On every SSB software product
Flat 30% discount on any software
School, hospital, hotel, restaurant, HR, lab, pharma and the rest of our software. Ask for a quote and we apply 30% on the price. Write to info@ssbconsultancy.com.
AI is useful when it drafts, sorts or finds something your team already stores. It is a liability when it is asked to invent a price, a diagnosis or a promise you will be held to.
The demo is fed friendly questions and answers well. On Monday a customer asks about a policy that changed last month. The tool answers from an old PDF, confidently. Staff stop using it, or worse, they do not notice.
We start from the task: sort these requests, draft this reply, find this clause. The documents are the ones you trust. A human remains on the output until you have seen the failures.
We will not put AI on a medical decision, a fee waiver or a legal promise and call it automation. Those stay with the person accountable for them.
A repeated task, with a person still checking the result
Answers come from documents you name, not from the open internet
The tool can say it does not know
Nothing customer-facing goes out unchecked at the start
The pilot is judged on a demo script instead of last month’s real requests.
The tool answers from stale or unapproved documents.
It cannot admit uncertainty, so every answer looks final.
Staff have no way to correct it, so the same mistake repeats.
A customer-facing promise is sent without a person reading it.
The pilot uses a pile of real requests you choose, including the awkward ones.
The source set is an explicit list. If it is not on the list, it is not an answer.
Low confidence is shown as “ask a person”, not as a fluent guess.
A correction is kept so the next draft can be checked against it.
Customer-facing use waits until you have accepted the error rate on the pilot.
What a person does today, how long it takes, and what a wrong answer costs.
Only approved sources. Old files are excluded on purpose.
You compare drafts with what your staff would have written.
Internal first. Customer-facing only if you still want it after the pilot.
Not at the start. The first use is a draft a person sends. Unattended replies are a later decision, after you have seen the mistakes on your own questions.
Often we do not need to. Many jobs are search plus a draft over documents you already have. If training is actually required, it is a separate scope with a statement of what data is used.
A chatbot is one possible surface. If it has no approved source, we will not install one. A widget that invents answers is worse than a contact form.
Twenty real examples of the task, and the document a good staff member would have used. If those do not exist, the task is not ready for AI.
The project starts from a paragraph and grows in chat. A custom software engagement writes the first release, the price and the milestone you can click.
The happy path works. The cancel, the duplicate and the empty field do not. QA is that list, run before customers find it.
The tool is bought, then the process is bent to fit it. Consultancy is the decision before the purchase: what the work is, and what not to buy.
Tell us the operation that is breaking. We will walk that screen and send a written quote.
We reply instantly
