PARMA AI Strategy: 12 Sections, 15 Departments
R&D as "startup within company", ROI 10x target, ₽31M actual / ₽48M projected, 15 departments with AI plans
What doesn't work
IT company with 600+ employees: AI was applied chaotically (<5% productivity contribution), effects unmeasured, initiatives never completed, no unified strategy. No infrastructure for scaling AI tools. Licenses and instruments chosen ad hoc. Competitors (Gazprom, T-Bank) were already systematically deploying AI.
Architectural approach
Corporate AI strategy with 12 sections: (1) R&D group as "startup" with trust overdraft and balance sheet, (2) deployment across 15 specializations, (3) training and ambassadors, (4) case execution, (5) governance, (6) metrics and KPIs, (7) ethics and legal, (8) financial model, (9) communication, (10) AI-OC (support), (11) infrastructure, (12) initiative portfolio. Effect attribution model by stage: hypothesis 15% → search 10% → validation 15% → creation 45% → deployment 15%.
What made it hard
AI adoption <5% at start — people didn't believe AI would help them specifically, not just 'somewhere at Google.' R&D as a startup with overdraft — needed to simultaneously show quick wins and build long-term strategy. 15 departments with completely different specifics — AI plans for Backend and InfoSec are two different worlds. Measuring effect: every department tried to either inflate or dismiss AI's contribution.
My role & contribution
CTO
Author of the corporate AI strategy (1000+ pages, 12 sections). Created the R&D group as a 'startup' with dual balance sheet. Developed the financial model (effect formula, 5-stage attribution). Prepared individual AI plans for 15 departments. Justified to executive leadership.
How it looks
How it works
R&D group: startup overdraft (2 architects × 3 months), dual balance (actual + projected), target — breakeven in 3-6 months, then 10x ROI. Formula: Effect = Σ(departments) + R&D + individual − capital expenses. Example: ₽45M departments + ₽6M individual − ₽20M = ₽31M actual, ₽48M projected. RACI matrices for every process. TFS tracking: granular tasks ≤16h with AI tool attribution. 15 departments with individual plans: Backend, Frontend, BA, Testing, Design, InfoSec, SysAdmin, DB, System Analysis, PMO, Tech Docs, Functional Architecture, Technical Architecture, Support, Internal Security. POC scenarios, tool registry, AI code of conduct.
Why this way
R&D as "startup with balance sheet" instead of centralized AI department
AI Center of Excellence: separate department with budget, deploying AI top-down
Centralized department: no link to department economics, unmeasurable effect, adoption depends on administrative pressure. R&D startup: trust overdraft + balance sheet + joint effect attribution — both sides are incentivized.
R&D with dual balance (actual/projected), staged effect attribution, 10x ROI target. 15 departments with AI plans
Results
- 01
- R&D ROI: 10x target (Q1 2026)
- 02
- Effect: ₽31M actual / ₽48M projected
- 03
- 15 departments with AI plans
- 04
- 12 strategy sections, RACI for each
- 05
- AI productivity contribution: up to 25%
- 06
- Effect attribution: 15%/10%/15%/45%/15% by stage
- 07
- Tool and model registry
Impact on business
R&D recoups startup overdraft in 3-6 months, targets 10x ROI. Corporate effect: ₽31M actual (departments ₽45M + individual ₽6M − capital ₽20M). 15 departments receive individual AI plans — from Backend to InfoSec. Joint effect attribution (R&D + department) accelerates adoption. AI productivity contribution reaches 25% in support and development ops. Example: document processing automation (2h→1h × 200 cases/month = ₽200K/month savings).
Algorithms & patterns
Technologies
- LLM
- GPT-4
- AI Assistants
- Prompt Engineering
- TFS
- Confluence
- RACI
- POC Framework
- AI Governance