Understand the product
I hired editors for each product’s technical or marketing needs. I want editors to understand the context and contribute to product decisions.
See it in practicePERSONAL RECORD / 001
Ivan Vukolov
I lead content teams for voice and AI products.
I’m responsible for the team, how we work and the quality of our content. I work closely with product, design and ML.
At MTS, I lead a content team working on VoiceTech and Trust & Safety services.
Previously — Marusya, VK’s voice assistant.
PDF · 2 pages
I want editors to understand the product and have a say in how it works.
PROTOCOLS / 001–004
How I lead
I set direction, agree on responsibilities with other teams and give editors room to work independently.
Take work through to approval
I agreed on responsibilities and approvals with product, design and legal teams. Editors took copy from the initial brief through to approval.
See it in practiceTry a new task myself first
I completed the reference-dialogue pilot for Marusya myself. That helped me understand the work, estimate the effort and allocate it among editors.
See it in practiceDelegate ownership
I initially designed and approved changes to MTS Secretary’s dialogue flows myself. Once the process was in place, I gave an editor responsibility for the work; QA teams joined the testing.
See it in practicePROJECT ARCHIVE / 003
Content as part of the product.
Three leadership challenges at MTS and VK.
01Built a content team from scratch.I hired editors to match each product’s needs. Product teams gained a clear process from brief to approved copy.02The content team took ownership of dialogue.I persuaded leadership to put the content team in charge of dialogue. We redesigned the flows with product and ML; average dialogue depth increased.03Built a content workflow with ML.I completed a pilot myself, then organized the editors’ work. The team delivered the dataset on time and met the ML team’s quality expectations.
RECORDS02
TXTMaking a voiceNatural speech, ML and voice coachingVIDTalk recordingTalk and Q&A · YouTubePROJECT RECORD / 01
Built a content team from scratch.
I hired editors to match each product’s needs. Product teams gained a clear process from brief to approved copy.
The situation
MTS’s voice products needed copy for websites, marketing, bots, SMS and push notifications. There was no content team in the division. I proposed taking responsibility for organizing the work.
My decisions
I hired editors for the technical or marketing needs of each product. With product teams, I agreed on priorities and responsibilities, and introduced workload planning. When the team became smaller, I reassigned product areas and revised our working agreements.
What changed
Editors developed product knowledge and took copy through product, design and legal approval. Product teams knew who to approach and how the process worked.
Team projects
The team now works on VoiceTech and the Trust & Safety service Zashchitnik. I also supervised an editor’s work on an updated, youth-oriented brand voice for QIQ. QIQ is an MTS mobile service managed through an AI chat.
PROJECT RECORD / 02
The content team took ownership of dialogue.
I persuaded leadership to put the content team in charge of dialogue. We redesigned the flows with product and ML; average dialogue depth increased.
The situation
MTS Secretary answers calls on a user’s behalf. Fixing individual responses wasn’t enough; the conversation logic also needed to change.
My decisions
Research and call drop-off data guided the work. I proposed a brief “Hello” instead of a long introduction. Together with an editor, the product manager and the ML team, we refined responses and how the assistant chose a path based on the caller’s meaning. I initially designed, built and approved changes myself. Once the process was in place, I gave an editor responsibility for the work; QA teams joined the testing.
What changed
The content team analyzed conversations, designed dialogue flows, and built and tested scenarios. Developers reviewed the changes before release. Calls progressed further through the scenario on average, giving callers more opportunity to explain why they called.
Average scenario steps per call1.3 → 4.5
Before1.3
After4.5
How far the conversation progresses through the dialogue flow. Scale: 0–5 steps.
PROJECT RECORD / 03
Built a content workflow with ML.
I completed a pilot myself, then organized the editors’ work. The team delivered the dataset on time and met the ML team’s quality expectations.
The situation
Marusya is VK’s voice assistant. For its conversation feature, the team was preparing a move to replies generated by a large language model (LLM). We needed reference dialogues in natural Russian, with replies short enough to follow by ear.
My decisions
I completed the pilot work myself to estimate the effort and understand what the editors would need. I then allocated work and set up quality review. With ML, we agreed on the criteria, volume and format of the dialogues.
What changed
The dialogues met the ML team’s quality expectations. The editors gained experience with a new type of work, and we continued collaborating with ML on other projects.
VOICE PRODUCTION / MTS SECRETARY
Helped MTS Secretary sound natural.
For MTS Secretary’s new voice, we wanted natural speech: pauses, hesitations and everyday phrasing. I made the case for that direction with leadership and aligned the brief with ML requirements.
