AI in the company

AI implementation on your own processes and data.

An assistant that reads a generic web page is a demo. One that reads your offers, your tickets and your communication history is a tool somebody uses on a Tuesday morning. I start from one process, connect the model to the data that process already runs on, and write down what it must never say before anything reaches a customer.

Book a call
Network delivery illustration
AI in the company

Six applications I start implementations from.

Integracje HubSpot

Replies to enquiries

A first draft of an answer built from the company's own materials and the history of that customer, ready for a person to check and send. The measure is whether checking is faster than writing, which it is not in every process, and I say when it is not.

Finanse i rozliczenia

Ticket classification and handling

Incoming tickets sorted, routed and answered where the answer already exists in the knowledge base. Rolled out on a test channel first, watching the gaps report, before anything answers a customer directly.

Obsługa zamówień

Search across communication history

A question asked in plain language against years of notes, emails and call transcripts: who asked about this product, where a conversation stopped, why a customer left. In one company that turned a base of 8,500 records into a list worth calling back.

Operacje wewnętrzne

Reports, notes and summaries

Meeting summaries, weekly reports and record notes built from data the company already holds, with the rule about when a meeting may be recorded written down alongside.

Synchronizacja danych

A bot in the team's channel

An assistant in the tool the team already talks in, answering from internal documents. It gets adopted because nobody has to open a new window for it.

Raporty i alerty

The team working with models day to day

Company accounts instead of private ones, usage rules on one page, cost limits, and a clear line around what data must not be pasted anywhere.

See AI training
AI in the company

An implementation stands on your data, on written boundaries and on the systems around it.

Warstwa no-code

Company data as the source

The model answers from your documents, records and history, not from what it absorbed on the internet. Where that data is messy, cleaning it is part of the project rather than a surprise in week three.

Warstwa kodu

Boundaries written before launch

The list of topics the assistant must not touch gets written before it goes live. I have seen a chat answer a question about a company's client mix, and that single answer ended the project. Since then the list comes first.

Warstwa danych

Integration with the systems

An assistant that cannot write back to the CRM or the ticket queue leaves the work half done. The integration is usually the larger half of the project.

See integrations and automation
Natywnie w HubSpocie

Language and the company's voice

A model left without instructions writes like a translation engine. The company's way of writing gets captured, along with the list of things a text must never contain, and a quality gate stays on the human side.

AI in the company

Audit. Pilot. Rollout. Training and upkeep.

01Mapuj

Audit of the process

One process, measured: minutes per run, runs per month, people involved. That number decides what gets built and whether it is worth building.

02Mapuj

Pilot

The smallest version that produces a real result, on real data, used by real people for a couple of weeks. A pilot that only I use proves nothing.

03Mapuj

Rollout

Integration with the systems, monitoring, cost limits and the boundaries enforced rather than described. Anything customer-facing goes live in stages, with a person in the loop first.

04Mapuj

Training and upkeep

The team learns the finished tool on its own data, and the same number from step one gets measured again. Then the review: what stuck, what went back to old habits.

Scope of the rollout

A pilot starts with the process that eats the most hours.

The process today, the systems around it and the data the model is meant to work on. The rest I confirm on the call.

Umów rozmowę
Contact

Let me hear about the process that runs on reading and retyping today.

On the call I go through which process it is, how often it runs and which data it touches. A work email address and a topic are enough to book it.

Starter offer

An audit day or a training day to start working together.

Both have a fixed scope and take one working day. I quote a larger project only once I know what is really in it.

HubSpot8 hours

HubSpot portal audit day

  • Half an hour with leadership on the one number the company cares about
  • A read-only pass through the portal
  • Twenty minutes with each team member, up to four people
  • A document with fixes priced in hours, plus an hour to walk through it
Book a call about the auditSee the audit scope
AI in business8 hours, online

AI training day for your team

  • Six blocks in one day
  • Opening with leadership, 30 minutes
  • Hands-on work with the model on your team's materials
  • Recording and a starter kit for every role
Book a call about the trainingSee the training programme
AI in the company

Questions that come up on a first call.

How do we start with AI in the company?

With one process where the work is reading and retyping, and with the number of hours a month it takes. One team, one process, a measured result, then the next one. Training everybody at once ends with a tool nobody opens after a month.

How much does an AI implementation cost?

It follows the number of processes, how clean the data is and how much integration the application needs. There is also a running cost paid to the model provider, which is separate from my work and worth knowing before you start. I quote after the audit of the process, because before it the integration effort is unknown.

What AI solutions actually exist for companies?

In practice three groups: assistants that draft text from your own materials, classifiers that sort and route incoming work, and search across your own history. Everything else I have seen sold is usually one of those three with a different label.

What about data security?

It starts with which data the tool may see, agreed before anything is connected. In regulated industries the first question is whether the provider signs a data processing agreement. I have had to shelve a technically trivial idea for exactly that reason, and that is the right order.

Will AI replace our people?

In the applications here the model writes a first version and a person checks and sends it. What changes is the mix: less writing from scratch, more verifying. Moving to anything that reaches a customer without a person is a separate decision, taken much later.

How do we know the model will not tell a customer too much?

Because the list of what it must not discuss is written before launch, enforced in the configuration, and tested on purpose. Customer-facing rollout also starts on a test channel with a gaps report, not on the front page.

Do we need HubSpot or another CRM for this?

No. This works in companies without a CRM, and the search and summary applications often work best on an inbox and a document store. If you do run HubSpot, part of this is available natively inside the portal.

Related services

Often combined with this service.

AI training for teams

A training day on the processes and materials the team works with.

See the training programme

Integrations and automation

Connections between systems and automation of work done by hand today.

See integrations and automation
AI in the company

I start with one process and the data you already hold.

<p>Tell me which process eats the most reading and retyping in your company. After the call you have the scope, the boundaries that need writing down and the number worth measuring before and after.</p>

Umów rozmowę