Neurogoose
An AI assistant for presales at Intelsy: agents break an incoming project into tasks and ask the right questions, and a parametric model — not the AI — prices it by the company's own rates.
Case on intelsy.ru ↗overview
- role
- Product designer & engineer
- client
Intelsy- type
- Internal AI service
- result
- 300 → 60 hours on 10 estimates a month
At Intelsy every incoming request used to burn a week of senior architects' time, and estimates floated: each decision-maker counted a little differently, and wrong assumptions surfaced only after the work had started.
I came up with Neurogoose, designed it and built it. You describe a project in a sentence — say, a B2B marketplace with a personal account and a 1C integration — and get back a breakdown into 148+ atomic tasks, hours per role and an estimate you can defend.
Ten complex estimates a month used to take about 300 hours of senior time. With Neurogoose they take about 60.
screens
context
Presales is a classic bottleneck in IT services. Requests piled up faster than the team could estimate them, estimates were made by gut feeling, proposals were rewritten again and again for lack of structure and questions to the client, and mistakes in assumptions — integrations, non-functional requirements, roles, dependencies — came out only after the project had started.
The goal: make the first estimate fast without making it a lottery, and turn the expertise in a few people's heads into rules the whole company can use.
role
Everything: the idea, the product design, the agent pipeline and the parametric model behind it.
process
The key decision was to split the work between AI and math. Agents read the request, ask clarifying questions where the brief is thin, and break the project down into atomic tasks grouped by role. The price is never generated: a parametric model calculates it from the company's rates and rules, so the estimate cannot hallucinate.
Rates, project templates, coefficients and formulas belong to an administrator role. They are set once and changed centrally, so every estimate follows the same rules, whoever runs it.
The result can be checked task by task, approved and exported to XML, CSV or PDF for the proposal.
result
Measured at Intelsy: ten complex estimates a month went from about 300 to about 60 hours of senior time. At a senior rate of 3,500 ₽ an hour, that frees up about 840,000 ₽ of payroll a month.
Estimates became reproducible — the same request gets the same breakdown — and Intelsy kept using Neurogoose after I left.