How it works

Unlock your organizational intelligence.

Ferrata Labs gets a first AI agent live in under 90 days and leaves your team able to build the next one. Four stages, one engagement: we assess where AI pays, anchor the Brain on your own data behind your firewall, put one agent live on real transactions, then make your team AI native. Every consequential action starts behind an approval gate, and autonomy is earned with evidence.

01Assess2 weeks
02Anchor3 weeks
03First agentUnder 90 days
04AI nativeOngoing

Why via ferrata: a fixed steel line bolted into rock so people who are not technical climbers can cross terrain that would stop them. The line goes in before anyone climbs. Foundation first, agents second, your team third.

The engagement in four stages

2 weeks · you keep the output either way

Assess

Where AI pays, and where it doesn't.

What we do

We map workflows, not org charts: transaction volumes, exception rates, where people re-key between systems, and what errors cost when they escape.

In parallel we audit your environment: cloud posture, data residency, and what your security team will allow.

Then we rank every candidate workflow by the money it moves and how ready it is, and choose the first one with you.

How to find AI business use cases in your own workflows →

What you get

Ranked workflows with a cost attached to each
A baseline for the number the first agent has to move
A target architecture for your environment
A model plan: what runs on private open-weight models inside your firewall

3 weeks · the asset you keep

Anchor: set up the Brain

Your ground truth, in the Brain, behind your firewall.

What we do

We bring your ground truth into the Brain: your ERP, documents, email and portals, connected and searchable for agents, with every answer traced to its source.

It runs inside your environment, behind your firewall. A model gateway sends every task that touches your data to open-weight models served there, and we fine-tune them on your workflows so the output fits how your business works.

Frontier models are used only for work that carries none of your data, through the same gateway, under a written policy. Your information never leaves your environment and never reaches a frontier lab.

Evaluation, cost controls, audit logging and identity are built in from day one. It belongs to you: the difference between renting agents and owning the capability to run them.

What you get

The Brain: retrieval over your systems of record and documents
A model gateway, with open-weight models served in your environment
Models fine-tuned on your data, inside your firewall
Evaluation, cost controls and audit logging
Identity and access wired to your directory

Under 90 days from start

First agent live

Gated first. Autonomous once it has earned it.

What we do

One agent, one workflow, live on real transactions, configured to your rules, your pricing logic, your approval thresholds and your escalation paths.

Phase one is gated. Every consequential action waits for a person, every decision is visible, and every approval and override teaches the agent how your business decides.

Phase two is earned. When the record shows the agent gets a class of action right, and you are comfortable, that gate comes off. The agent executes on its own and exceptions still go to people. You decide, one class of action at a time.

Hypothetical example (illustrative only): an accounts payable agent matches a vendor invoice to the PO and receipt and queues the posting. In phase one a person approves every posting. In phase two clean three-way matches post on their own, and mismatches still come to a person.

What you get

One agent in production on your own numbers
Nothing releases cash, commits spend or contacts a customer without approval until you remove that gate
Every action logged and attributable, with a rollback path defined before go-live
An accuracy record for each class of action, so gates come off on evidence

Ongoing, or until you don't need us

Your team, AI native

Your team builds the next one.

What we do

Our AI Velocity POD is two forward-deployed engineers who learn your business inside out. They sit with your business teams, work hand in hand with your IT team, and run the whole engagement end to end, from assessment to agents in production, with Ferrata's architecture review and validation behind every release.

More agents follow as trust builds. All along your people learn the platform, the ADLC and the runbooks by building with us, so by the end your own team builds the next agent.

The engagement is designed to end. Keep us on for managed operation if you would rather; you never have to.

What you get

More agents as trust builds
Your people and IT team trained on the platform and the ADLC
Runbooks and an evaluation harness your team can extend
Managed operation if you would rather not

What changes, and where.

Transformation is not a model licence. It happens in three places at once, and skipping any one of them is why most of it stalls.

