# girard-sfeir-ai4it-vs-ai4business-budgets-2027-2026-06-24

## Veille

In-depth opinion piece (point of view) published on **sfeir.com** on June 24, 2026, by **Didier Girard** (Managing Director, SFEIR). **Central thesis**: in 2024 everyone was betting on **AI4Business** (AI in business processes) as the great value reservoir; by 2026 the picture has **reversed** — it is **AI4IT** (AI to produce the information system: code, SDLC, software factory) that is creating **measurable** value. The article *grounds* this thesis in the firm's tech watch: AI4Business disappointment (the MIT study "95% of pilots without ROI," contested but revealing; an **organizational** blockage / Mollick's Hayekian problem) versus quantified AI4IT evidence (Salesforce, Intercom, Raiffeisen, AWS/Bedrock, Atlassian, DORA). Mechanistic explanation: **code verifies itself** (compilation, tests, CI) whereas business processes have neither a compiler nor an immediate feedback loop. **2027 budget consequence**: a **CapEx→OpEx** shift, token price dynamics (rising peak — Fable 5 at 2× Opus — vs inference ÷280 and downward pressure from open weights/desktop), and **AI FinOps** driven by **cost per outcome**. Closes with **4 recommendations for the COMEX**.

## Titre Article

AI4IT vs AI4Business : le renversement, et ce qu'il fait à vos budgets 2027

## Date

2026-06-24

## URL

https://www.sfeir.com/articles/ai4it-vs-ai4business-renversement-budgets-2027/

## Keywords

AI4IT, AI4Business, reversal, 2027 budgets, AI FinOps, cost per outcome, token price, CapEx OpEx, software factory, SDLC, AI ROI, engineering productivity, verification loop, Hayekian problem, shadow AI, MIT 95% study, J-curve, Jevons paradox, Claude Fable 5, Claude Opus 4.8, GLM-5.2, open weights models, desktop inference, Arthur Mensch, Mistral AI, McKinsey, headcount accounting, Didier Girard, SFEIR, thought leadership, change management, pre-sales, CIO, COMEX

## Authors

**Didier Girard** — Managing Director (CTO / DG) de **SFEIR**, ESN française (~1 000 personnes, France · Belgique · Luxembourg · Suisse). Auteur de l'article ; voix éditoriale du cabinet sur la transformation IA des DSI.

## Ton

**Profile**: long-form executive point of view (thought leadership), first person (*"I build 2027 budgets with executives"*), **strategic** register aimed at **COMEX / CIOs**. Technical level kept accessible: it discusses value, ROI, budget trade-offs, FinOps, without burying the decision-maker in jargon.

**Style**: structured by thesis-subheadings (*The 2024 bet → The AI4Business disappointment → The AI4IT reversal → Why this is not a coincidence → The economic consequence → What I recommend to a COMEX*). **Controlled honesty and an owned stance**: Girard states he was *"among the first to challenge the method"* of the MIT study, then turns it around (*"the symptom is true even when the figure is false"*). The argument is **evidence-backed** (every claim is anchored to a quantified, dated case study). Marker phrases: *"the numbers have settled it. And they settled it the other way around,"* *"we're equipping those who already know how to equip themselves,"* *"the next budget battle will not be about the price of the token, but about the cost per outcome."* Crafted metaphors: code that *"verifies itself,"* the J-curve (transition tax), the token-fuel.

