A specific task
We choose a focused scenario and define what a good answer or processing result looks like.
06 Service
AI implementation
into business processes.
Searching company materials, drafting a reply, extracting document data or routing enquiries. We choose a task with an outcome we can verify.
When it helps
I'll put AI to work where it has a useful role.
The model is only part of the solution. It needs good sources, working rules and a clear handover to a person. Before rollout we gather test cases, check the answers and agree on limits.
- 01
Your team answers the same questions repeatedly.
- 02
Information is hard to find in company materials.
- 03
Incoming text and documents have to be manually broken down into fields.
How the work goes
Four steps
from brief to launch.
- 01
Choosing the task
We find a narrow scenario with a verifiable result: answers, search across materials, document processing.
- 02
Materials and examples
We gather knowledge sources and test examples of good and bad answers.
- 03
Prototype and testing
I build the assistant, we test it on examples and set up rules and handover to a person.
- 04
Rollout
We connect the assistant to your workflow and evaluate quality and model usage costs.
What's included
What the work
consists of.
Sources and integrations
I connect the chosen sources and build the assistant into your workflow.
Testing and human involvement
We test on examples and plan what happens when data is missing and how difficult cases reach a team member.
What you get
- Task description and limitations
- Knowledge source preparation
- Working assistant flow
- A set of test examples
- Difficult cases handed to a person
- Model usage cost estimate
We'll confirm the exact scope after discussing your project.
Related guide
Worth
reading.
3 min read
Business chatbots: what they do and how to launch one
Why a business needs a Telegram or Viber chatbot: what it handles, how to design the flows, connect it to a CRM and hand chats to a manager.
2 min readImplementing AI in business: which tasks to start with
How to start using AI in business without unnecessary risk: suitable tasks, preparing materials, checking answer quality, keeping people involved and model costs.
Related services
You can also
add these.
Example: a draft answer grounded in your materials
We use an agreed document set and real questions. The assistant drafts an answer, states when it cannot verify information and hands the question to a person. Automatic sending and data access require a separate agreement.
- Before launch we evaluate answers, limitations and model costs.
Before we start
A clear scope.
Clear terms.
What the estimate covers
Pages or workflows, content preparation, integrations and validation. The proposal records deliverables, milestones and acceptance criteria.
Separate costs
Domain registration, hosting, CRM subscriptions, external services and AI usage are separate unless explicitly included in the proposal. You retain the accounts and access.
Timing and changes
We agree a schedule after reviewing materials and access. Additional pages, features or integrations are estimated separately before implementation.
After launch
I hand over source files, access and instructions. We agree support, updates and further changes separately, including scope, communication and cost.
Need something beyond the services listed here? Let’s discuss custom features, integrations, website improvements or process automation. Describe your current setup and desired outcome. I will assess feasibility, clarify constraints and propose the work, with scope, timeline and cost agreed before development.
Project estimate
What affects
timeline and cost.
- 01Type of task and answer quality requirements
- 02Volume and format of company materials
- 03Integrations with work tools
- 04API costs of the chosen model
I give an estimate of timeline and cost after reviewing the task, before any work starts.
The details
Questions
about this service.
Can the assistant make mistakes?
Yes. That's why the project includes answer checks, rules for when information is missing and a human in the loop for important actions.
Can it answer questions using our documents?
Yes, that's a common scenario. First we need to look at the format, volume and quality of the materials, and any access restrictions.
How is model usage paid for?
The model provider usually charges for API usage. I estimate these costs on your examples and discuss them separately from development.
Does it need to handle the whole process from day one?
No. It's easier to start with drafts or suggestions for your team, then give the assistant more to do as it proves itself.