I prepared scripts, briefed the voice actors, listened to takes and coached their delivery. Everyday situations helped: imagine getting to work and telling colleagues what happened on the way.
almost58hours of recorded speech
From the second recording cycle, dedicated to creating a new voice from scratch.
VIDEO ARCHIVE / 001
How content teams shape a voice product.
I share how our team redesigned dialogue for MTS Secretary, a call-answering assistant, and worked on its conversational voice. Includes a Q&A on dialogue design and writing for speech.
The full talk is in Russian.
EDITORIAL ATTRIBUTES / 007
S.P.E.C.I.A.L.
What I do myself, what I lead and what I’m experimenting with.
Strategy
I set directionI agree on priorities and the content team’s responsibilities with product teams.
See the workPeople
I leadI hired editors to match product needs, allocated work and delegated ownership.
See the workEditing
I write and editI worked on Marusya’s responses, dialogue flows and editorial guidelines, and prepared scripts for MTS Secretary’s new voice.
See the workConversation
I designI redesigned MTS Secretary’s conversation logic and responses with product and ML.
See the workAI and ML
I organize the workI agreed on requirements with ML and led the preparation of reference dialogues for Marusya’s LLM.
See the workAnalytics
I use evidenceI used research and call drop-off data supplied by the product manager to guide dialogue changes.
See the workPrototypes
I experimentI experiment with AI-assisted prototypes to try out ideas in practice.
EMPLOYMENT RECORD / 006
From media to voice products.
At VK, I progressed from editor to head of content for Marusya. Before voice products, I worked in news, video and special projects.
- 01
Aug 2024 — Present
MTS
Head of Content, Voice Products
The team currently works on VoiceTech, MTS’s voice and AI services, and Zashchitnik, a spam and fraud-call protection service within Trust & Safety.
- 02
Jan 2023 — Aug 2024
VK · Marusya
Head of Content, Marusya
Led product content for VK’s voice assistant and the team’s joint projects with ML.
- 03
Mar 2021 — Dec 2022
VK · mail.ru / Marusya
Content Editor, mail.ru and Marusya
Worked on responses, dialogue flows, the knowledge base and editorial guidelines for the voice assistant.
- 04
Jan 2019 — Mar 2021
TASS
Senior Editor, Video
Led a team of editors, scriptwriters, designers and video editors.
- 05
Mar 2018 — Dec 2018
RBC
Video Producer
Prepared scripts and organized shoots and broadcasts.
- 06
Aug 2017 — Mar 2018
RTVI
Editor
Worked on news and special projects.
[ Download resume ]PDF · 2 pages
CONSULTING / 003
Where I can help.
We can look at a team process or a specific challenge in your work as an editor.
Work is piling up.
We’ll review workload, priorities and approvals, then outline process changes and clearer responsibilities for content and product teams.
You’re unsure where AI fits.
We’ll identify tasks AI can handle, where editorial judgment is needed and how to check the results. Then we’ll outline a workflow for your needs.
The bot frustrates users.
We’ll review the dialogue flow and responses, prioritize fixes and define how to assess them. We can also ask whether a bot is the right approach for the task.
A single consultation, a focused audit, several sessions or ongoing team support — depending on the task.
COMMUNICATION CHANNELS / 002
Let’s talk about work. Or just say hello.
A leadership role, consulting or a voice product — tell me what you have in mind.
WORKING TOGETHER / FAQ
Working together: FAQ
Build a team or develop an existing one?
Both interest me. At MTS, I built the content team from scratch; at VK, I developed Marusya’s existing team. I value clear goals, broad responsibility and the ability to make decisions.
What do I take responsibility for?
People, processes and content quality: hiring and team development, workload planning, approvals and the content team’s contribution to product decisions.
What can I help with as a consultant?
Improving content team operations, bringing AI into content workflows, or understanding why people struggle with a conversational product.
What could working together look like?
A single review, a focused audit, several sessions or ongoing team support. We can agree on a format and scope to suit the task.
TERM-LINK PROTOCOL / 07
Hack the dossier
Find the password among the words. After a wrong guess, the terminal shows how many letters are in the correct positions.
Likeness counts letters in the same position. Brackets offer two bonuses.
TERM-LINK / 07
Find the password among the words. After a wrong guess, the terminal shows how many letters are in the correct positions.
A note: does it need a bot?
When reviewing a conversational product, start with a question: does the bot help someone get a task done? Sometimes the responses aren’t the problem; the chosen interaction is unnecessary. That is a useful finding too.