## Pense-betes

- **Thesis to remember**: the reversal **AI4IT > AI4Business** on **proven** value creation in 2026. AI4Business is not dead — it is **slower** (reorg, data classification, legal liability) and will come **later**, driven by capabilities built while doing AI4IT.
- **Why the reversal** (the core): AI excels where the output is **verifiable structured text** with an **immediate feedback loop** → **code verifies itself** (compilation, tests, CI; the agent breaks a test in seconds). Business processes have neither a compiler nor a test suite, their truth is diffuse, their liability legal. *"We're equipping those who already know how to equip themselves."*
- **AI4Business disappointment**: **2025 MIT study** ("95% of GenAI pilots without measurable ROI," $30-40B) — **Girard disputes its method** (sample, vague ROI, age) but reads its **persistence** as the real signal of genuine dissatisfaction (*"the symptom is true even when the figure is false"*). Shadow AI thrives where the official project stalls. Cf. [[mit-nanda-genai-divide-95-percent-pilots-fail-legalio-2025-08-23]]. The blockage is **organizational** = Mollick's Hayekian problem (cf. [[mollick-roon-asi-consulting-forward-deployed-engineering-2026-05-10]]).
- **AI4IT evidence cited** (all already documented):
- **Salesforce**: +151% Effective Output, 18× migration, −5% incidents → [[salesforce-tallapragada-how-engineering-became-agentic-2026-05-27]]
- **Intercom**: 3× R&D productivity, 93.6% agent PRs, −50% cost/PR → [[curran-intercom-fin-ideas-2x-nine-months-later-3x-rd-productivity-2026-04-16]]
- **Raiffeisen Bank Ukraine**: −8% headcount, −70% blocking incidents, 7 AI products → tatsyi-raiffeisen-ukraine-ai-engineers-different-not-just-faster-2026-05-05
- **AWS / Bedrock**: 6 people/72 days vs 30/18 months, 100% AI code → ifttd-351-aws-summit-julien-lepine-2026-04-08
- **Atlassian**: +19% PRs, +59-87% for 3-5 adopters → atlassian-ai-native-sdlc-paying-off-rovo-dev-2026-05-31
- **DORA × Google Cloud**: 39% ROI / 8-month payback, greenfield 35-40% vs legacy ≤10% → dora-google-cloud-roi-ai-assisted-software-development-j-curve-2026-04-21
- **2027 budgets — 3 accounting shifts**: 1. **CapEx→OpEx**: the token becomes a **variable, non-deterministic** OpEx charge (5-10× variance, O(n²) context, retry tails) → cf. **Token Budget Wars** gupta-token-budget-wars-marginal-token-utility-2026-05-28. **Order of magnitude cited**: **Arthur Mensch** (CEO Mistral AI), testimony before the **French National Assembly's inquiry commission** on digital vulnerabilities (May 2026) → ~**10% of the payroll budget** allocated to tokens among advanced adopters. 2. **Token price = double trap**: at *constant capacity*, inference ÷**280** in 2 years (Stanford AI Index); **but the peak is rising** — **Claude Fable 5 at $10/$50** per M tokens = **2× Claude Opus 4.8 ($5/$25)**. Downward pressure via **open weights** (**GLM-5.2 ~$1.40/$4.40**, ~1/6 of frontier, **zero marginal cost when self-hosted** → artificial-analysis-glm-5-2-gdpval-aa-open-weights-2026-06-22) and **desktop inference** (DGX Spark, Mac M5 Max). **Jevons**: consumption rises faster than the price falls. 3. **AI FinOps**: **cost per outcome** (not token counting), **rule-based allocation**, **token-to-outcome attribution** as a strategic asset.
- **4 COMEX recommendations**: (1) **fund AI4IT first** (proven ROI, payback < 1 year), separate the AI4IT line from the AI4Business line; (2) **budget for the J-curve** (transition dip); (3) **install token FinOps before** drift sets in; (4) **redefine headcount accounting** (McKinsey 40,000 humans + 20,000 agents → sternfels-mckinsey-60000-people-20000-agents-officechai-2026-01-14; SFEIR *"1,000 people, capacity of 10,000"* → cf. bfmtv-tech-co-business-ia-developpeurs-disparaissent-2026-05-05).
- **Closing line**: *"the next budget battle will not be about the price of the token, but about the cost per outcome."*
- **Meta**: this article is the firm's **published editorial synthesis** of its tech watch — it is the Gold version of SFEIR's doctrine on the topic made public (to be linked to the deliverable `gold/ai4it-vs-ai4business-renversement-2027-2026-06.md`).

## RésuméDe400mots

In this opinion piece published on sfeir.com (June 24, 2026), **Didier Girard** (Managing Director of SFEIR) argues a thesis: the **AI4IT vs AI4Business reversal**. In 2024, the consensus saw **AI4Business** — AI poured into business processes (sales, support, finance) — as the great productivity reservoir; **AI4IT** (AI to produce the information system) was seen as an engineers' topic. Two years later, *"the numbers have settled it, and the other way around."*

**The AI4Business disappointment**: the 2025 MIT study ("95% of GenAI pilots without ROI") is, by Girard's own admission as he disputes its method, questionable — but its **persistence** is the real signal of genuine dissatisfaction: many executives do not see the promised value in their processes. *"The symptom is true even when the figure is false."* The blockage is **organizational** (Mollick's Hayekian problem), not technical.

