AI as a Knockout Criterion in ERP Tenders: Why SMEs Are Getting It Wrong

It sounds like a forward-thinking decision: a mid-sized company adds AI-powered forecasting and intelligent automation as a hard knockout criterion to its ERP requirements specification. After all, the new system shouldn’t be outdated before it’s even implemented.

I see this regularly in tenders that cross my desk. And every time, I say the same thing: This is one of the most expensive mistakes you can make in ERP selection. This is a central aspect of my ERP consulting for SMEs.

AI Requirements in ERP Tenders: Why the Market Cannot Deliver

If you judge ERP vendors by their presentations, you might think AI is already standard. Automated invoice processing, intelligent demand forecasting, AI-powered anomaly detection – it’s all in the current product brochures.

Reality looks different.

The actual AI maturity level of most ERP systems in the DACH market is still far behind the promises in 2026. Many advertised AI features are in beta, require extensive data foundations, or are only available for specific industries. Anyone who formulates AI as a hard knockout criterion reduces their vendor market to a handful of large platforms – and potentially excludes the very systems that would be the best functional fit.

AI as Knockout vs. Evaluation Criterion in ERP Selection

In systematic criteria assessment – following the UfAB methodology or MoSCoW framework – there is a fundamental difference between a knockout criterion (A-criterion) and an evaluation criterion (B-criterion).

A knockout criterion means: if a vendor doesn’t meet it, they’re out – regardless of how well they score in every other area. An evaluation criterion, on the other hand, feeds into the weighted overall assessment. A vendor without AI functionality can compensate for this disadvantage through strengths in other areas.

In practice, I regularly see requirements catalogues where 70 to 80 percent of all criteria are classified as knockout criteria. That’s far too many. Recognised frameworks like MoSCoW recommend a maximum of 60 percent must-have requirements. Anything above that is counterproductive – because the more knockout criteria, the fewer vendors remain.

For AI functionality, the recommendation for 2026 is clear: classify it as an evaluation criterion – with a clear weighting factor that reflects its strategic importance without unnecessarily restricting the market.

The Four-Step Decision Tree for AI Requirements

Instead of blanket-setting AI as a knockout criterion, I use a structured decision tree in my projects:

Step 1 – Legally mandatory? Is there a legal requirement for AI use in this area? In most cases: No. So no automatic knockout criterion.

Step 2 – Core capability? Is AI necessary for the core operation of the ERP system? Financial accounting, procurement and sales work without AI. The answer is almost always: No.

Step 3 – Mission-critical? Would the absence of AI jeopardise business operations? Rarely – though there are exceptions for data-driven business models.

Step 4 – Available in the market? Do enough vendors offer this functionality? For advanced AI integration: not yet across the board.

Result: AI requirements land at level B or C in most cases – as a weighted evaluation criterion or as an information criterion for future-proofing.

AI Readiness in ERP: What Actually Belongs in the Requirements Specification

Rather than formulating “AI-powered forecasting” as a knockout criterion, I recommend concrete, verifiable requirements:

• Open API architecture (knockout criterion): The system must enable integration of external AI services through documented interfaces.
• Data export and access (knockout criterion): All relevant data must be exportable in standardised formats for use with external analytics tools.
• Vendor AI roadmap (evaluation criterion): The vendor must present a documented AI strategy with a concrete timeline.
• Existing AI modules (evaluation criterion): Already available AI functionality feeds positively into the evaluation.

This ensures your new ERP system is AI-ready without unnecessarily restricting the vendor market.

The AI Readiness Trap

Another mistake I frequently see: companies demand AI functionality without checking their own prerequisites. AI-powered forecasting requires clean, historical data in sufficient volume. Anyone who has been working with inconsistent master data for years won’t get usable results even with the best AI module.

Before adding AI to the requirements, take an honest look at your own data quality. Without a clean data foundation, any AI requirement is pointless.

And for those wondering about the regulatory framework for AI use: the EU AI Act 2026 brings concrete obligations that should already be considered during system selection.

Bundling Three AI Types? Even Riskier

It becomes particularly problematic when companies don’t just require individual AI functions but bundle different AI types as a single knockout criterion: predictive analytics, process mining and generative AI in one criterion. Each of these areas has a different level of market maturity. Bundling them in one knockout criterion is the surest way to reduce the vendor market to one or two players.

The better strategy: define each AI area individually as an evaluation criterion with its own weighting. This preserves differentiation and shows you where each vendor has its strengths.

Conclusion: AI Yes, But as Investment Protection – Not as an Exclusion Reason

SMEs need ERP systems that are ready for AI. They don’t need an ERP selection process that shrinks to two vendors because of overambitious AI requirements.

Those who formulate AI functionality as a weighted evaluation criterion rather than a hard knockout criterion get:

• A broader vendor selection
• Better negotiating positions
• A system that works today and is AI-capable tomorrow

Experience from numerous ERP selection projects shows: the best results come when companies clearly distinguish between indispensable core functions and strategic future investments. AI clearly belongs in the second category in 2026.

The question isn’t: “Does the ERP have AI?” The question is: “Can the ERP do AI?”

Those who understand this distinction will make the smarter decision in 2026.