6× Defect Reduction
Formalized requirements engineering at Axioma drove a six-fold drop in production defects and doubled feature-delivery throughput.
Practitioner Notes
Aleksey Tolmachev
Practitioner watching the BA/SA role dissolve.
14 years as a Business & System Analyst across retail (Metro, MediaMarkt, Leroy Merlin), banking (Alfa-Bank), and iGaming. I think the role as a separate function is on a 5-7 year clock — product managers absorb most of it, AI agents absorb the rest. This site documents the transition: what's already moved, what AI can't do, what part of the craft endures.
Systemic Rigor Where the Practice Comes From
40M+ Customers
Specs I authored at Alfa-Bank served 40M+ retail customers and 7,000+ call-center agents across five regions.
€1M+ Programs
As Head of BSA at Metro: €1M+ programs, up to 20 specialists directed, 3–5 concurrent enterprise initiatives.
30+ APIs
Designed 30+ APIs integrating call-center to core banking — voice and chat — at Alfa-Bank.
20+ Interfaces
Managed a single integration landscape spanning 20+ system interfaces during the Metro Lead tenure.
~15 Analysts Mentored
Coached BSAs from junior to mid-senior since 2022, codifying tacit craft into reusable artifacts.
Recent Observations
View Archive → Latest Articles
Why I am writing acceptance criteria four times for the same agent
Gartner predicts 40% of enterprise AI agents will be demoted by 2027. Why I now write acceptance criteria four times for the same agent — one stack per autonomy level.
shift
The Old Analyst Wrote Reports. The New Analyst Writes Instructions.
AI didn't replace analysts. It replaced what analysts produce. The cognitive work is the same. The artifact is different — and that changes hiring.
shiftProblem Framing: Clients Don't Know What They're Asking For. The SA's Job Is to Surface It.
Stage 1 of the SA System Design Cycle. The work most analysts skip — and what it costs when an electronics retailer asks Axioma to 'just build e-commerce'.
Experimental Projects
The Lab
Testing how LLMs respond to search queries. The Proof pillar's home — concrete AI experiments rather than abstract takes. Where the r=0.009 null finding came from.
getaisearchscore.com →