**The AI4IT reversal** rests on quantified evidence: Salesforce (+151% Effective Output, migration 18× faster, −5% incidents), Intercom (3× R&D productivity, −50% cost/PR), Raiffeisen Bank Ukraine (−8% headcount but 7 new products, −70% blocking incidents), AWS (Bedrock rebuilt by 6 people in 72 days), Atlassian (+19 to +87% PRs), DORA × Google Cloud (39% ROI, 8-month payback). **Why?** Code **verifies itself** (compilation, tests, CI); business processes do not. *"We're equipping those who already know how to equip themselves."*

**The 2027 budget consequence** comes down to three accounting shifts. (1) **CapEx→OpEx**: the token becomes a variable OpEx charge — Arthur Mensch (Mistral) puts it at ~10% of the payroll budget in tokens among advanced adopters. (2) **Token price, a double trap**: at constant capacity, inference has been divided by ~280 in two years, but the peak is rising (Fable 5 at $10/$50 = 2× Opus 4.8), while open models (GLM-5.2) and desktop inference push costs down; the Jevons paradox drives consumption up faster than the price falls. (3) **AI FinOps**: think in terms of **cost per outcome**, allocate by rules, treat token-to-outcome attribution as an asset.

Four recommendations for the COMEX: fund AI4IT first (payback < 1 year), budget for the J-curve, install token FinOps before drift sets in, redefine headcount accounting (humans + agents). Conclusion: *"the next budget battle will not be about the price of the token, but about the cost per outcome."*

## GrapheDeConnaissance

- Didier Girard —dirige→ SFEIR (ORGANISATION, 0.99)
- Didier Girard —publie→ AI4IT vs AI4Business : le renversement (article SFEIR) (DOCUMENT, 0.98)
- Didier Girard —affirme_que→ c'est l'AI4IT, et non l'AI4Business, qui crée la valeur mesurable en 2026 (AFFIRMATION, 0.97)
- AI4IT —surpasse→ AI4Business (CONCEPT, 0.92)
- Didier Girard —affirme_que→ l'IA excelle là où la sortie est un texte structuré vérifiable avec une boucle de feedback immédiate (le code se vérifie tout seul) (AFFIRMATION, 0.93)
- Didier Girard —s_oppose_à→ la méthode de l'étude MIT (95 % des pilotes sans ROI) (DOCUMENT, 0.9)
- Didier Girard —affirme_que→ le symptôme est vrai même quand le chiffre est faux (CITATION, 0.92)
- AI4Business —est_basé_sur→ problème hayékien (information tacite et distribuée) (CONCEPT, 0.9)
- Salesforce —mesure→ +151 % d'Effective Output et migration 18× plus rapide (MESURE, 0.93)
- AWS —mesure→ redéveloppement de Bedrock par 6 personnes en 72 jours (vs 30 / 18 mois) (MESURE, 0.92)
- coût d'inférence —mesure→ divisé par ~280 en deux ans (Stanford AI Index) (MESURE, 0.9)
- Claude Fable 5 —surpasse→ Claude Opus 4.8 (capacité) au prix de $10/$50 le M tokens, soit 2× (MESURE, 0.9)
- Arthur Mensch —affirme_que→ ~10 % du budget salarial est alloué aux tokens chez les adopteurs avancés (AFFIRMATION, 0.85)
- Didier Girard —recommande→ financer l'AI4IT d'abord (ROI prouvé, payback < 1 an) (AFFIRMATION, 0.95)
- Didier Girard —recommande→ piloter le budget 2027 au coût par outcome plutôt qu'au prix du token (AFFIRMATION, 0.95)
- Didier Girard —recommande→ redéfinir la comptabilité des effectifs (humains + agents) (AFFIRMATION, 0.9)

---
Canonical: https://www.thekb.eu/en/fiches/girard-sfeir-ai4it-vs-ai4business-budgets-2027-2026-06-24/