Chapter 03
Research dimension: Tech Foundation
What technology foundations does AI need?
3 min readPublished Published by Teklens
AI needs three technology foundations: infrastructure that runs solutions at scale, integration with the existing core systems and an architecture in which AI components can be swapped. In the DACH analysis, architecture is the strength at 3.22 of 5. Infrastructure (2.70) and compatibility with core systems (2.66) sit below the scale midpoint. Prototypes get built; the path into operation is the bottleneck.
Contents · Short answer
Three findings
- 01
Architecture is more flexible than operations
Q9 architectural flexibility sits at 3.22 ("manageable" to "standardised"). Q7 infrastructure (2.70) and Q8 integration (2.66) stay below 3, both with SD 1.14.[Probst 2026, Appendix B]
- 02
The second-widest gap between the profiles
Tech Foundation separates the advanced profile (4.17) and the early stage (1.95) by 2.22 scale points; only adaptability separates more sharply (F = 76.7 versus 83.1).[Probst 2026, Appendix C]
- 03
The three items measure one picture
With α = 0.863, Tech Foundation is among the two most reliable dimensions. Infrastructure shows the highest item-rest correlation of all 19 items at 0.82.[Probst 2026, Table 3]
Evidence
- Q7 IT Infrastructure2.70
- Q8 Compatibility2.66
- Q9 IT Architecture3.22
Select an item to see its wording, levels and statistics.
Takeaway: Architecture (Q9, 3.22) is the strength. Infrastructure (Q7, 2.70) and integration with core systems (Q8, 2.66) sit below the scale midpoint.
- Unit
- Mean on the maturity scale 1 (least developed) to 5 (most advanced)
- Population
- DACH analysis: 64 cleaned self-assessments from Germany and Switzerland, mostly senior decision-makers
- Denominator
- n = 64 organisations
- Source
- Probst 2026, Appendix B · Appendix B – Mean, standard deviation, median, minimum, maximum and item-rest correlation of all 19 items
Data as a table
| Item | Dimension | M | SD | Median | Min | Max | r (item-rest) |
|---|---|---|---|---|---|---|---|
| Q7 IT Infrastructure | Technology foundations | 2.70 | 1.14 | 3.00 | 1 | 5 | 0.82 |
| Q8 Compatibility | Technology foundations | 2.66 | 1.14 | 3.00 | 1 | 5 | 0.75 |
| Q9 IT Architecture | Technology foundations | 3.22 | 1.03 | 3.00 | 1 | 5 | 0.66 |
Analysis
From prototype into operation
The Tech Foundation dimension holds IT infrastructure, compatibility and IT architecture. Its mean of 2.86 sits below the scale midpoint. The thesis deliberately splits the factor the literature ranks most important, IT infrastructure, in two: operational scalability (Q7) and architectural flexibility (Q9).[Probst 2026, Table 1]
The median of 3 on Q7 stands for "standardised operation: reproducible environments for development and operation". Level 2 reads "individual AI solutions run stably; scaling consumes massive IT capacity". On Q8, level 3 describes "connected to core systems, but via workarounds; high manual maintenance effort". Both distributions spread widely: some organisations run AI-native platforms, others isolated point solutions.[Probst 2026, Appendix A]
Technology separates the profiles almost as sharply as adaptability
For the early stage, Tech Foundation at 1.95 is the second-lowest score after adaptability (1.84). The intermediate profile sits at 3.12, just above the scale midpoint, the advanced profile at 4.17. All three pairwise differences are significant.[Probst 2026, Table 5]
Teklens interpretation
For your product team
The study describes organisations across sectors. Applying it to software product teams is our interpretation, not an empirical finding about product teams.
Our recommendation for engineering leads standing between prototype and operation.
Integration before model choice
The question "which model" is already well answered on Q9. The question "where does the data come from and where do the results go" (Q8) decides whether a feature reaches operation.
Reproducible environments first
Level 3 on Q7 is the threshold from which AI can be run permanently. Whoever lacks it should build it before the next automation.
Context instead of point-to-point
Every integration that only connects one tool with another grows into a maintenance problem. A shared context across Jira, documentation and code keeps the connections in one place.
Scope & limits
- The items measure respondents' assessment, not measured operational indicators such as availability or latency.
- The sample leans towards technology: 43.7% of organisations come from technology, media and telecommunications.
- Q9 asks about swapping AI components such as LLM providers; the answer depends heavily on whether any AI components are in use at all.
Questions & answers
What is the difference between infrastructure and architecture?
Infrastructure (Q7) means the compute, storage and operating resources that carry AI workloads. Architecture (Q9) means the structure of the software that makes AI components interchangeable.
Why is compatibility so low?
Because AI tools often run next to ERP, CRM and databases rather than with them. Level 4 requires automated data flow via APIs, which few organisations reach yet.
Does a cloud platform help automatically?
The data say nothing about that. The items do not distinguish by hosting, but by whether operation, integration and interchangeability are solved.
Sources
- Probst 2026, Table 1 · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Table 1 – Dimensions and factors of the readiness framework (Section 2.5). Sample: Conceptual, no sample · Scale: – · Limits: Factor mapping by the author, based on Ali & Khan (2025) and Dey & Ghose (2025).Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Table 3 · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Table 3 – Reliability of the six dimensions (Section 4.2). Sample: n = 64 cleaned responses, Germany and Switzerland · Scale: Cronbach's α and McDonald's ω · Limits: Three to four items per dimension; α ≥ 0.6 is treated as acceptable in exploratory research.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Table 4 · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Table 4 and Figure 4 – Means and standard deviations of the six dimensions (Section 4.3). Sample: n = 64 cleaned responses, Germany and Switzerland · Scale: Maturity scale 1–5 · Limits: Self-assessment, exploratory, not representative.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Table 5 · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Table 5 and Figure 7 – Dimension scores by profile (Section 4.4.2). Sample: n = 64; profiles of n = 12, 27 and 25 · Scale: Maturity scale 1–5 · Limits: Profiles from K-means on the same six dimensions; differences between profiles are large by construction.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Appendix A · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Appendix A – Full survey instrument (19 items, five levels). Sample: Instrument, no sample · Scale: Maturity scale 1–5 · Limits: Seven of the 19 items carry level labels only, without a written behavioural anchor.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Appendix B · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Appendix B – Mean, standard deviation, median, minimum, maximum and item-rest correlation of all 19 items. Sample: n = 64 cleaned responses, Germany and Switzerland · Scale: Maturity scale 1–5 · Limits: Self-assessment, exploratory, not representative.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.
- Probst 2026, Appendix C · Capturing AI's Potential: How Ready Are Organisations in the DACH Region? (2026). Appendix C – One-way ANOVA and Tukey HSD per dimension. Sample: n = 64 cleaned responses, Germany and Switzerland · Scale: F values, df (2, 61), p · Limits: Describes how sharply the clusters separate on the clustering variables themselves; not a hypothesis test.Probst, F. (2026). Capturing AI's Potential: How Ready Are Organisations in the DACH Region? An Empirical Analysis Using an Integrated TOE–DCT Framework. Master's thesis, University of St.Gallen, Institute of Information Systems and Digital Business. Supervisor: Prof. Dr. Ingrid Bauer-Hänsel; co-supervisor: Prof. Dr. Benjamin van Giffen. St. Gallen, 21 August 2026. Conducted in cooperation with Teklens.