Questions and answers
What companies ask on the first call. If your question is missing, write to me: I will answer and add it here.
Tools and data
Which tool will you use?
The one that fits the task and what you already have. If you pay for Microsoft 365 Copilot, I start there. If a model through the API is needed, usually Claude or OpenAI. If everything happens in Salesforce, Agentforce and Prompt Builder. You get a proposal with pros and cons after the first call.
Will our data be used to train models?
Under the vendors’ terms, data sent through the API and on business plans is not used for training. I check those settings during implementation and record them in the documentation. Personal data is anonymised or stays in your systems. Staff’s personal accounts are another matter, and the first thing to tidy up.
We already bought Copilot. Does that not cover it?
For summaries and drafts in Outlook and Word it often does, and then we start there. It does not cover connecting to your CRM and databases, reports at a fixed time, an approval queue or chat on your website. I check what Copilot already does for you before proposing anything alongside it.
Where is our data while AI works?
In your systems. What goes to the model is the fragment the task needs, not the whole database, through a business account with training switched off. The log lives with you. I do not copy data to my machines beyond what the prototype needs, and I delete that when it is done.
What about GDPR?
Anonymisation or data staying in your systems, business accounts with the vendors, a log, a data processing agreement where required, and a list of systems and vendors for your records of processing. You agree the documents with your lawyer; I make sure the technology does what they say, and I supply the list that is needed for them.
Mistakes and control
What if AI gets it wrong?
That is why actions with consequences (sending, changing, deleting, paying) need a person to confirm, and every result leads to its source. A mistake should be noticed before anything comes of it. How that works in practice is in the article on the human in the loop.
Can AI make things up?
It can, so it does not get tasks where making things up would go undetected. A query counts the figures, answers cite a source, and when there is no source the assistant is supposed to say “I do not know”. I test that on the prototype with questions the data cannot answer.
Who is responsible when something goes wrong?
I am responsible for the mechanism: permissions, the log, the approval queue, the tests. The company is responsible for decisions a person approved, as with any tool. The log lets you establish what happened without guessing.
Will AI have access to everything?
No. A separate account in each system, read-only where that is enough, sensitive fields excluded, with an expiry date. AI sees only what that account sees. That is the rule every implementation starts from.
Working together and costs
Do we need our own IT person?
No. I need someone who knows the task and can judge the results: the sales manager for a report, someone from support for the inbox. I handle the technology and leave the documentation behind.
What does it cost?
A fixed price for the prototype, agreed after a call. Then by scope: a job, a block of hours or ongoing care, billed for time worked. Plus the AI tool costs, which I show up front and separately. No price list on the site, because the tasks vary too much.
Does a small company even need this?
If one task takes someone more than two hours a week and repeats endlessly, usually yes. If not, I will say so on the first call and will not push.
When will we see the first result?
A prototype within 1–2 weeks of getting access to the data. A working version on your own examples, not a presentation. Full go-live usually within the following two weeks, then a month of observation.
We are on Salesforce. Does that change anything?
Yes, for the better: Salesforce has its own AI tools and MCP server, and I have worked on the platform since 2020. Start with the free permissions audit, then have a look at the site about AI in Salesforce.
We would rather our team learnt this themselves.
That is a good route for tasks everyone does in their own chat window: emails, notes, analysis. I run AI training for teams. An implementation is needed where AI has to run on its own, at a fixed time, on data from your systems.
Is there a task someone at your company does by hand every week? Describe it in one sentence.
I reply within one business day. The first call is free of obligation.