Why Anvira
Why a large organisation can choose a young company
Scale is a reasonable proxy for safety, and we are not going to argue that it is worthless. What we will argue is that the risks scale is meant to cover can be engineered out directly, and that this is cheaper and more verifiable than buying them off with a brand name.
Six commitments
What we will put in the contract
Not positioning. Each one names the mechanism that enforces it, because a commitment without a mechanism is a sentence in a proposal.
If you are not here in three years, what happens to us?
Nothing structural. The code, the database schema, the documentation and the deployment scripts are handed over as they are written, into your repository, and the system runs in your cloud under your accounts.
We build on standard technologies and on the systems of record you already run. There is no proprietary runtime and no licence to renew, so any competent team can take the work forward. That independence is deliberate.
Who will actually do the work?
The person who designs your system builds it. There is no account layer between the two, no handover from the team that sold to the team that delivers, and no one learning your process at your expense.
You will know the name before you sign, and it does not change halfway through.
Will this survive our information security review?
Zero data retention on every model endpoint. No training or fine tuning on your data. A sandboxed build and test environment, an append only audit trail of every machine and human action, and access enforced at the data layer rather than the screen.
This is not a policy we wrote for the tender. It is how we have already delivered inside a client's own cloud and artificial intelligence policy, with the model restricted to read only throughout.
How do we limit exposure before committing?
Stage gates. A short paid discovery produces the architecture, the control design and a cost model built from your numbers rather than a vendor benchmark. A sandbox proof of concept then runs on your real documents against success criteria agreed in advance.
You can stop at either gate, and you keep everything produced up to that point.
What happens when the model gets something wrong?
It is designed on the assumption that it will. Intelligence reads, extracts, matches and drafts. A person makes every write to a system of record and every payment decision, and the audit trail records who decided what.
An item that does not match is routed to an exception queue with a reason code rather than guessed at, and a refusal names the next correct action instead of simply blocking.
What do we give up by not choosing a large integrator?
The layers, and the rate card that pays for them. What you gain is senior time on your process and a shorter distance between the problem and the person fixing it.
Where a programme is genuinely beyond our capacity we will say so at the first meeting rather than staff it thinly. That is the trade, stated plainly.
Staged commitment
Three gates, and you can stop at any of them
The size of your commitment rises only as the evidence does. Nothing about this arrangement requires you to trust us in advance.
Discovery, measured in days
We walk the process as it is actually performed, count volume and variance, and identify the control points. You receive the reference architecture, the control design and a cost model built from your own numbers. If the case is weak, this is where we say so. You keep the analysis whether or not you proceed.
Sandbox proof, measured in weeks
A narrow slice runs on your real documents in an isolated environment, against success criteria written down before it starts. It either meets them or it does not, and both outcomes are useful. You keep the working sandbox and the results.
Build and rollout, measured in months
Data model first, then operations, then interface, with something demonstrable at the end of each cycle and billing tied to those milestones. Rollout runs one process and one team at a time, with the existing path still available. You keep the source from the first commit, not at the end.
Where we say no
The work we turn down
A supplier who never declines anything is telling you something about how they will handle your programme when it becomes inconvenient. These are the situations where we will recommend someone else, or nobody at all.
Saying this in public costs us some enquiries. It costs us fewer than a failed engagement would.
Automating a broken process
If the process should be removed or redesigned, automation only makes the wrong thing faster and harder to unwind. We will say so during discovery, before you have spent the build budget.
AI where rules are correct
A deterministic rules engine is cheaper, faster and auditable. Where the problem is arithmetic and tolerance rather than language and judgement, we build the boring thing.
Programmes beyond our capacity
We would rather lose the work than staff it thinly and discover the gap in month four. If the scale needs a hundred people, it needs a firm with a hundred people.
Systems we cannot hand over
If a constraint means you would be permanently dependent on us to operate what we built, the design is wrong. We change the design, or we do not take the work.
Contracting
The terms that make the answers enforceable
Everything on this page is only worth reading if it survives contact with your legal and procurement teams. These are the terms we expect to sign, and we raise them ourselves rather than waiting to be asked.
- Intellectual property in the delivered system assigned to you on payment
- Source code committed to your repository from the first week, not at handover
- Confidentiality in both directions, with your data never used to train a model
- A named individual accountable for delivery, identified before signature
- Billing tied to milestones you can inspect, not to elapsed time
- Termination for convenience at any gate, with everything produced to date retained by you
We hold no security certification yet. We will answer your questionnaire honestly, control by control, and tell you where the answer is a practice rather than a certificate.
We have no decades of trading history. The systems on the solutions page are the evidence we do have, and every figure on this site can be traced to one of them.
We are not a body shop. If your need is volume of hands rather than design and build, we are the wrong firm and will tell you so early.
We do not run a twenty four hour operations desk today. Support arrangements are agreed explicitly per engagement rather than implied.
Test the claims on this page.
Bring your information security questionnaire and your hardest process to the first meeting. You will learn more in an hour than any proposal will tell you.