Where AI Value Lands

Gültekin, Pinarbasi, Yazici & Adiguzel — Commercialisation of AI in Türkiye (2025)

9 min read

Gültekin et al. — Commercialisation of AI: A Research on Entrepreneurial Companies (2025)

Partial ingest

Per author’s instruction, not folded fully into the wiki. Ingested only for what bears on weber-taxonomy-2026-gaps. The paper’s main thrust (Turkish-ecosystem-specific commercialisation challenges and BAH-vs-Claessens process models) is left unmaterialised — it is a development-economics / national-ecosystem study, not a taxonomy paper.

Citation

Gültekin, D. G., Pinarbasi, F., Yazici, M., & Adiguzel, Z. (2025). Commercialisation of artificial intelligence: a research on entrepreneurial companies with challenges and opportunities. Business Process Management Journal, 31(2), 605–630. Emerald. DOI: 10.1108/BPMJ-10-2023-0836. Received 26 October 2023; accepted 23 August 2024.

Why this paper exists in the wiki

It is the most recent (2025) peer-reviewed empirical study in the wiki that operationally uses Weber et al.’s (2022) four-pattern AI-startup taxonomy as a classifier, on a sample collected after the cutoff of the original Weber paper. It therefore provides a small but specific test of how the taxonomy holds up as new evidence accumulates — and is the first independent confirmation in this wiki that researchers two-and-a-half years downstream still reach for the Weber classification when they need one. See weber-ai-startup-business-models and ai-startup-business-archetypes-weber for the classifier.

Method and sample (one paragraph)

Semi-structured online interviews with 10 founders / C-level executives of AI-as-core-product startups in Istanbul and Antalya technoparks. Interviews conducted November 2022 (≈one month before ChatGPT’s launch — the sample is effectively pre-FM-era operationally, even though the paper was published 2025). MAXQDA20 structured qualitative content analysis. Sample composition: 5 computer vision, 4 predictive data analytics, 1 language processing. Firm sizes: median 14 staff, < 50 customers, except the single language-processor (LP1) which has > 100,000 customers via consumer transcription/voice service.

What it says about weber-taxonomy-2026-gaps

Headline contribution

On re-reading, this paper’s most decisive contribution to the gaps diagnostic is not that it weakly evidences the existing six dimensions, but that it surfaces a seventh dimension Weber’s morphological method could not encode: a process dimension — commercialisation motion / GTM posture — distinct from the structural channel-geometry of the original gap 4. Section §5 of the paper (BAH vs Claessens) is the empirical anchor; the dimension now lives on the gaps page as gap 7. This is the first gap-page dimension empirically derived from an ingested source rather than from wiki-internal triangulation.

Where it confirms / qualifies the diagnostic

Gap 1 — Model sourcing (in weber-taxonomy-2026-gaps §1). Weak indirect confirmation. Out of 10 firms, only one (CV1, computer vision 3D scene generation) classifies as ai-development-facilitator; nine classify as ai-charged-product-service-provider or ai-data-analytics-provider. This is consistent with the wiki’s reading that the facilitator pattern is the most-squeezed-by-FM-commoditization slot, even on a sample still pre-ChatGPT. But the paper does not ask the model-sourcing question itself — the dimension remains under-evidenced and the Strand A ingest is still needed.

Gap 4 — Distribution architecture, channel-geometry slice only (§4). The single B2C self-serve firm (LP1) is the only firm in the sample with > 100,000 customers; all nine B2B firms have < 50. Direction consistent with distribution-moat — but on close reading this is not actually about channel geometry: LP1 is both B2C and self-serve, while the nine B2B firms are both B2B and demo-led / network-channelled — so the dataset cannot separate the channel effect from the motion effect. The paper does not address the channel-geometry characteristic set on the gaps page (direct / marketplace-listed / platform-embedded / wholesale-API / FM-store / OSS-distributed). The real signal in §4 turns out to be process, not channel — see gap 7 below.

Gap 7 — Commercialisation motion / GTM posture (§5, §5.1, §5.3 — new dimension). Strong empirical support. Section §5 contrasts two literature commercialisation process models:

  • BAH (Booz, Allen & Hamilton 1982) — linear, sequential: new product strategy → idea generation → screening → development → commercialisation (last stage) → post-launch review. Marketing arrives only after development. The dev-first paradigm.
  • Claessens (2015) — interleaved: idea generation → screening → concept dev & testing → marketing strategy & business analysis → product development → market testing → commercialisation → post-launch review. Marketing-strategy and market-test phases occur before full commercialisation.

The paper finds Turkish AI startups default to BAH despite none of them having a deliberate approach or model:

“Choosing the commercialisation process seems not to be a conscious or intentional decision on businesses’ part, it is rather the consequence of the previously implemented practical solutions. Many start-ups tend to follow the BAH model, which places less emphasis on marketing activities… no matter which model was used in practice, the respondents stressed that the process lacked clear-cut steps; it was depicted as a fuzzy process.”

