Data, AI and the Death of SaaS

For CTOs, digital leads, and ops teams navigating the intersection of SaaS and AI

TL;DR

AI success depends on well-structured data; SaaS tools often make that harder. Throwing away those SaaS tools won’t solve the underlying problem. Instead, organisations need to develop data skills and allocate time to identifying and fixing data structure issues.

The Death of SaaS

Since Klarna’s Sebastian Siemiatkowski announced a downsizing of their use of SaaS, there’s been a lot of discussion about the future of SaaS in the new world of genAI. I don’t think AI will lead to the death of SaaS - despite the headline (you probably know Betteridge’s Law by now, which says “any headline that ends in a question mark can be answered by the word no.”). However, companies must change how they manage their platforms to take advantage of modern AI tools.

The SaaS Model

SaaS is everywhere. Most businesses use Salesforce, Hubspot, or another CRM; they use external CMSs like WordPress or Shopify; manufacturers use Oracle or SAP ERPs. Most of these are simple database apps - with interfaces to view, update and report on business info. Good SaaS companies think hard about the problems their users are trying to solve (managing sales, or websites, or inventory) and design their interfaces to match users' workflows and needs.

Before Salesforce gained so much traction, many software companies developed in-house CRMs -but most eventually adopted Salesforce or an alternative, abandoning their custom efforts to refocus on core products.

Where SaaS Starts to Break

Today, most companies look at alternatives, find a platform they like, and subscribe. The platforms are invariably managed by the domains in which they are used (a sales or sales ops person manages Salesforce admin, marketing or marketing ops manages the CRM, etc). This is where the model starts to break down, because real businesses don’t fit neatly into generic templates.

Most businesses are unique in some way. Even commodified manufacturers try to find and control a niche, whether based on budget, geography or some other factor. SaaS platforms are invariably modified to accommodate the needs of the business - a flag to denote a high-value client, a custom field to group similar products in a CRM, or a dummy BOM to manage a challenging logistics situation.

This quickly takes many companies to the same point - you have a two-page document about how to run a sales report without crashing the app, or you have to add a special prefix to your image files to make sure they appear on your website, and so on. Over time, your SaaS setup becomes brittle - overly tailored, hard to change, and riddled with workarounds. Now, you can’t experiment with new ways of managing clients without breaking your processes. Changes that used to be small now take weeks.

Will AI save the day? In this situation, you can automate those reports or double-check your image files before publishing. Still, you are adding another fixed layer on top of an already brittle platform, making more significant future changes harder rather than easier.

Why SaaS Data Structures Break Down

The ultimate cause of these problems is poor data structure.

The relationship between a company and its clients, or between products and product ranges, is complex. Adding a flag, a dummy code, or some other construct is quick to make a report run. But now, the database that stores your information is a worse reflection of your situation than before.

When a data structure does a good job of reflecting reality, then incremental changes are more straightforward. When a data structure does a bad job of reflecting reality, there is a much higher chance that an incremental change will break the bad model, meaning that the model must be fixed first in some way before the change can be made.

What does AI have to do with it? Companies are becoming increasingly aware of these issues as they pursue and struggle with AI initiatives. Effective AI products need well-structured data. Without it, each AI project also necessarily includes time to fix historic data issues. Sometimes this is known in advance, and other times it adds months onto a project that was only supposed to last weeks.

So, Should I Ditch Salesforce?

You don’t have to rebuild everything from scratch. Swapping one SaaS for another will not solve this issue. The best way to solve this problem is to comprehensively review your SaaS data. If you have a digital work programme, you should bring in someone with data design skills to review your SaaS platforms. Focus on data structures that have grown organically, especially those that have evolved without clear ownership - these are often the source of the most painful complexity. Work with domain teams to rationalise those structures.

You will always have to accommodate some changes to your data over time. Whether entering a new market or launching a new product, expect to make data schema changes with some frequency. Ask yourself who in your organisation has the skills to assess changes, and ensure that processes are in place to keep your data clean and well-structured.

What Skills Will Win the Future?

A friend observed that these data issues are much less prominent in (in-house) product data than SaaS data. I think it is likely that product managers and backend engineers tend to also be good at mapping information to data, and since they manage product change requests, those changes tend to come with fewer issues. Product managers and engineers are trained to think in systems and schemas. We need to bring that same thinking to our SaaS ecosystems.

The companies that can rise to this challenge will be at a significant advantage - every effective AI product is built on a strong foundation of data. Can an AI make sense of your company data today? Or is it a jumbled and hard to interpret mess? If you plan on going deeper into AI, I would strongly recommend exploring your SaaS data and seeing what state you’re in.