ANVE: the private AI office, on the record. See what that means

ANVE

Private AI glossary · updated 2026-10-07

Proof of concept
A short, small test of whether an AI idea can work at all, before a pilot.

Short answer. A proof of concept (PoC) is a short, deliberately small test of whether an idea can work at all under realistic conditions, before an organisation commits to a pilot or a purchase. For an AI tool it answers questions such as: can this model transcribe our Hinglish meetings well enough, or answer correctly from our own circulars? The UK government’s service manual describes a similar stage, alpha: build only what is needed to test your riskiest assumptions, expect to throw the code away, and usually finish in six to eight weeks. NIST’s AI Risk Management Framework asks that an AI system’s performance be demonstrated in conditions similar to where it will be used. A PoC differs from a pilot: a PoC tests feasibility, while a pilot tests the tool in real work with real users.

01

Proof of concept, pilot and rollout

StageQuestionScaleOutput
Proof of conceptCan it work for our task at all?A few people and sample documentsGo or no-go, with measured results
PilotDoes it work in daily work, and is it worth it?One team, real work, several weeksUsage, error rates, time saved and costs
RolloutHow do we run it for everyone?The whole officePolicy, training and support
02

Scoping an AI proof of concept

  • Pick one task and one test set: ten real meetings, or fifty questions with known answers from your documents.
  • Write the pass marks before testing, for example every action item traceable to the transcript, or 45 of 50 answers correct with the right source.
  • Use realistic inputs: your languages, scan quality and document types, not the vendor’s demo files.
  • Decide data handling first: anonymised or synthetic data for a cloud PoC unless a contract covers real data.
  • Record failures as carefully as successes; they decide the pilot’s scope.
03

Test the riskiest assumptions

The UK manual’s advice is to test the assumptions that would sink the project. For office AI these are usually accuracy on your own mix of languages, whether answers can be traced to sources, whether the tool fits the systems you already use, such as Microsoft Office, Google Workspace or Zoho, and whether data can stay where your rules require. A PoC that shows a demo working on clean English files has tested none of these.

04

Data and contracts

Section 8(2) of the DPDP Act lets a data fiduciary involve a data processor, for activities related to offering goods or services, only under a valid contract; that applies to a PoC that uses real personal data. A PoC on a box inside the office avoids sending documents out. With ANVE, the demo call is free, and pilots are paid, with the fee agreed in writing before installation.

05

Questions people ask

How long should an AI proof of concept take?
Weeks, not months. The UK government’s guidance puts its comparable alpha stage at six to eight weeks, and a narrow office PoC can be shorter.
Should a vendor charge for a proof of concept?
Practice varies. What matters is that the scope, the data handling and what happens afterwards are agreed in writing, whether or not money changes hands.
How is a demo different from a proof of concept?
A demo shows the vendor’s chosen examples; a PoC tests your own task with your own data against pass marks you set.
06

What is not real yet

No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.

Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.