Skip to how we do it
Armakuni Contact us

Make your data AI-ready.

In a business of a hundred or two hundred people, teams already ask AI agents for numbers and get confident answers that disagree. We give every AI agent one agreed meaning for each number and the exact place it lives, so two AI agents asked the same question give the same number, with its working.

Book a free workshop

One day with the engineers who’d do the work. AWS funding typically covers the workshop.

Part of Guaranteed AI Outcomes

We built the analytics that give an executive search firm’s CEO and board their own answers on revenue, margin and pipeline. An AWS Premier Tier partner, one of about 300 among more than 130,000.

Every word on one signpost, with one road to where it lives.

An example business.

Every team’s revenue is right.
That’s the problem.

Sales counts revenue when a deal closes, finance when the invoice goes out, and operations when the order ships. Each of them is right, and everyone knows which one they mean. An AI agent doesn’t, and it won’t ask. It adds up whichever table it finds first, answers without a flicker of doubt, and the number ends up in a board deck.

  • Three different numbers share one name. Nobody’s is wrong, so nobody checks which one the AI agent took.
  • When someone renames a field in the CRM, the AI agent carries on without a sound and reports a pipeline of $0.
  • These numbers are an example. The workshop finds your own.

September’s revenueToday

numbers called revenue

What each team means by it

Example

Your people agree what a number means.
AI agents do the adding up.

Each answer is worked out from the meaning your people agreed, and it comes with where it came from. They own every meaning and hear first when one changes. The AI agents fetch and add up, so the hours someone spent pulling reports go on the questions only a person can answer. Here are six examples from six kinds of business.

A dental group, three clinics

How many new patients did we see in September?
184new patients
Means
A first visit attended, at any of the three clinics
Agreed by
The practice manager
From
The booking system

New patients, counted one way

The front desk counted bookings and billing counted first claims. Now both read one meaning, and the numbers match.

An online shop

How many active customers do we have?
12,480customers
Means
Bought in the last 12 months, with refunded orders left out
Agreed by
The head of marketing
From
Orders and returns

Active customers, one list

Marketing, support and finance send to the same customers, and nobody is counted twice.

A wholesaler

What’s our margin on olive oil this quarter?
18.4%margin
Means
Sale price less landed cost, with freight and duty in
Agreed by
The finance director
From
The accounts and purchase orders

Margin with the freight in

Each sale is matched to its purchase order on the agreed key, so no line goes missing from the sum.

A property manager

How many of our units are empty?
23of 410 units
Means
No signed lease on the 1st; units under notice counted apart
Agreed by
The head of lettings
From
The leasing system

Vacancy, as the owners count it

Units under notice show up on their own line, so the owners’ report and the lettings team agree.

A home services company

Which jobs lost money last month?
7of 212 jobs
Means
Hours at the loaded rate, plus parts, came to more than the invoice
Agreed by
The operations director
From
Timesheets, parts and invoices

Job cost across three systems

Timesheets, parts and invoices are joined the way your team agreed, and each of the seven comes with its working.

An accounting firm

Which of my clients are behind on their filings?
4of your 38 clients
Means
A deadline within 14 days and a document still missing
Agreed by
The managing partner
From
The practice system and the document portal

Only the clients you may see

Each accountant’s AI agent asks as that accountant, so it answers for their own clients and no one else’s.

An AI agent will always find a number.
The work is making it the one finance signs.

Connecting an AI agent to your tables is quick. Trusting what it tells you about them is engineering, and each decision below stops a wrong number before anyone acts on it. Our forward deployed engineers make them with your team, working inside your business until it’s live. They bring the layer we’ve already built to give every AI agent one meaning of your data. It’s called Axiom, and it comes with every Guaranteed AI Outcomes engagement.

  1. 1

    One name for each number, agreed by its owner.

    Your data owners agree each definition: revenue, a customer, margin, a new patient. We write each one down once, with its calculation attached, in one catalogue that every AI agent reads.

    • Where two teams count differently, both numbers keep their own names, such as bookings and revenue. Your decision owner settles which one the board hears.
    • A definition changes only when its owner changes it, as a new version that keeps the old one on record.

    WhyOne question gets one answer, not three.

  2. 2

    Every word mapped to exactly where it lives.

    We map every business term to its exact tables, columns and join keys, so an answer is worked out from the agreed meaning instead of guessed. AI agents read your data through that layer, never straight from the tables.

    • Reports and dashboards can read the same layer, so a number means the same thing wherever it shows up.

    WhyA join the model guesses can drop records without anyone noticing.

  3. 3

    It asks as the person it works for.

    Through Amazon Bedrock AgentCore Identity, each AI agent queries as the person asking, so the row and column rules in your data store still apply. A sales rep’s AI agent sees totals, not salaries.

    • The rule lives in the data layer, because a prompt enforces nothing.
    • All of it runs in your own AWS account.

    WhyPayroll doesn’t go to whoever asks nicely.

  4. 4

    Every number comes with its receipt.

    Every answer is traced back to its query, with the definition and its version, the sources it read and who asked. Anyone can open a number and see its working.

    WhyA doubt takes a minute instead of a meeting, and a figure nobody can check stays out of the board pack.

  5. 5

    A change in a system is caught before it answers.

    When a field is renamed or a table is split, we catch the change before any AI agent reads through it. The answers that depend on it pause, and the owner of each definition hears what changed.

    WhyNobody finds out on Friday that the pipeline has read $0 since Tuesday.

  6. 6

    Tested on questions you already know the answers to.

    Your team picks real questions with checked answers, such as last quarter’s board pack, and each one becomes a test.

    • The tests run again on every change to a definition or a source, and a pass has to hold on every run.
    • When someone corrects an answer, the correction becomes a test it has to pass from then on.

    WhyFixing one number doesn’t quietly break another.

