the largest enterprise software providers to automate its services Artificial intelligenceAnd the company says it could potentially eliminate more.

Co-founder of Klarna and CEO Sebastian Siemiatkowski the financial institution, which explained the rationale in a recent conference call Seeking Alpha reportedKlarna is no longer using Salesforce, a platform that collects and packages sales and marketing data for businesses. The company has also removed HR and recruiting platform Workday from its tech stack, a Klarna spokesperson confirmed to Inc.

Here are a few things that require additional thought/review.

1. Some capabilities are particularly difficult to build in some applications. For an AI coding tool, creating an entry screen to enter debits and credits should be a breeze. However, financial software is full of really complex functions that you won’t find coded in accounting principles. For example, producing financial statements (such as balance sheets, income statements, and sources and uses of funds) is a very complex problem. For decades, programmers have tried to collect certain values ​​to paint into a specific summary balance field, perform calculations on it, etc. Each company has different accounting collection rules, and may place different values ​​in different accounting periods based on the currency of a transaction, transaction date, effective date, etc. Also, let’s not forget that consolidated financial data is more than just a simple aggregation. Accumulated balances may also require the addition of accruals and reversals. While AI tools can certainly automate large portions of accounting activities, it’s the exceptional cases that will confuse AI tools. And when it comes to AI writing code to mimic the structure of financial statements, that may be a step too far for now.
2. Payroll is an extremely cranky HR app. In addition to having to deal with potentially thousands of local, state, and federal government requirements, payroll solutions must also comply with union requirements, integration requirements (for example, a payroll-general ledger interface or the need to connect to hundreds of different benefit providers), complex overtime calculations, shift differentials, garnishment rules, etc. Yes, AI tools can be trained on regulations, policy guidelines, and more.to learn‘Many of these requirements but will they happen?’interpret‘ Are these true in every case? I doubt it. And let’s not forget how much variability there can be in a company’s payout requirements globally.

The Inc. article also included these observations from HR industry expert Josh Bersin:

“Human resources technology analyst Josh Bersin is skeptical that the payments company can effectively replace Workday. “Systems like Workday have decades-old workflows and complex data structures, including payroll, time, and attendance,” he told Inc. “If Klarna wants an engineering team to build all of this, they’re going to end up in a black hole of system features, not to mention user experience.”

During the 1960s – 1980s many companies set up their own Payroll systems. I coded, developed, repaired, etc. many of these. But one by one, companies abandoned them because each came with its own long-tail support. The software was expensive to build and maintain. Worse, some new requirements (for example, from new regulations or internal policy changes) often came as a surprise and had to be implemented immediately.

This particular world requires a lot of people: people who research problems, people who design and add functionality, people who respond to user queries, IT people who run and patch the application, etc. These solutions also need other software (e.g. security software, time and attendance software/hardware, backup software, databases, etc.) to work.

Bersin is right when he describes a black hole. And worse, it is the misallocation of a company’s capital, personnel, and technologies to develop specialized applications for non-strategic functions. Here is a classic graphic from the 1990s that describes this exact decision dynamic:

As a result, custom solutions have often been replaced by packaged software because packaged solutions:

• Releasing a number of technical customer staff to work on more strategic applications
• Less expensive to maintain compared to custom options
• Developed and kept up to date by the vendor for technical and compatibility reasons
• Developed and supported by teams with a form factor that is typically larger than what a single private user firm would assemble
• The technical debt that a custom application can accumulate is rarely

Also, as Bersin points out, even if an AI tool can guess what an application’s data dictionary should contain, remember that successful packaged software applications are often part of larger application packages. AI tools also need to understand the exact data dictionary requirements and all the explicit and implicit relationships between all the data elements. This is a big AI training challenge. If you miss a few of these, the resulting application software can produce really bad results.

This brings us to Klarna’s bold statement

Putting aside the hype surrounding this announcement, we can have a reasonable discussion about what software developers should be creating, what they should be using AI to assist them in their efforts, and what software should be left to third-party vendors. For example, applications like Fixed Asset Accounting have been around for decades. A lot tactical and provides little to no strategic value. Except for a few software developers at a few software companies, no company should allocate capital or personnel to building a dedicated Fixed Asset system.

My opinion

• Some applications, such as succession management, can be completely replaced by an AI tool. The new capability should be able to dynamically generate succession plans on demand and create any possible succession scenario based on a user’s wishes. If a company has a well-populated HR database that an AI tool can train on, that’s a good use of AI.
• Some applications, like payroll, will continue to thwart AI tools. You can make a good piece of payroll code that works, but payroll is not something you can do yourself.kind of‘ Let it be true. Companies will 100% always want a perfect payroll solution, otherwise they will end up in litigation and regulatory hell. Some of the largest lawsuits in the history of application software (and most of them by volume) are Payroll related. Therefore, for litigation prevention and mutual indemnity reasons, I strongly advise companies to leave Payroll functionality and processing to companies that make it their primary (if not sole) responsibility.
• Some companies will use AI to create large pieces of code for their operational systemsThese solutions may have fewer regulatory requirements, but some applications will likely have some poor functionality that may be difficult for an AI tool to fully and accurately grasp.
• Nowhere have I seen companies address the potential issue of technical debt associated with custom/AI-built apps. I’m still working on this potential issue.
• Every major wave of innovation in technology causes companies to re-evaluate their application software strategies. That’s what companies need to do right now. Sometimes companies try to fight innovation (remember the cloud deniers who wanted to stay on-premises forever?). Some will be early adopters of new innovations. But many will wait a bit to see how others tackle new approaches, strategies, and innovations.

Klarna’s announcement makes us think and question our perspectives. That’s fine, and maybe it’s time to really look into it. Something Jon Reed pointed out, and something I want to strongly emphasize, is that software vendors REALLY need to have a serious discussion about this. Why?

App vendors have already started using AI to develop large pieces of app code. However, vendors don’t seem to want to pass on the cost savings to their customers. This is a greedy strategy that could really backfire on vendors over time. Every vendor needs to ask themselves how they can significantly reduce their cost profile to be super competitive in every market in the world.

Suppliers also need to develop product roadmaps that reflect much more thinking than AI sprinkled over legacy applications and legacy process workflows. Suppliers are far behind in this. We need to hear suppliers deliver brand new types of applications that solve new problems that aren’t 400 years old, like accounting.

It’s time to rethink….