Layer 01

People

Our AI Velocity POD, two forward-deployed engineers, works inside your team on your systems and your rules to unlock the organizational intelligence you already have. Your people learn the framework while it is being built. By the end they are building the next agent on it themselves.

Layer 02

Product

Agents wired into the systems you already run, acting on live transactions. Nothing is re-platformed: the agents are infused into the products your team already works in. The approvals they stop at happen on the platform we stand up with you.

Layer 03

Process

Agentic workflow that runs on the rules your business already follows: your approval thresholds, your escalation paths, your guardrails. Encoded, audited and governed rather than reinvented, so the agents work on top of the framework you have.

ADLC

One foundation, then the loop.

ADLC, the Agent Development Lifecycle, is the framework every Ferrata Labs engagement runs on. Anchor lays the foundation once: your Brain, set up behind your firewall. Then every agent goes round the same seven-phase loop, from Scope to earned autonomy, and the next agent starts on the same foundation.

The four stages are how we take you through it. Assess decides which agents go round the loop and in what order. Anchor lays the foundation. Your first agent is the first trip round the loop, gated until it earns its way out. Stage four is your own team running the loop.

ADLC is the practice. Our AI Velocity POD, two forward-deployed engineers inside your business, is who runs it, with your team alongside.

SDLC, built for codeADLC, built for agents

Code does the same thing every timeAgents decide, so they are scored on your real cases

Tests pass or fail before releaseEvaluated before go-live and watched after it

Ship it, then patch itGated first, autonomy earned on evidence

Bugs get fixedEvery override becomes the improvement signal

See the full lifecycle

The harnesses, runbooks and gate rules the loop produces are documented and handed to your team in stage four. The lifecycle is the transferable part; the agents are what it produces.

Questions we get asked early.

The ones that come up in nearly every first conversation, answered the way we’d answer them on the call.

Where does our data actually go?

Nowhere outside your environment. The platform stands inside your firewall, retrieval runs over your own systems of record, and every model that touches your data is an open-weight model served there. Frontier models only ever see work that carries none of your data, and that line is written into policy in Assess before anything is built.

Do you train on our data?

Yes, for you and only for you. Where it makes a workflow better, we fine-tune open-weight models on your data so the agent learns your vendors, your codes and your exceptions. That training happens inside your environment, behind your firewall, and the model it produces is yours. Your data never leaves your environment, never reaches a frontier AI lab, and is never used for anyone else's work.

What happens if an agent gets it wrong?

It gets caught at the approval gate, because agents start behind one and only earn their way out with evidence. Every action is logged and attributable, and a rollback path is defined before go-live rather than after an incident.

Frontier models or open-weight?

Both, with a firm line between them. Anything that touches your data runs on open-weight models inside your environment, tuned on your workflows. Frontier models are used only for work that carries none of your data, such as reasoning over public material or drafting templates. Which is which is decided per workflow in Assess and written into the gateway's policy.

Are we locked into you?

No. The platform runs in your environment and the runbooks are yours. Stage four exists so your own team builds the next agents: our AI Velocity POD trains your people and your IT team by building alongside them. We would rather you outgrow the engagement than depend on it.

How long until the first agent is live?

Under 90 days from the start of the engagement: Assess for 2 weeks, Anchor for 3 weeks to set up the Brain, then one agent on one workflow against real transactions with an approval gate on every consequential action that moves cash, spend or customers.

What does this cost?

It depends on the workflow and how much of your environment already exists, which is why Assess comes first and why you keep its output either way. We give a range on the first call rather than making you sit through a process to find out.

Bring us one workflow.

Thirty minutes, no deck. Pick the process that costs you most in people, errors or delay, and we'll pressure-test whether an agent is genuinely the right answer for it.

An honest read on whether that workflow is a good agent candidate
What the first deployment would involve, and roughly what it costs
An honest answer if the sequencing is wrong and you should fix something else first

Book a discovery call