Questions we get asked

What does Metis actually build?

We are an AI product studio and consultancy. We help teams work out where AI pays off, then design, build and ship it — AI agents, workflow automations, internal tools and full AI products, from a first prototype to production software.

We are not a staffing firm and we are not a reseller for someone else's platform. Every build engagement ends with software that runs in production and a person on our side responsible for it continuing to run.

Are you an AI consultancy or a development shop?

Both, on purpose. The consultancy side works out what is worth building; the studio side builds it and stays responsible for the outcome.

A recommendation with nobody accountable for building it rarely survives contact with production, and a build with no strategy behind it tends to solve the wrong problem. We do both so neither happens — which also means we say no to projects we would not be able to operate.

What industries do you work with?

Finance and insurance are where we do the most work. We also build internal software for technology companies and lead systems for real-estate tech.

The common thread is operational: work that is real, repeated and expensive to do well by hand — the kind of problem where AI earns its keep rather than decorating a slide.

How big is the team?

Small and deliberately so. You will work directly with the people building your system.

There is no account layer between you and the engineering. For the size of engagement we take on, that is a feature — questions get answered by whoever wrote the code.

How do engagements usually start?

With something small and fixed: a Discovery Sprint to decide what to build, or a scoped pilot with a fixed fee, a fixed timeframe and a number we agree in advance that tells us whether it worked.

We prefer this to an open-ended retainer because it puts the risk in the right place. If the pilot does not hit the number, you have spent a defined amount and learned something concrete rather than committing to a year.

What does it cost?

Discovery Sprints and pilots are quoted as a fixed fee. Larger builds run either as a fixed-cost project or as an embedded team on a monthly rate.

We do not publish a single number for builds because the honest answer depends on scope — how many of your systems we integrate with, how much data work is needed, and what has to be true for it to run in production. A scoping call takes about twenty minutes and produces a real number.

How long before something is live?

We usually put a working prototype in front of you within two to four weeks.

The engineering is rarely the constraint. Access to data and systems, and decisions on your side, are what set the pace.

What do you need from us to start?

A clear problem, access to the data and systems involved, and one person on your side who can make decisions.

The last one matters most. Projects stall on unanswered questions far more often than on technical problems.

Do we own what you build?

Yes. Custom work we build for you is yours, set out in the engagement terms before we start.

Where a system depends on third-party services — model providers, cloud platforms, messaging — those remain under their own terms, and we will tell you which ones you are taking on before you commit.

Can you prototype before we commit to a full build?

Yes — that is how we prefer to work. We build the riskiest part first, so you see whether the idea holds before paying for the rest.

A prototype that answers the hard question is worth more than a polished demo of the easy parts.

Do you build production systems, or just prototypes?

Production systems. Prototypes are how we start, not where we stop.

That means the parts demos skip: access controls, monitoring, evaluation of the AI's output, and a handover your team can actually run. Compliance, security and observability are designed in from the start rather than added after launch.

Which AI models do you use?

Whichever fits the job. We are not tied to one provider.

We design systems so the model behind them can be swapped — for cost, quality or regulatory reasons — without rebuilding everything around it. Our note on DPDP and AI explains why that matters for Indian customer data.

Will people know they are dealing with AI?

Yes. We do not build systems that pretend to be human.

People are largely fine with software that is quick, useful and honest about what it is. They object to being deceived, not to being helped by AI.

What happens after launch?

We can keep running, monitoring and improving what we built, or hand it over and train your team to run it — your choice.

Either way, we do not hand over a demo and disappear. The engagement is not finished until someone on your side can operate the system without us.

Where does our customer data go?

Into the systems required to run what we build for you, under a contract that covers processing on your behalf — and nowhere else.

We will tell you exactly which third parties are in the path before you commit, what each one receives, and how long it is kept. If that list changes, you hear about it before it changes.

Can Indian customer data be sent to an AI model hosted abroad?

Currently yes. India's DPDP framework permits transfer to any country the government has not specifically restricted, and no country has been restricted to date.

This is the opposite of the European position and it surprises most people. The obligations that do apply are consent, notice, retention and breach reporting. We set out the detail, including why you should still keep your model provider swappable, in our note on DPDP and AI.

What happens to the data our AI systems collect?

Where it identifies people it is personal data, so it gets a defined retention period tied to the purpose it was collected for, and it is deleted at the end of it.

Logs, transcripts and uploaded documents are the artefacts teams accumulate silently and never schedule for deletion. We set the schedule at the start of the engagement rather than after a breach makes it urgent.

Do you use our data to train models?

No. Your data is used to run your system.

We also check the terms of the providers in the path on your behalf, and confirm the same holds there.

Do you work with regulated businesses?

Yes — finance and insurance are where we do the most work, and the rules shape the design from day one rather than being checked at the end.

That covers data protection under DPDP and sector rules that reach into how software contacts customers. For insurers, for example, service and transactional calls had to move to TRAI's 1600 number series by 15 February 2026; we set out the detail in our note on the 140 and 1600 series.

Where are you based?

India. We work with clients across the country and remotely.

Do you work with companies outside India?

Yes, though our deepest advantage is in the Indian market.

The regulatory knowledge that makes our Indian work good does not automatically transfer to another market. We will say so if a project would be better served by someone local to it.

What is the smallest project you take on?

A Discovery Sprint or a scoped pilot. Below that, an honest conversation is usually worth more to you than an engagement.

If what you need is advice rather than a build, we will give it in the first call and not invoice for it.

How do we get in touch?

Through the contact form. It reaches us directly and we reply within a working day.