“We have an AI mandate and fifteen competing ideas.”
First stepReadiness and baseline sprint, 2–3 weeks
How the readiness sprint worksCookies on zokforce.com
We’d like to use Google Analytics to understand how visitors find and use our website, so we can improve it. It sets cookies in your browser only if you allow it. Privacy policy
We use Google Analytics to understand how visitors find and use our website, so we can improve it. It sets cookies in your browser, and you can turn it off at any time. Privacy policy
Consulting
Models are probabilistic. Your operations can’t be.
Enterprise AI rarely fails inside the model. It fails where the model meets your systems of record, your edge cases, your budget and your auditors. We engineer that boundary inside your cloud, then hand the method to your team.
What we do
Five services, combined to fit the problem.
Each is a defined piece of engineering with deliverables you can count. Most engagements combine two or three.
The model is the smallest part of the system.
A pilot needs only the model. Production needs everything around it, and that is where most enterprise AI stalls. It is also what we build.
Human sign-off
Without it: Nobody is accountable for what the model wrote.
Evaluation gates
Without it: A fix for one case quietly breaks others.
Guardrails
Without it: Security blocks the launch, and they’re right to.
The model
A frontier API, an open-weights model, or a small model trained on your data, chosen by evaluation results. It’s the part every demo shows.
Cost and trace telemetry
Without it: The bill grows and nobody can say why.
Connectors
Without it: Staff copy and paste between the AI and the real system.
Your systems of record
- ERP
- EMR
- Core banking
- CRM
- Document stores
We connect to them. We don’t replace them.
Where to start
Start from the problem you have.
Most teams come to us with one of three problems. Each has a short, fixed-scope first step. You don’t need to know which of our services you need.
“The pilot works in the demo. It can’t go live.”
First stepThin-slice sprint on one workflow, 3 weeks
How we connect existing systems“We bought AI coding tools. Releases didn’t speed up.”
First stepADLC sprint in a flagship repository
How ADLC worksHow we work
Something you can use within weeks.
Each stage is short, leaves you with something your team can use on its own, and ends with your decision. Durations are typical, not fixed. No unpaid proofs of concept and no multi-year commitment.
Diagnose and baseline
We work inside your environment, map the systems and data the workflow touches, and measure what it costs today.
You leave with
- A ranked shortlist of workflows, scored on value and feasibility
- A target architecture and a security and privacy evidence pack
- A signed baseline of today’s cost, cycle time and error rate
Then you decide: build, or stop with a plan you own.
Build into production
We build the workflow end to end on the agreed scope and hand over code, tests and runbooks your team can maintain.
You leave with
- Connectors to your systems of record, deployed in your cloud
- Evaluation gates, guardrails and cost tracing around the model
- Human sign-off on every write to a system of record
Then you decide: run it yourselves, or keep us on.
Operate and improve
Providers change model behaviour and retire versions. We keep accuracy up and cost down after launch, or coach your engineers to do it.
You leave with
- New models tested against your evaluation set before any switch
- Reviewer corrections turned into new test cases
- A monthly cost, drift and audit report
It ends when your team is ready to run it alone.
Four things in every contract.
We work in your cloud
Code, evaluation sets and models run in your environment, on your keys and your model-provider accounts, under your access controls.
You own what we build
Connectors, prompts, evaluation sets and fine-tuned model weights are yours. Our reusable templates come with a perpetual licence.
A person signs off
Models draft. A named person approves every change to a system of record, and AI never merges code on its own.
No lock-in
No required licence and no proprietary runtime. If one of our own tools helps, it’s optional and replaceable.
How we charge
A smaller fee up front. The rest only if it works.
The base fee covers the cost of the work. What we earn on top is a share of the value we count together after go-live, for an agreed period and up to a cap. So we stay with it after launch: part of our fee depends on how it runs.
A typical fixed-price project
The whole fee is paid before anything runs in production.
Base fee plus a value share
Less to pay before go-live. The rest comes from results we count together after it.
An illustration of when fees are paid, not of amounts. Terms are set in each statement of work.
Baseline before build
Unit costs and the value formula are agreed at the end of Stage 1, before any build starts.
Counted, not estimated
Value is counted from telemetry in your environment, and only on outcomes a person signed off.
Capped and time-limited
Every value-share contract has a written ceiling and an end date, so finance knows the range before signing.
Need a fixed price?
Every stage can be a fixed price, paid against milestones you accept. Most public-sector and enterprise work is bought this way.
Who you work with
You work with the people who build it.
We have worked on many AI projects for clients, from first pilots to systems running in production. We also build and run AI products of our own, among them MedScribe AI for clinical documentation and ZokLens for the cost, quality and security of LLM applications, using the same method we bring to your team.
Your project is staffed by engineers with real experience building and operating production AI systems, not by a separate advisory practice.
More about usWhat leadership teams ask first.
Do we need an in-house AI team?
No. Many clients don’t have one. We build the system and coach the engineers you do have, so they can run and change it after we leave.
Which models and clouds do you work with?
The ones you already use: frontier APIs and open-weights models, through your own AWS, Azure or Google Cloud accounts or private endpoints. We pick models by how they score on your evaluation set.
How is our data protected?
The work runs inside your environment, on your keys and your model accounts. Sensitive data is masked before it reaches an external model, and every call is logged for audit. We don’t ask you to send data to us.
Do we have to buy your products?
No. ZokLens and MedScribe AI are separate products. If one would help your project we’ll say so, but nothing we build for you depends on them.
Can you work through public-sector procurement?
Yes. Public-sector work uses fixed-price, milestone-gated stages, and our readiness work produces the technical evidence an algorithmic or privacy impact assessment needs. How to buy through government
Tell us where it’s stuck.
A 30-minute call about where your AI work is stuck. If we’re not the right team for it, we’ll tell you.
Hi, I’m Zoey, the AI assistant at Zokforce. Ask me about our AI services, our products, or how AI could work in your business.Writing an answer…YouZoeyYou’re sending messages quickly. Wait a minute, then try again.Zoey can’t answer right now. Please try again shortly, or email service@zokforce.com.