Four services, run in order. Identify and Report is a complete engagement on its own, and you own the plan whether or not you ever call us again. Build and Shift turn that plan into working infrastructure. Here is what each one asks of you, and what lands at the end of it.
In 1865 the economist William Stanley Jevons noticed something that did not fit. Steam engines had become dramatically more efficient, so England should have been burning less coal. It was burning far more. Efficiency had made coal cheaper to use, and cheaper use created uses nobody had bothered with before. Demand did not fall with the price. It chased it.
That is the question AI puts to every task in your business. When the cost of doing this work drops by eighty percent, does the world want eighty percent less of it, or does it want vastly more? The honest answer is different for every task you do, and it is the difference between a line of business that triples and one that stops existing.
But demand alone does not tell you who gets paid. A task can explode in volume and still be worthless to you, because the person who used to buy it from you can now do it themselves. So there is a second question, and the two together are the whole diagnosis.
Every task the business does, placed by those two answers. Nothing here is about job titles.
Nobody needs more of this than they already do, but the value still lands with you. The gain is margin, not growth. Take the cost out and keep the difference. Do not expect the top line to move.
Cheaper means more of it, and you are still the one who gets paid for it. This is where reclaimed hours belong. Defend the reason the value stays with you, because that is the whole position.
The worst corner. Volume will not grow and the price is falling toward the cost of running the tool. Automate it hard, price it as a utility, and move the people out of it before the market decides for you.
The volume is real but so is the do-it-yourself risk. There is far more of this work about to happen, and most of it will not come to you. Win it back on judgment and standards, or productize it and sell the tool.
One task, one placement, one decision. A single job title routinely lands in three of these corners at once, which is exactly why we do not map roles.
There is no survey to fill in and no software crawling your systems. Identify is human beings interviewing your people, analyzing what they hear, and uncovering how the business actually runs.
We sit with the people doing the work and ask what their week really contains. Where the hours go. Which parts they dread. Which parts they would never hand to anybody. What quietly breaks when one person is out. Who they call when something goes wrong, and whether that person is on the org chart at all.
Then we go over it again with the people one level up, because the two accounts almost never match, and the gap between them is usually where the real exposure is. Only once we understand the business from the inside does any of it go onto the matrix.
Your people, and enough calendar to talk to them properly. Whatever process documentation exists is welcome, and if there is almost none, that is normal and it is not a prerequisite. We are not asking you to prepare anything.
Several days for a focused business, several weeks for a complicated one. It runs as long as it needs to, because a map built from three conversations is a guess. We tell you which one you look like before we start.
An honest account of every task the business does and where each one sits. No recommendations yet. The point of this step is that you stop guessing about your own exposure, and so do we.
This is the deliverable, and it is not a slide deck of principles. It is a document with scores, money, owners and dates in it, written so that someone who was not in the room can act on it. Most of what follows is the same structure every time. The content is yours.
Each line rated 1 to 4, from severe threat to clear opportunity, built from four factors rather than a feeling: how short the adaptation window is, how much revenue and reputation ride on it, how far the value can slide to the buyer, and how much of the work is already commoditized.
The full task map with each item placed, and an intensity read on both axes so you can compare two tasks that landed in the same corner for very different reasons.
What each score is worth in money. Current pricing and how it is structured, what AI compression does to that structure, and the expected gain per line if you execute. This is the section that decides whether the plan gets funded.
Every line placed by urgency against importance, so the sequence is an argument rather than a preference. The shortest adaptation windows come first whether or not they are the most interesting problems.
Every committed move in one place, each with a person's name against it. A plan with no owner is a wish, and we have watched enough of those to stop writing them.
Which tools, at which layer, and who inside your business is responsible for each. Including the ones you already pay for and should stop paying for.
Each measure carries a current figure, a target, and a directly responsible individual. If we cannot get a number for today, it does not go in as a KPI.
Sprint by sprint, from the exposed lines through the margin plays to measuring the leverage. Dependencies stated, so nothing in week six quietly depends on something nobody started in week two.
Every recommendation moves work off a person and onto a system. The people doing that work have a craft and a reputation attached to it. This section says who will react which way, and what to say to them before the rollout rather than after.
