AI implementation · Process automation · Integration

AI implementation for business: the process first, then the model

When AI implementation in a business process fails, it usually fails before anyone chooses a model: the effect was never calculated in money, or the data it needs does not exist. So we start with the process and a check of the data, and choose the technology last: AI, an algorithm, an ERP system or a mix.

For companies where manual work has become the bottleneck (processing documents and requests, support, recruitment, quality control, planning) and where there is a way to measure the result.

What is included

Process audit and choosing where to apply it

We work out where the time and money go, and choose a process whose effect can be calculated. Quite often part of the work can be handled by ordinary automation with no AI.

Data check

We check whether the data exists, what shape it is in and whether a model can learn from it. Without data there is no next step. You will learn this from the diagnostics, before the pilot and before the main spend.

A pilot with a measurable hypothesis

We frame the hypothesis so that it can be disproved: which metric should move, and by how much. The pilot runs on a real part of the process, on live data.

Integration into your working systems

AI is useful when it works inside the systems you already have: 1C, CRM, ERP, email, messengers, internal services. People soon stop opening a separate window they have to visit by hand.

On-premise deployment and security

If your data cannot leave the perimeter, we deploy on local models inside your infrastructure. Access rights and the action log are designed together with the system, from day one.

Handover to your team

We hand the system over with documentation and train your staff so they can run it without us. If only the contractor can maintain a system, what you own is a dependency rather than an asset.

How we work

  1. A conversation about the process

    A short conversation to understand the task and call things by their names: where the effect is, where the data is, where the constraints are.

  2. Rapid diagnostics

    We look at the process and the data, pick 1–2 scenarios with a calculated effect and check how realistic they are.

  3. Pilot

    A working system on a limited part of the process. The metric is fixed in advance, and we compare the result with how things were before the pilot.

  4. Production

    Integration, load, fault tolerance, running costs, access rights and audit.

  5. Growth

    We watch quality on real data and carry the approach over to neighbouring processes where the effect can also be measured.

Why us

Our methods in production

StudyNinja, the learning platform our methods run on, is in production (so far as a closed beta): a knowledge graph, telemetry, answers checked by code. We put the running cost of such systems into the calculation from day one.

Open code and published negative results

Some of our work is open: the Balansis numerical library on PyPI and the Agent Native Universe agent runtime under MIT. We publish negative results alongside positive ones.

An honest stop

If AI will not pay off for your problem, we will tell you before any work starts and suggest something cheaper.

Case studies

Aviation · aircraft components trading and MRO

Pilot

Aircraft parts procurement: from email to a managed pipeline

An AI agent reads incoming RFQs, compares supplier quotes and checks trace documents. A separate deal-tracking system moves each deal through mandatory stages.

Full loop
the AI loop ran end to end on a live procurement process, from an incoming RFQ to a team notification. This was a pilot and the figures are our own so far: the client has not confirmed them yet
3
types of trace documents checked by the agent: FAA 8130-3, EASA Form 1 and COC. Like the figure above, it still awaits the client’s confirmation

Distribution · Partner networks

In production

A sales and support platform for a distributed partner network

Partners ask who on their team is inactive and what blocks the next level, and get numbers for their own part of the network only. Access is enforced at the SQL level.

Their own slice
a partner gets figures for their own team only: a filter on the current user is added to every SQL query
6 layers
of data protection between the language model and the database, from blocked personal-data columns in SQL to rate limits; counted in the code

Marketplace e-commerce · Sales analytics

In production

Profit per SKU across three marketplaces, checked against manual figures

Five seller accounts on WB, Ozon and Yandex Market merged into one SKU-level P&L. Variance from the client’s manual figures on a control SKU for one month: 0.00%.

0.00%
variance vs the client’s manual benchmark across units, revenue, COGS, profit and margin for a control SKU over a month; the threshold was 1%
1.18M rows
of sales across 5 accounts and 12 months processed; 10.7% were ordered and settled in different months

EdTech · Mobile app

In-house product

A mobile client for an AI learning platform, built without waiting on the backend

Flutter client for an exam-prep platform: adaptive micro-lessons, a streaming AI assistant, a knowledge map. The UI was built alongside the backend via a mock layer.

13 weeks
from first commit to a working app with sign-in, goals, lessons and AI chat
41 → 81 endpoints
API coverage doubled from MVP to the extended version, with 14 → 31 modules on the same architecture

IT consulting · Lead qualification

In-house product

A pilot plan for the visitor, a qualified lead for sales

We replaced the contact form with an AI consultant: it interviews the visitor, returns a pilot plan with KPIs and risks, and sales gets a lead with budget and timeline.

1 day
to build a working MVP: frontend, backend and deploy configs
16 days
from first commit to production with HTTPS and an issued certificate on its own domain

Frequently asked questions

How much does it cost to implement AI in a company?

The cost depends on the state of the data, the number of integrations and the deployment requirements; the model matters less. Cheapest are tasks with ready data and one integration point. Most expensive is work with scattered sources, an on-premise setup and strict quality requirements. We give a price range after the rapid diagnostics, and you can work out the economics of your process beforehand in the AVA calculator.

How long does implementation take?

Implementation goes in stages, and each one starts when the condition of the previous one is met. Diagnostics ends when it is clear whether there is an effect and the data to support it. The pilot ends when the metric has been measured on a real part of the process. The scope of the full rollout is set mainly by the integrations and the deployment requirements. We give you a plan with stages and transition conditions after the diagnostics.

What if we have little data?

That is common, and it does not always get in the way: some tasks are solved by ready-made models working with your documents and rules, with no training on historical data. If the task needs history and there is none, we will say so plainly and suggest starting with data collection.

Our data must not leave the company. Is that possible?

Yes. We deploy on local open-weight models inside your perimeter, with no calls to external providers. The infrastructure costs more and performance needs more attention, but the setup works fully.

How is this different from a chatbot?

A chatbot answers questions. Implementing AI in a process means the system does the work: it parses documents, prepares a decision, passes the result to your system and leaves a trail for audit. The conversational interface is only a detail here.

How do we know the implementation worked?

By the metric we agreed on before the start: processing time, share of manual operations, cost per operation, number of errors. If the metric did not move, the hypothesis is disproved, and you learn that before the main spend.

Work it out on your own figures

The AVA calculator works out the economics of your process from your own figures, with no form and no price from us. If automation does not pay off, it says so.

Estimate the payback

Tell us what you are dealing with

Describe the problem in your own words. If it can’t be solved or won’t pay off, we’ll say so before any work starts.

In the reply we will say what we understood, what we need to clarify and whether it is worth doing.

or email us directly: hello@xteam.pro

See case studiesNew write-ups on Telegram