Services

AI services, from one task to the whole firm.

Four connected services. Each one answers a question before the next begins, and not every firm needs all four. Every engagement is fixed-fee and scoped before it starts.

One delivery path

Start with one task. Prove it. Then roll it out.

  1. 01 · Find Half a day · $2,500 fixed

    Working session

    Pick one task your team repeats every week and build a first version together, on your own files.

    Best for: Firms that know AI could help but aren't sure where to start.

    What you get
    • A one-page blueprint of the task
    • A working first version, set up in the AI tools you already have (for example a Copilot agent, a custom GPT, a Claude project or a Gemini Gem) and tried on your own files
    • The before number the pilot will be judged on
    • An honest read on where AI helps, and where it won't
    • A fixed quote for the pilot, with the $2,500 session fee credited toward it, or our advice to stop
    Start with a working session →
  2. 02 · Prove About 4 weeks · from $10,000

    Pilot

    A small group uses it on live work, with hours measured against how the task was done before.

    Best for: Firms with a task worth testing and a few people ready to try it.

    What you get
    • Set up in your accounts, under your access rules
    • Tested against past work where the answer is known
    • Hours, turnaround and review rounds measured
    • Bad inputs, failures and monthly running costs tested and written down
    • Weekly check-ins, with problems raised as soon as we see them
    • A go/no-go report and a fixed deployment quote
  3. 03 · Deploy Scoped from the pilot

    Deployment

    Roll it out to everyone who does the task, connected to everything it needs, then hand it over.

    Best for: Firms whose pilot numbers justify a rollout.

    What you get
    • Rollout to everyone who does the task, in stages
    • Connections to the rest of your folders, templates and archives
    • Role-based training plus how-to videos
    • A named owner on your team, trained to run and adjust it
    • Admin walkthrough, written docs and a results report
  4. 04 · Run Monthly

    Care

    Keep it accurate as your tools and templates change, and take on the next task when you're ready.

    Best for: Firms without in-house AI or IT staff to look after it.

    What you get
    • A monthly check-in
    • Accuracy spot-checks on real output
    • Updates when your tools or templates change
    • The next task, scoped when you're ready
What "deployed" means

Done is when your team runs it without us.

What a finished deployment looks like at your firm.

  • Runs in your accounts

    On your real folders, templates and permissions. Not a sandbox, not our servers.

  • The whole team uses it

    Not one enthusiast with a clever prompt. Everyone who does the task does it the same way.

  • Review is built in

    The right person checks and signs off before anything leaves the firm.

  • People are trained

    Short live sessions plus how-to videos your team can rewatch.

  • It's measured

    Hours, turnaround and review rounds against the before numbers, in a one-page results report.

  • It's yours

    Instructions, connections and docs handed over, with an admin walkthrough. No lock-in.

Why it's built this way

Where AI projects stall, and what we do instead.

Many AI pilots never make it into everyday work. These are the usual reasons, and how each stage is set up to avoid them.

  • A roadmap, but nothing built

    Every stage ends with something running on your files, starting with a first version in the working session.

  • Discovery that becomes its own project

    Discovery is half a day, at a fixed $2,500, with a set list of outputs. The fee is credited to the pilot.

  • A demo that never reaches real work

    The pilot runs on live work. It isn't done until bad inputs, failures and running costs have been tested and written down.

  • Lots of activity, unclear results

    We agree on one before number in the session. The pilot and the deployment are judged against it, not on effort.

  • Nobody owns it after launch

    A named person on your team is trained to run it, with the docs and admin access. Care is there if you want it, not required.

  • A bill that grows with the hours

    Each stage is a fixed fee for an agreed result. Changes you ask for are quoted before we do them, and our own mistakes are ours to fix.

Method

Every task comes down to five questions.

We answer them with you in the working session, build to them in the pilot, and they become the spec for deployment. Here is one filled in for a common task: drafting a proposal.

You leave the working session with a page like this for your own task.

Task blueprintExample · Proposal draft
  1. 1Start
    What kicks it off, and who asks?
    A new request for proposals lands and the principal decides to pursue.
  2. 2Know
    What does it need to know?
    Past proposals, project sheets, team résumés, the request itself.
  3. 3Do
    What does it produce?
    A first draft of the proposal with the right projects and people.
  4. 4Check
    Who reviews it, against what?
    The PM checks project facts; the principal signs off.
  5. 5Deliver
    Where does it go, and who hears?
    A formatted draft in the pursuit folder, with a note to the team.

What each step looks like

Run live with your team over a video call. The order flexes to your team.

Part 1

How AI tools grew up, and where yours sit

From a single question and answer, to chat, to assistants inside your files, to agents that run on their own, shown live on a familiar task.

Part 2

Pick one task

Something your team already does every week and already tries in ChatGPT, Claude or Copilot.

Part 3

Break it into five questions

Start, Know, Do, Check and Deliver, answered in plain business terms on a one-page blueprint.

Part 4

Build a first version

Built live on your own files, so you see it work on real material.

Part 5

Map it to your tools

The same blueprint translated into Microsoft 365, Google Workspace, Claude or ChatGPT.