The engagement can end here, and for some businesses it should. The Report is yours. If you never speak to us again you still have the scores, the economics, the owners and the sequence, and any competent team can run it.
If you want to see one end to end before you commit to anything, there is a complete worked report on this site.
Plenty of consultants will sell you the Report. Far fewer will wire it into how the business actually runs, which is the only step where anything changes. The plan becomes one connected system: your data at the bottom, the models above it, the workflows that run on them, and the places your team already works. Guardrails and measurement run beside all of it rather than getting bolted on at the end.
AI prepares. People decide. Your data stays yours; Gainspring builds and supports every layer above it. Security, privacy and measurement apply at every layer, not just at the end.
There is an approval step in front of anything that reaches a client, a customer or a ledger. What the approver decides feeds back down and improves the next run, which is the part that makes the gain survive past the pilot instead of dying with a single champion.
We build and support everything above it. We scope who can see what, keep sensitive material where it belongs, and can run the whole system in an isolated environment when the work demands it. There is a longer answer to this below.
Training is part of the build, not an afterthought sold separately. A system only your vendor understands is a dependency, and we are not interested in selling you one.
This is the step everyone assumes is automatic, and it is the one that fails most often. A task gets faster, nobody decides what to do with the difference, and the time disappears into the day.
So the Report already says, task by task, which of two things happens to the hour. Either the task retires and the hour moves to something with a higher return, or the task keeps its hour and the person spends it raising the standard of the output. Both are deliberate. Neither is a headcount decision.
Then we measure it on a schedule, because a shift nobody is checking on quietly reverses. That includes going back to the people whose work changed and asking how it landed, using the same survey and the same questions, so the second reading means something.
The thinking behind the two movesMost of what follows is not really an objection. It is something a smart person believes about how these systems work that stopped being true a year or two ago, and it is worth saying plainly why.
This is the belief sitting behind almost every other objection, and it rests on a picture of how these systems work that is a few years out of date. Using a model does not mean handing your data to anyone, and it does not mean the model learns from it. There is a ladder of options here, and every business we work with lands on one of the rungs rather than choosing between all in and nothing.
The business tiers from Anthropic, OpenAI and Google do not train on what you send and do not retain it beyond the request. This is contractual, not a setting you hope somebody ticked. It is where most work belongs.
The same class of model, running through AWS Bedrock, Google Vertex or Azure, inside the tenancy you already hold. Your data does not leave the boundary your security team has already reviewed and signed off on.
For work that genuinely cannot leave the building, open-weight models run on your own machines with no outbound connection at all. Air-gapped in the literal sense. Capable enough for the great majority of document, drafting and analysis work.
Independent of which rung you pick: these systems read what they are handed at the moment of the request and retain nothing afterward. Your knowledge base stays a database you control. Revoke a document and the next answer no longer knows about it.
Different work can sit on different rungs in the same business. Client files under one rule, internal drafting under another.
They have to, and it is not optional in any business with clients. The system inherits the permissions you already run: it can only retrieve what that specific person is already cleared to open, so an assistant asking a question does not get an answer built out of the partner's files. Every retrieval and every output is logged, which usually turns out to be better traceability than the shared drive it replaced.
No, and we would tell you if we thought otherwise. Every recommendation in the Report ends with a person still holding the decision. What changes is what they are responsible for. If your goal is a headcount reduction, we are not the right partner for it. Not on principle, but because the businesses that cut first tend to lose the judgment that was the only thing keeping their value from sliding to the buyer.
Individual use and an operating system are different objects. A handful of people getting good at prompting produces real gains that are invisible on the P&L, disappear when those people leave, and never touch the tasks nobody enjoys enough to volunteer for. The gap is not capability. It is that nothing is connected to your data, nothing is measured, and nothing survives the champion.
The matrix does not care what industry you are in. It asks two questions that apply to a law practice, a construction company and a payroll bureau in exactly the same way. What differs is where the tasks land, and that is the whole reason we map yours instead of handing you somebody else's answer.
Then you stop after the Report. It is written to be executed by your own team and it is yours outright. We would rather you run it internally than pay us to be in the room for something you can do.
You do. The workflows, the prompts, the configuration, the documentation. No part of the system is licensed back to you and nothing stops working if you stop working with us.