Revenue, SeptemberTue 9:41

$368,950

Answered by the AI agent for Sam, the owner

Example

What a client said

NHS Wales needed data access measured in minutes, not days. Armakuni built the platform and transferred every piece of knowledge to our team. When they left, we ran everything.

Director of Technology, NHS Wales

One day to score your data.
About 90 days to go live.

We start with Data readiness for AI, one day with three to six of your people: whoever decides, the finance lead who owns what revenue means, someone from each team that uses the number, and whoever keeps your systems. They bring the questions they ask most, with answers you already know.

  • In the morning we map what you run today. You watch an AI agent answer “What was September’s revenue?” for an example business, with its receipt. Then each team’s figure for the same word goes on the wall, with where it lives and who owns its meaning. If the figures already agree and live in one place, we’ll say so, and an AI agent can be asked about them today. We go through the systems behind them: how they join, who may see what and how often they change.
  • In the afternoon we ask your own questions, before and after. We put the questions whose answers you already know to a sample of your data, or a look-alike, and score the answers. Then we write down the meanings agreed in the morning and ask the same questions again. With you, we score seven readiness questions for sales, finance and operations, from 0 (nobody knows) to 3 (enforced and tested): meaning, source, quality, access, change, proof, and history and documents.
  • Then we design the target: the first area to make ready, its entities and how they relate, where the meaning layer sits over your systems and which keys join them, and how each person’s access is enforced.
  • You leave with three things, yours to act on with us or without us:
    • A findings report, with the verdict for each area (ask now, ready in weeks or data work first) and your questions scored before and after.
    • A target architecture: the meaning layer over your systems, its joins and its access rules, with the words that matter, their owners and where each one lives.
    • A costed roadmap: the first area, its known questions (last quarter’s board pack, say) as the first tests, the week 4 measure, what it costs and what to fix yourselves first.
  • Then there’s a go-or-stop decision in week 4, on a scorecard you sign. If it’s stop, you keep the map, the baseline and the scorecard. If it’s go, we’re paid on the result from then on.

What we’ll tell you not to do

  • Don’t point an AI agent at an area that scores data work first. Where one customer is five records, or the history isn’t kept, that’s a data project of its own, and we’ll say so before any AI agent goes near it.
  • Keep the report if one report already answers the question. Not every question needs an AI agent, and a dashboard your team trusts costs less.
  • Do it yourselves if a written definition and a renamed column would do. You leave with the list, and we’ll tell you it isn’t worth paying us for yet.

Board-pack questions answered rightLive at week 13

known answers,
at week 13

Example

We’ve untangled data like this before.

GORUCK

A rugged outdoor and fitness brand, selling direct and running events nationwide

Before
Nine platforms that didn’t talk to each other. The same person could exist as five disconnected records, and analyst-built reports took days.
What we built
A customer data platform that unifies all nine sources. We built it and maintain it: the warehouse, the pipelines, the analytics layer, and segments sent automatically to email and ad platforms.
  • A single customer view across nine platforms
  • 40% higher ROI on targeted campaigns
  • Customer insights 30% faster for marketing, eCommerce and event teams

An executive search firm

Global C-suite and board search

Before
A manual platform, locked to a vendor that shut down in December 2025, and analytics in pieces.
What we built
Six pieces of work, opened by a discovery fully funded by AWS. One was an analytics build on Amazon QuickSight and Amazon Q.
  • The CEO and board get self-service answers on revenue, margin and pipeline
  • Zero disruption through the vendor’s shutdown
  • All business data stays inside AWS

Checked by AWS. An AWS Premier Tier Services Partner, validated by AWS for agentic AI consulting, with more than 500 customer launches on AWS and 1,300+ engineers. Of the proofs of concept we scope, 70% reach production.

How Guaranteed AI Outcomes works Read the case studies

Reasons to wait,
answered.

Owners give us good reasons to wait, starting with “our data isn’t ready yet.” Here are five, and what we tell them. Got another? Bring it to the workshop.

Our data is a mess. We’ll clean it up first.

Every business’s data is messy, and cleaning all of it before anything else is how these projects stall. We start with the handful of words your decisions turn on, such as revenue, a customer and margin. We map those and test them against answers your team already trusts.

Cleaning happens where a test fails, one source at a time. The workshop tells you which areas an AI agent can be asked about now, and which need data work first.

We’d need a data warehouse first.

Usually not. The meaning layer sits over the systems you already have and reads the data where it lives.

Where a source can’t be read safely or quickly enough, we say so at the workshop. Sometimes the answer is a small, clean copy of the few tables that matter.

There’s no budget for it this year.

AWS funding typically covers the workshop, and AWS has funded part of the work itself for several of our clients. At the workshop we’ll tell you whether yours could qualify.

After that you decide at week 4, on a scorecard you sign. If it’s go, we’re paid on the result from then on.

Our own team can handle it.

Some can, and we build it with them, inside your business, so they run it afterwards. We bring engineers who’ve done it before, and Axiom, the meaning layer we’ve already built, so your team doesn’t start from a blank page.

If you have no data team, we build it with the people who keep your systems. Then they run it, or we do.

We’re replacing a system next year. We’ll wait.

The meanings outlive the system. Revenue means the same thing in the new one.

We catch a change before an AI agent reads through it, and the answers that depend on it pause until their owners have seen what changed. A system you replace is mapped again, and the known answers prove the numbers still match before anyone relies on them.

Bring the number your teams argue about.

We start with Data readiness for AI, one day with the engineers who’d do the work. AWS funding typically covers the workshop. You leave with a findings report, a target architecture and a costed roadmap: which words need one meaning first, where each one lives, and which questions an AI agent can answer from your data today. If it isn’t worth doing, we’ll say so plainly.

Book a free workshop