For twenty years, getting numbers you could trust meant running a project. Scope it, budget it, staff it, survive it. Six months if you were lucky, six figures almost always, and at the end — a data warehouse, a set of dashboards, and a quiet new dependency on whoever built them.
That era is ending. Not as a prediction, and not because anyone declared it. It’s ending because the arithmetic underneath it broke.
This is the argument in full. The short version lives on The Shift ; this is the part that has to survive scrutiny.
How reporting became a project in the first place
It’s worth being fair to the old model, because it wasn’t stupid. It was correct — for its time.
In the 2000s and 2010s, if a mid-market company wanted management reporting, there was genuinely nothing to adopt. Source systems were closed; connectors barely existed. Every data warehouse was bespoke because it had to be. Every semantic model was built from zero because no reusable one existed. And the people who could do the work were consultants, priced by the day, learning your business on your budget.
So reporting became a project, with a project’s anatomy: requirements, build, delivery, handover. And a project’s failure modes: scope that grows, timelines that slip, and — the part nobody budgeted — maintenance. The project ships, the business changes, the dashboards decay. Three years later, someone proposes a new project.
Every CFO who has been in the seat longer than one cycle knows this loop personally. The loop wasn’t a flaw in execution. It was the model working as designed: bespoke things decay, and bespoke was the only option.
Then, over about five years, all three conditions that justified the model stopped holding.
Force one: AI commoditised the output layer
Look at what the implementation project actually sold you: dashboards, a monthly pack, commentary on the variances. The visible layer. The deliverables in the demo.
All of it is now the cheap part. A language model drafts a competent variance comment today. Dashboard tooling stopped being differentiating years ago — the tools are excellent and roughly equivalent. The board pack that took an analyst three days assembles itself, if the data underneath allows it.
If the data underneath allows it — that’s the entire hinge. AI didn’t make numbers trustworthy; it made untrustworthy numbers more fluent. A model will comment on a wrong margin as confidently as a right one, and more persuasively than the spreadsheet ever did.
So the scarce thing moved. It used to be the ability to produce the output. Now it’s whether the number entering the output is right: one definition per metric, reconciled to the ledger, owned by someone who answers for it. The project era priced the output. The output is now nearly free, and the layer underneath it — the governed one — is what’s worth paying for.
Force two: the middle layer became a product
Here’s the observation that took us longest to accept, because it worked against our own billing model at the time: finance logic repeats.
A distributor, a SaaS company and a manufacturer differ in their drivers — stock cover, net revenue retention, yield. They do not differ in what a P&L bridge is, how working capital ties to cash, what a reconciled ledger means, or how a consolidation eliminates intercompany revenue. The structural layer of financial reporting is the same problem, solved again and again, one company at a time, at full price each time.
Solve it enough times and something changes. The model stops being a build and becomes a catalogue. We watched this happen across 70+ companies in 11 countries, on 12+ different ERP and accounting systems: each engagement’s marginal build got smaller, until one day the honest description of what a new client receives was no longer “we will build you a reporting solution” but “we will connect you to one that exists — and has been refined by every company before you.”
That is the difference between implementing and adopting, and it isn’t a pricing gimmick. It’s what happens to every category of bespoke work that matures: ERP itself went this way, payroll went this way, e-commerce infrastructure went this way. Nobody builds their own webshop platform anymore. Reporting infrastructure is simply the next layer down the same path.
Force three: the people never scaled
The third condition is the one hiring markets keep re-proving every quarter.
The person who can do this work properly is fluent in two languages at once: accounting — what an accrual is, why the trial balance matters, what the auditor will ask — and data engineering — how to move, model and govern the numbers at scale. People with genuine fluency in both are rare, expensive, and concentrated in places that aren’t mid-market finance departments.
The old model asked every company to hire or rent its own. The maths never worked: a company that needs this skill for twenty hours a month was asked to compete for a full-time salary, or pay consultancy day rates for it. Most did neither and asked an overworked controller to become a part-time data engineer instead — which is how most in-house dashboard estates actually got built, and why they look the way they do.
Scarce expertise always ends up shared. It happened to legal, to audit, to payroll processing. A shared team serving many companies is not a compromise on the old model — it’s the only arrangement under which this particular skill set is available to the mid-market at all.
The arithmetic that stopped clearing
Put the three forces together and price the two paths honestly.
Build: six to nine months before the first trusted number, a six-figure project cost, then the part the proposal never itemised — the analyst or agency retainer that keeps it alive, and the rebuild when the business changes shape. Cost of ownership, not cost of project.
Adopt: connection in days, first governed reports in weeks, a monthly fee that includes the upkeep — because the model being maintained is the same one every other client runs on, so maintenance is a product function, not your payroll line.
When the same outcome arrives ten times faster at a fraction of the lifetime cost, a market doesn’t debate the shift. It just moves. You can already see the incumbents conceding the point in their own language: every BI consultancy now sells “accelerators”, every tool promises insight “out of the box”. The market has admitted the pure build is over. It just hasn’t finished updating the price tag.
What doesn’t change
An argument this convenient to our business deserves its honest boundaries, so here they are.
Your ERP stays. The transaction systems — the ledger, the invoicing, the statutory filings — are good at their jobs and are not what this is about. The shift happens above them, not instead of them. Anyone using this argument to sell you a new ERP has misread it; most companies asking for a new ERP need the data already inside the one they run .
Judgment stays. The governed layer makes numbers trustworthy; it does not make decisions. Someone still reviews, still signs, still answers the question behind the question. If anything, the shift raises the value of that judgment, because it stops being spent on assembling the numbers and starts being spent on what they mean.
And building is still right for some companies. If you have a real data team, finance is not its bottleneck, and reporting infrastructure is a competency you want to own — build. Seriously. The methodology is not a secret; most of it is in this Knowledge Hub. The end of the implementation project doesn’t mean nobody should build. It means building stopped being the default — the thing you do because there’s no alternative.
What this means for the four proposals on your desk
If the argument above is right, it has one practical consequence, and it’s the one we built The Shift around.
The proposals arriving on a CFO’s desk right now — the reporting project, the ERP replacement, this quarter’s tool, the controller search — are all reaching for the same thing: numbers you can trust, on the systems you already run, delivered as an outcome, by people who know what they’re doing. They look like four decisions because four different industries are selling them. Underneath, it’s one: keep buying the era that’s ending, or adopt what replaced it.
The implementation project had a good twenty years. It solved a real problem in the only way its time allowed. But the conditions that justified it — closed systems, unreusable models, unshareable expertise — are gone, and they aren’t coming back.
The era isn’t ending because someone declared it. The arithmetic ended it. The only question left is how long your company keeps paying the old price.