Part 6

Next steps

A fixed quote for the pilot, or our honest advice to stop here.

Step 1

Set up

Connect to the folders, templates and accounts the task needs, and agree on the measures.

Step 2

Build and test

Run it against past work where the right answer is known, and fix what it gets wrong.

Step 3

Live use

A small group uses it on real work, with review in place.

Step 4

Measure and decide

Compare with before. You decide go or no-go, and get a fixed quote for deployment.

Phase 1

Expand

From the pilot group to everyone who does the task, in stages that fit how your firm is organized.

Phase 2

Connect

The rest of the folders, templates, project archives and permissions it needs to work for everyone.

Phase 3

Train

Live sessions by role, plus short how-to videos people can rewatch.

Phase 4

Hand over

Admin walkthrough, written docs, and a results report against the before numbers.

Chad Kumabe
Who does the work

Chad Kumabe, co-founder. Every engagement follows Chad’s method, and he stays accountable for the result from the first call through handoff. About Second Wave →

The evidence

We build and run our own AI products, including AiToVid, a live product that turns slide decks into narrated video. See what we've shipped →

Your data and your stamp

Your files stay yours. Your professionals stay in charge.

Drawings, specs, proposals and client files are what firms worry about most. Here is how we handle them.

  • Built in your accounts

    Everything runs inside your Microsoft 365, Google Workspace or AI accounts, under the permissions you already have. We don't copy your project files to our own servers.

  • Not used to train AI

    We set it up on business-tier plans whose terms exclude your files from model training. If you're not on one yet, we tell you before we start.

  • Only the access the task needs

    We ask for the specific folders one task needs, through accounts you control. You can see and remove that access at any time, and we ask you to when the engagement ends.

  • Sensitive work stays out of scope

    Projects under strict client confidentiality terms or controlled information rules stay out of scope unless your current tools are already approved for them. We ask about this on the first call.

  • AI drafts. Your people decide.

    Every output has a named reviewer. Nothing goes out under a stamp or seal without the responsible professional’s review, and AI never signs anything.

  • Yours to keep

    Instructions, connections, docs and training videos stay in your accounts when we’re done. No lock-in.

  • Mutual NDA first

    We sign a mutual NDA before you share anything with us.

  • Insured

    We carry professional liability, general liability, and cyber & privacy liability insurance. Certificates are available on request.

Questions firms ask

Before you book.

How long until it's fully deployed?
It depends on team size and how many offices and tools are involved. We quote deployment as a fixed fee at the end of the pilot, once we've seen it work on your real material.
Do we have to switch tools or buy a new platform?
No. We deploy in what you already pay for: Microsoft 365, Google Workspace, Claude or ChatGPT. If something genuinely needs a new tool, we'll say why before you spend anything.
Where does our data go?
It stays in your accounts, under your existing access rules, on business-tier plans where your files aren't used to train models. If you're not on one yet, we'll tell you before we start. We sign a mutual NDA before you share anything. More on how we handle your data.
What does it cost?
The half-day working session is $2,500, fixed, and credited in full toward the pilot if you go ahead. Pilots start at $10,000, fixed, and are quoted at the end of the session once we know the task. Deployment and care are quoted the same way. Every step is a fixed fee, quoted in writing before it starts, so you know the total before you commit to anything. Each quote comes from what we have actually seen on your material rather than a guess.
What is the "first version", exactly?
A working assistant set up inside the AI tools you already pay for: for example a Copilot agent in Microsoft 365, a custom GPT, a Claude project or a Gemini Gem. It has the instructions and reference files for one task and is tried on your own material during the session. It lives in your account, so your team can keep using it, and the pilot hardens it for real work.
How do you measure hours without adding time tracking?
No new timesheets. In the session we pick one or two simple measures, like hours per proposal, turnaround in days or rounds of review, and record them for a handful of real examples before the pilot starts. During the pilot we record the same measures on the same kind of work and compare.
Does it work with our project and document systems?
We start with the files and tools the task already uses. If what it needs lives in a project, document or markup system, we check in the working session whether it can be reached from your AI tools, and say plainly if it can't before you pay for a pilot.
You're in Hawaiʻi. Does that matter?
No. The work happens over video calls and in your shared folders, and we schedule sessions and check-ins around your time zone.
Can we stop after the pilot?
Yes. If the numbers don't justify a rollout, you keep what was built and the results report, and we part as friends.
What if the pilot misses the target?
You'll hear about it at a weekly check-in, not at the end. The report says plainly what fell short and why, and whether we think it's fixable. You're under no obligation to continue.
What do you need from our team?
One person who knows the task well, some time each week to review output (we agree how much up front), and access to the folders it needs. You set priorities and make the go/no-go call. We manage the build, testing and fixes, and tell you early when something is blocked.
What if the scope changes partway through?
The fee covers the result we agreed on, not our hours. If you want something added, we quote it before doing it. If something we built doesn't work, fixing it is on us.
What do we own at the end?
Everything built in your accounts: instructions, connections, docs and training videos. You're not locked in to us to keep it running.

Start with one task.

Tell us what's piling up. We'll reply with an honest read and, if it fits, a date for the working session.