This is the fuzzy / emergent motion: nominally BAH-like in outcome (dev-heavy, marketing-light), but with no formal process at all in practice. Section §5.1 adds an orthogonal internal-vs-external spectrum (own resources / customer-driven iteration vs investor- and partner-led GTM). Sections §5.3 and §5.4 surface auxiliary motion-level mechanics: state-program contests as primary entry channel, demo-as-sales-tool, international-body lobbying as legitimacy-building, and the B2B sales-cycle reality (“one or two sales per year is good” — CV2).

The empirical signal is sharp: of 10 firms, only one (LP1) is on the self-serve / B2C-direct motion; it is the only firm with mass-market reach. The other nine — across computer vision, predictive data analytics, and one development facilitator, across founding years 2000–2021, across staff counts 2–46 — are all on the dev-led, demo-driven, contest-channelled motion, and all have < 50 customers regardless of AI capability or industry. In a sample of 10 this is anecdotal, but it directionally points the same way as the distribution-moat thesis (Cursor’s self-serve motion is half its moat). It is also a process dimension that Weber’s morphological method could not have surfaced — only a qualitative interview method can.

This was the missing dimension in the original six-gap diagnostic. Promoted to gap 7 on the gaps page.

Gap 6 — Ethics / safety / compliance as product (§6). Confirms the dimension exists in the discourse but is not yet productized in this sample. Participants explicitly raise transparency, accountability, and “social benefit integration” as concerns for sustainability (§5.4):

“The focus is on improving transparency and accountability in AI algorithms to prevent harm to society. This will help build trust and open up more possibilities for using AI.” None of the ten firms sells ethics, safety, or compliance as a product feature; the concern is articulated as a macro worry about the field, not as a micro product surface. This independently corroborates Weber et al.’s own 2021 expectation that ethics should have surfaced as a dimension but didn’t — the asymmetry between concern-level salience and product-surface absence is a stable feature across two pre-2023 samples on different geographies. Strand A ingest still required to evidence post-2023 ethics-as-product emergence.

Where it has nothing to say (gaps the paper does not inform)

  • Gap 2 — Autonomy level. The interview guide does not distinguish copilot / agent / autonomous-worker. The four-pattern classification absorbs everything into “Autonomous robots & bots” (none of the sample) or “Process & task support” (most of the sample). No 2026 evidence here.
  • Gap 3 — Orchestration depth. Pre-ChatGPT sample; chains and multi-agent systems are not in the design space.
  • Gap 5 — Data flywheel. The paper repeatedly emphasises customer feedback as the dominant resource (§5.2) but this is human conversational feedback, not the behavioural / trajectory data flywheel of continual-learning-paradigm or distribution-moat. No instances of the Cursor-style behavioural-data accumulation in the sample.

Net effect on weber-taxonomy-2026-gaps confidence

  • Adds a new dimension — gap 7 (Commercialisation motion / GTM posture) — empirically anchored in §5 of this paper. First gap-page dimension derived from an ingested source.
  • Splits the original gap 4 into a pure channel-geometry slice (under-evidenced here) and the process slice that became gap 7.
  • Weakly evidences gaps 1, 6 (under-population of facilitator slot; ethics-as-concern-not-product asymmetry confirmed across two geographies).
  • Leaves gaps 2, 3, 5 unaddressed (sample pre-dates ChatGPT).
  • Does not retire any dimension.
  • Does not validate the fifth-archetype claim (agentic-workflow provider) — the sample has none, consistent with a Nov 2022 cutoff.

The page’s Status and Confidence should remain working-draft until a 2024–2026 post-ChatGPT taxonomy study is ingested (Strand A) — but with the asterisk that gap 7 now has a peer-reviewed empirical anchor, which none of the other six dimensions has. The taxonomy has not been abandoned by the research community; it is still the off-the-shelf classifier of choice in 2024 BISE-adjacent venues — but it now has a documented process-dimension blind spot.

Other notable observations (logged for completeness, not promoted)

These are interesting but bear on the user’s Türkiye-specific commercialisation questions, not on the gaps diagnostic, so they live here as a record rather than as new concept pages:

  • The paper distinguishes BAH (1982) and Claessens (2015) commercialisation process models and finds Turkish AI startups default to BAH (technical-development-heavy, marketing-light) despite none having a “deliberate approach or model for commercialization.”
  • State-program funding (KOSGEP, TÜBİTAK) supports R&D but not marketing/sales — a structural ecosystem mismatch.
  • “Made in Türkiye” bias from both domestic and international buyers / investors; several firms establish secondary entities in Germany to bypass it.
  • Customer-as-co-creator is the dominant resource in the sample — but as human iterative feedback, not as a data asset.

If a future paper section needs emerging-market AI-startup commercialisation patterns, this source is the anchor. It is not the anchor for taxonomy work.

See also

Graph View

Backlinks