File No. 1 · current
↑ to contentsMarch 2026 — present
Head of customer service department
Proxy infrastructure service, B2C and B2B · Moscow
A round-the-clock support department: four agents on the line, an AI agent and the head. I joined a place with no quality standard, no training system and no cost calculation. Five months later all of it exists, and it was built under hard resource constraints: there is no budget for off-the-shelf tools, so everything is assembled by hand and from cheap parts. Hence an AI agent for thirty dollars a month instead of a commercial platform, training and methodology written in-house, and a replacement hire costing 0.73 of a monthly salary instead of the market three to six.
Economics
01Calculated the cost per contact, a metric that did not exist in the company in any form: I set the boundaries, the method and the assumptions myself and had them signed off by finance.
02Brought the cost of replacing an employee down to 0.73 of a monthly salary against a market benchmark of three to six. The difference is ramp-up speed: a new hire reaches the line in a week instead of a month of training and six months of reduced output. Tested in practice: one person left, the next started the same day and had the best feedback coverage in the team within days.
Cross-functional
03Ran an independent review of management reporting: found four discrepancies between the product and sales reports and established that cost of goods and margin per product were not being tracked at all. Raised it with the owners.
04I push decisions through neighbouring functions without formal authority over them: an escalation procedure for corporate clients with a 10—20 minute sales response, a one-sided supplier warranty taken apart with the lawyers, an employee moved from contractor status onto the payroll through finance and HR. The head of sales asked me to be his mentor.
05Turned presales out of support into a process: training, a financial model of its contribution to revenue, a method for crediting a sale, bonuses. The B2B lead flow into sales runs on my own script that classifies conversations by seven triggers.
06I run the department payroll budget and the unit economics of the line: cost per contact of 213 rubles for the department including my own payroll, cost per minute of department work around 10 rubles, the economics of hiring and replacing a person.
07Completed “Financial management in service for non-financial managers”, Apex Berg, 2026. I came in with one specific goal: to talk to owners in the language of P&L rather than ticket counts.
Quality and speed
08Raised quality against the standard from 54 % to 85 % and hold it in the target range. The standard itself is mine too: forty-odd versions, with the effect of every change measured.
09Raised customer satisfaction from 42 % to 94 % against a target of 80 %. Relaunched the incentive scheme, introduced procedures and weekly feedback against the quality standard, built training materials, a self-check question bank and certification.
10Hold first response at a median of 12—13 seconds against a 30-second target; time to a substantive answer is down from a hundred seconds to forty.
11Carried the department through a 50 % growth in contact volume with no drop in quality.
Artificial intelligence
12Put an AI agent into production work on the night shift. It closes about a third of the flow at 97.7 % quality, scored against the same standard as people, with 99.95 % availability. Running the model costs around thirty dollars a month.
13Launched an AI prompter as a corporate solution, choosing our own server over a ready-made platform. It guarantees a minimum answer quality regardless of who is on the line today.
14Automated the analytics loop itself: weekly agent scoring against sixteen parameters, detection of corporate customers inside conversations, payroll calculation. A test run against real data matched the manual figure to the ruble.
People and training
15Rebuilt the incentive scheme into two tiers: a base plus added value, with two parameters that zero out the bonus. Talked it through with every agent personally. The effect shows on a short horizon: in the first month after launch quality rose from 76—79 % to 88 % and the quarterly target of 85 % was met ahead of schedule.
16Built a five-stage qualification pipeline: ten courses and a portal, a bank of 284 questions, an AI conversation simulator, an exam and a grade.
17Documented the hiring pipeline with a funnel, roles, a three-lens scorecard and red flags. Started regular one-to-ones with the whole team and weekly individual reviews.
Systems and analysis
18Assembled the department documentation: twenty-three procedures in the corporate knowledge base and a separate agent-facing base covering over ninety percent of questions. Updates run through a four-loop pipeline with full audits for contradictions.
19Wrote the incident plan for mass outages: twelve failure directions, roles, thresholds, compensation, a log. Built from a review of thirty-plus real cases.
20Ran a gap analysis against ISO 18295 across twenty-two sections: seven requirements closed, eleven partially closed, four gaps named openly. This is a starting point, not a finish line: certification lies ahead, but we know where the department stands.
21Diagnosed the main source of friction: a third of all contacts comes from supplier quality, not from how the service works. That finding changed the conversation with the product owner.
22Built a register of the company's critical processes — thirteen processes, eight of them end-to-end — and a founders' dashboard of thirteen indicators, each with a sentence linking it to money and an honest status: measured or not yet.
