AI integration
from
$2700/ project
You see what AI is doing elsewhere and wonder what it could do for you. We start by identifying one specific task that is costly and repetitive - then we check on your own data that it works.
Book a first call45 minutes, no commitment, a few quick questions when you book.
who it's for
You're in the right place if
- Your teams spend hours on a repetitive task of reading, sorting or data entry
- You are piling up documents or messages that nobody has time to make use of
- You want to add AI to an existing product without rebuilding everything
- You have already tried a consumer AI tool and hit a wall with your real data
Another package will suit you better if
- You want to “do AI” without an identified business problem: let's start with a conversation, not a quote
- Your data doesn't exist yet or isn't usable: that needs to be addressed first
- You want to train a model from scratch: that is a different job and a different budget
the process
First conversation
Your involvement: 45 minutes, a few quick questions when you book
We look for the task that costs you the most time today. If no use case holds up, I tell you frankly - that is more useful than a project that delivers nothing.
Use case scoping
Your involvement: Access to your data and to the person who knows the business
We define what the system must produce, and above all how we will know it works: a measurable criterion, decided before we start. Without that criterion, an AI project has no way of ever being declared a success.
Proof of concept
Your involvement: You judge the results on your own cases
I build a first version on a sample of your real data, and you assess its quality. If the agreed criterion isn't met, we stop or change approach - before the budget for production has been committed.
Production rollout
Your involvement: Sign-off on edge cases
The POC becomes a reliable component of your product: error handling, behavior when the model gets it wrong, and control of API call costs. This is the step that separates a demo from a usable tool.
Handover and follow-up
Your involvement: Onboarding of your teams
I show you how to monitor answer quality and the API bill over time, and what to adjust as your data evolves.
deliverables
- The AI feature built into your existing product
- The complete source code, in your Git repository
- The measurable success criterion and the measured results
- Monitoring of model call costs
- A walkthrough for your technical teams
the budget
- The state of your data: usable as is, or in need of preparation
- The volume to process and the frequency of calls
- The required level of reliability - a tolerable error does not take the same effort as an unacceptable one
- Integration with the existing system and the number of tools to connect
what to prepare
To get started, all I need is
- One person who knows the business and can judge whether an answer is right or not
- A concrete task that costs you time today
We build the rest together
If you have none of this, that's normal and it won't hold us back: that's exactly what the scoping step is for.
- A sample of your real data, in whatever format it is in
- Examples of expected results - a dozen or so is enough
- The criterion that will define the project's success
- Your confidentiality constraints on the data being processed
- The monthly budget you are willing to spend on model calls
questions
- Why aren't API costs included?
- Because they depend on your actual usage, not on my work. They are recurring fees billed by the model provider, on your own accounts. I estimate them during scoping and design the system to keep them down, but billing them as a flat fee would mean selling you a margin on usage you control better than I do.
- My data is confidential. Where does it go?
- That is a question to settle during scoping, not afterwards. Depending on your requirements, we choose a provider that commits to not reusing your data, dedicated infrastructure, or a model hosted on your premises. Each option has a different cost, and I present them to you before you choose.
- What if the AI gets it wrong?
- It will. That is a property of these systems, not a defect to fix. The real work is deciding what happens when it does: human review before any action, flagging uncertain cases, or automatic correction. We define that behavior during the production rollout.
- What if the POC doesn't deliver?
- We stop, and that is planned for. The success criterion is set before we start precisely so that this decision is based on facts, not a debate. You will have spent the cost of a POC, not that of a full project that went off the rails.
get started
The first call is about finding out whether the project holds up. If it doesn't, I'll tell you - that's more useful to you than a quote.
Book a slot