AI, Measured

    Your team tried AI. It didn't land.

    See it work on your own work. Then decide.

    In three weeks we run one real use case on your own past work and measure it against what your people decided. You get a build, defer or stop for the board, with a number next to it. All of it happens inside your own Microsoft 365.

    • Three weeks
    • Measured on your own work
    • Build, defer or stop
    Tell Us What Your Team TriedSee a Measured Result
    A drawing of a laptop running a team's own proposal assistant pilot. It replies that it can't find the firm's past proposals or pricing. The pilot was last edited 14 weeks ago, and a sticky note on the screen says nearly there.Proposal assistant (pilot)Draft our proposal for the new jobI can't find your past proposalsor pricing. Can you paste them in?Last edited 14 weeks agonearlythere!

    Sound Familiar?

    Someone at the top asked for AI results. Your keen people had a go. Months later, there's nothing the board can weigh.

    The pilot stalled

    An agent your team built worked on five sample documents. On your real files, with real permissions and real volume, it stopped. It's been "nearly there" ever since.

    The licences sit unused

    Copilot was bought for everyone. Most people use it to tidy up emails. Nobody uses it for anything that matters.

    The board is asking

    AI is on the agenda every meeting. There's budget set aside and nothing to show under "results so far".

    Proposals are going into public AI tools

    Without something better, people paste your proposals and prices into whatever tool is open in their browser.

    "We Already Have AI Tools."

    Good. Copilot, ChatGPT, an agent your team built: keep them. The question nobody can answer yet is how good any of it is on your own work. That's what we measure.

    A drawing of three AI tools, Copilot, ChatGPT and Agent v2, all switched on, above a box asking: tested on your own work?CopilotChatGPTAgent v2ALL SWITCHED ON?Tested on yourown work?

    Another go on your own

    • Works on a handful of sample documents
    • Nobody checks it against your real past work
    • Can't tell the board what it would take to run for real
    • Ends "nearly there", again

    AI, measured

    • Runs on about twenty of your own past proposals
    • Measured against what your people decided, and what actually happened
    • Ends in build, defer or stop, with a number next to it
    • Done in three weeks

    What Happens in Three Weeks

    One use case, on a slice of your own past work, inside your own Microsoft 365. The three weeks start once that slice is loaded.

    1. Step 1

      Choose

      Pick one use case. By default, a proposal writer or a proposal checker. You export a slice of your past proposals, pricing and wins, and we load it into a contained space inside your own Microsoft 365. Not a live connection.

    2. Step 2

      Run

      The tool checks, or drafts, each of about twenty past proposals, won and lost, using only what's in your own files.

    3. Step 3

      Measure and decide

      We compare its calls with your team's and with what happened. You get the result, the gaps in your data, and a verdict for the board.

    How we measure it, in plain words

    1. 1We take about twenty of your past proposals, won and lost.
    2. 2The tool checks (or drafts) each one.
    3. 3We compare its call with what your own people decided, and with what actually happened.
    4. 4You see where it agreed, where it didn't and why, and which gaps in your data got in the way: proposals never filed, pricing kept in someone's head, a CRM that doesn't record what won.
    A drawing of four past proposals, three stamped won and one stamped lost, feeding into a checker. The caption reads: your real work, not a demo set.Proposal 2024-07WONProposal 2024-19LOSTProposal 2025-03WONProposal 2025-11WONCheckeryour real work,not a demo set

    Your data stays in your Microsoft 365

    Your proposals and prices never leave it. We work in a contained space inside it, with a copy of a slice of your past work.

    We don't touch your live systems

    Nothing is connected to, changed in or switched off in the systems your business runs on.

    What a Measured Result Looks Like

    This is what you'd see at the end of the three weeks. The firm below is made up, and so are its numbers. Your result will be your own.

    Bilbarra Engineering: proposal checker

    140 staff. Measured against 20 past proposals, won and lost.

    Example, fictional firm

    Checker against the team

    AgreedDidn't agree

    16 of 20

    proposals where the checker made the same call as Bilbarra's own people

    The 4 misses, and why

    • Proposal 3The pricing table was pasted in as an image, so the checker couldn't read it.
    • Proposal 9Won on the relationship, not on the document. Nothing in the file says so.
    • Proposal 14The scope changed after it was submitted. The filed version isn't the one that won.
    • Proposal 19The proposal was never filed. Only the covering email was.

    Data gaps found

    • Some proposals only live in people's sent emails, not in SharePoint.
    • Rates for two services are kept in one estimator's own spreadsheet, not the pricing file.
    • The CRM records whether a bid was won, but not why.

    Verdict for the board

    BuildDeferStop
    About 7 monthsto pay for itself
    About 60 hourssaved a month

    File every proposal in one place first, then build it for real. The costed plan sets out how.

    Fictional firm and figures, to show what you get. Not a client result.

    Three Answers, All Paid For

    At the end of the three weeks you get one of three answers, each with a number next to it. We have no stake in which one it is.

    Build

    It works well enough on your own work to be worth taking further.

    The number: what it costs to take to production, and how soon it pays for itself.

    Defer

    It could work, but something in your data has to be fixed first.

    The number: what needs fixing, and what fixing it takes.

    Stop

    It isn't worth it for you right now, and you'll know exactly why.

    The number: what you won't spend building the wrong thing.

    "Stop" is the cheapest AI decision you'll make.

    It's a paid-for outcome, not a failure. You find out in three weeks, on your own work, instead of after a year of building.

    Tell us what your team tried, in a 20-minute call.

    Book a 20-Minute Call

    What You Get

    A measured result, not another demo and not a score out of five.

    A measured result on your own work

    Where the tool agreed with your people and where it didn't, proposal by proposal, with the reason for each miss.

    Build, defer or stop, with a number

    A verdict you can take to the board, with the cost and payback, or an honest reason to wait or stop.

    One use case working

    Running on a slice of your own past work, so you've seen it, not just read about it.

    The gaps in your data, named

    Which proposals were never filed, where your pricing really lives, whether your CRM knows what won, and what it takes to fix each.

    A costed plan, if it's build

    What it takes to get from the slice to production, in order, with the cost next to each step.

    Nothing leaves your Microsoft 365

    A contained space inside your own environment, loaded from an export. Your live systems are never touched.

    Is This You?

    This works best for firms that have tried AI once already and need an answer they can stand behind.

    A good fit if

    • You have 50 to 1,000 people and run on Microsoft 365
    • Someone at the top has asked for AI results
    • Your own team tried, and it didn't land
    • You have a few years of proposals on file, won and lost
    • You don't have a data team of your own

    Not a fit if

    • You already have a data or AI team, who can build and measure it themselves
    • Nobody has asked for AI results yet
    • You want a readiness score for the board pack, not a result
    • You have fewer than 50 people
    • You don't keep a written trail of proposals, or of the work you want to try AI on
    • Not you? Talk to us about our other work

    Questions We Get Asked

    Tell Us What Your Team Tried

    A 20-minute call, no slides. Tell us what you tried and where it stalled. We'll tell you honestly whether a measured result would help.

    Book a 20-Minute CallSee the Example Result