Onboarding4 Oct 2026
TL;DR Xcellerate OPS now turns your knowledge base, documentation, product catalogue and resolved tickets into one permanent memory for its AI colleagues. Every resolved ticket becomes a reviewed solution. Summaries, classifications and translations are computed once and kept, and the Oversight page shows the credits you did not spend. The memory is governed: redacted before it is stored, erased together with the customer, listed in your AI Act register and hosted in the EU.
The problem: AI that starts from zero every time
Most AI in a service desk works in a similar way. It reads your knowledge base, looks for tickets that resemble the one in front of you and suggests something. That is useful, and nobody wants to give it up. But look at what happens underneath.
The same ticket is summarised when it arrives, summarised again when it is escalated and summarised once more when a manager opens it for a review. The same classification is worked out every time someone asks for it. The same translation is produced for the second, third and fourth reader. Each of those calls costs something, and you pay for it every single time, even though the content has not changed.
The second problem is quieter, but it costs more in the long run. What your technicians learned on Tuesday is gone by Friday. The fix that took an afternoon to find lives somewhere in a ticket thread, and unless someone makes the time to write a knowledge base article, the next technician starts the search all over again. The AI cannot help, because nobody ever told it what worked.
An AI colleague that forgets everything between tickets is a capable newcomer stuck on permanent day one. We wanted the opposite: a colleague that remembers.
One memory, wired into how you work
The memory in Xcellerate OPS is not a pile of text that the AI rummages through. It is structural. Categories are linked to products, playbooks, knowledge base articles and documentation, so the system knows which knowledge belongs to which kind of problem.
You do not have to maintain those links by hand. They are learned from what your technicians actually used to resolve tickets. When a playbook or an article keeps helping to close a certain type of ticket, that link is what the memory picks up. Knowledge that is never used does not clutter the picture.
Triage is deterministic first. Your routing rules decide before the AI does, so the behaviour you configured remains the behaviour you get. Where the AI does classify, its accuracy is measured against a floor instead of being taken on trust.
Both the routing engine and the AI colleagues read the same structure. That matters more than it sounds: rules and AI never work from two different versions of the truth, and every improvement to the structure benefits both at once. Read more about how an AI colleague works inside Xcellerate OPS, and how your knowledge portal is part of the same memory.
Every resolved ticket becomes a reviewed solution
When a ticket closes, Xcellerate OPS turns it into a solution record. Not a copy of the thread, but a solution with a few properties that make it trustworthy:
- Deduplicated. Ten tickets about the same printer driver do not become ten solutions. They all point to one.
- With provenance. Every solution shows where it came from, so a technician can go back to the original ticket and read the context.
- Promotable. A good solution becomes a knowledge base article in one click.
- Correctable. A team lead can correct a solution or delete it. If something was wrong, it does not stay in the memory.
That is the difference with a list of "similar tickets". Similar tickets show you what happened before and leave the judgement to you, every time again. A solution memory is curated: it grows with every resolved ticket, and your team stays in control of what it contains. Over time it becomes the written-down experience of your service desk, without anyone having to sit down and write it.
Computed once, and you can see the saving
Summaries, classifications, embeddings and translations are cached per content version, for good. If the content does not change, nothing is recomputed. If a ticket or an article is edited, only that item is processed again; everything else stays as it is.
AI usage in Xcellerate OPS runs on credits. What is computed once does not use credits a second time.
You do not have to take our word for it. The Oversight page shows the tokens and credits saved, so AI cost stops being a single line on an invoice and becomes something you can follow and explain. Your team can lean on the AI more without every repeated question being paid for again.
Governed by construction
A memory that holds your customers' tickets has to be governed from the start, not patched afterwards. In Xcellerate OPS the rules are part of how the memory is built:
- Redacted before it is stored. Nothing enters the memory before redaction.
- Erased with the customer. When you delete a customer company, what the memory holds about it is erased too.
- EU-resident embeddings. Embeddings come from a provider resident in the EU.
- Listed in your AI Act register. Your AI Act registry lists exactly what the memory holds.
- Per-field visibility. Admins decide, field by field, what the AI may see.
- Reviewable by people. The lessons the memory keeps can be reviewed by your team.
For an MSP in the EU, this is often the reason the feature is switched on at all. A memory you cannot explain to your customers is a memory you would rather not have. See how AI governance works across Xcellerate OPS.
What this means for your team on Monday
- Fewer repeated questions to your seniors. The answer a senior gave last month is in the memory, so the next technician finds it there first.
- New technicians learn from the team's history. They do not start from scratch; they start from what your team has already solved.
- AI cost you watch going down, not up. The Oversight page shows what was reused instead of recomputed.
Every ticket you close makes the next one cheaper.
