The history did not come with us

The bill for changing supplier arrives before the cost does, and the two are not the same number. That is the thing three people who have moved a business from one system to another say, in three different trades, without having met.

Money & World asked business owners, founders and operations managers two questions. What they switched and what triggered it, and what the switch cost that they had not expected. Twenty eight people answered in the first day. Three of them answered from their own operations rather than in general terms, and they were changing completely different things: an enterprise resource planning system, a search measurement platform, and the advertising accounts sitting behind a set of online stores.

None of them names the migration fee. All three name the record the old system held.

The part of the budget nobody quotes

Answer: Girish Songirkar says the enterprise platform moves he has worked on take around nine to fifteen months, that the first months on the new system carry a productivity fall of fifteen to twenty percent, and that cleaning the data before it can be migrated absorbs roughly thirty percent of the project budget and is usually missing from the quoted price.

Girish Songirkar is a delivery manager in enterprise software engineering at ArionERP. He describes what he has observed across large transitions from legacy on-premise systems to cloud ones rather than a single named project.

"When it comes to enterprise platform switches, the highest cost does not always come from implementation fees or subscription charges. In fact, the real cost of change manifests itself as productivity tax, which is incurred during the period of expected operational inefficiency."

"Based on my experience with large scale transitions of legacy on-premise ERPs to cloud-based systems, I have observed that transition processes usually take around 9-15 months to be successfully completed. Transition triggers usually include reaching a scalability ceiling or vendor refusal to assist with legacy modules. However, unfortunately, an unexpected large cost of switch is incurred on the stage of data cleansing that must precede any data migration."

The reason he gives for the cleansing bill is worth separating from the bill itself, because it explains why the cost lands at the moment of the switch rather than earlier.

"A frequent mistake that companies make is underestimating the price of muscle memory in that even though a new software is going to be faster, the first few months after switching the technology involve a serious drop in productivity, which can be as high as 15-20 percent. This sudden drop occurs due to the fact that employees have to relearn the processes as well as face the issue of several duplicates or incomplete records that old systems were able to tolerate and new systems reject as erroneous."

"To clean this data before migration firms spend, on average, around 30 percent of their overall project budget, which is usually not taken into account in quoted prices."

The duplicate and incomplete records were there the whole time. The old system tolerated them, which meant they cost nothing and were therefore invisible. The new system rejects them, and the accumulated tolerance of a decade turns into a line item on a project plan. Nothing about the data changed. The threshold moved.

His last point is about people rather than records, and he puts a failure condition on it.

"The most damaging aspect of a poorly planned transition is loss of trust within an organization. Whenever the IT directors s and operations directors do not agree on what a successful project looks like, the transition never gets completed. A transition never gets completed unless the expenses arising from retraining are treated as an investment rather than an inconvenience."

Two years of numbers with the meaning stripped out

Answer: Nassira Sennoune describes moving a property brokerage from one rank tracking platform to another and finding that two years of position history stayed behind. The export produced numbers without annotations or any record of which change was made in which week, and the question her director actually asked became unanswerable for several months.

Nassira Sennoune is an SEO consultant at Originn Properties. The trigger she gives is capacity rather than price.

"We changed rank tracking and site crawling platform, and the cost that hurt was historical data."

"The trigger was straightforward. We are a property brokerage tracking a lot of location-specific terms across several markets, and our plan's keyword limits meant we were sampling rather than monitoring. Every time we added an area we had to drop one. The decision took a couple of weeks of comparisons. Being genuinely operational on the new platform took about two months."

Then the part that does not appear in any comparison of the two products.

"The history did not come with us. Two years of position data stayed in the old tool. Exports gave me a spreadsheet of numbers with none of the context, no annotations, no record of which change we made in which week. So the question a brokerage director actually asks, whether a page is better or worse than last spring, became unanswerable for the first several months on the new platform until it built its own baseline."

Notice what did move. The numbers moved. A spreadsheet of positions came out of the old tool and went into a file, and by any technical description of data portability that is a successful export. What stayed behind was the layer that made the numbers mean something: which week the team changed a page, what they were testing, why a line moves in March. That layer is not a dataset. It is a set of annotations written by people who work there, and it lived inside the vendor's product.

She adds a second effect that is easy to mistake for a result.

"The numbers disagreed. The two platforms measure position differently enough that the same page showed a meaningfully different rank on the same day. That is not an error, it is methodology, but it means the moment of the switch looks like a performance change in any report that spans it, and I spent real time explaining that it was not."

And a third, which is the only one here that is straightforwardly financial.

"The price moved once we committed. The rate I evaluated was for the annual commitment at a seat count we outgrew within a few months, and the added seats did not get that rate."

Her own conclusion is a sentence about procurement rather than about software.

"The lesson: with any measurement platform, you are not just buying a tool, you are abandoning a baseline. Price that in, because the reporting gap is the part people notice."

Where the value actually sits

Answer: Igor Ivitskiy says an audit of thirty one Google Ads accounts covering 9.46 million search term rows found a median of sixty two percent of conversions coming from the top one percent of search terms by spend. On his reading, that means a migration which breaks conversion tracking for two weeks damages most of an account's measurable performance while total spending, and therefore the report, still looks normal.

Igor Ivitskiy is the founder of Doctor Ads, and he sees these switches from a position none of the other contributors occupy, which is downstream of somebody else's decision.

"I see these switches from the ad account side rather than from procurement, because when a company changes its store platform, CRM or analytics vendor, the part of the bill that lands in my lap is lost conversion history rather than migration fees."

His figures are from his own audits and are in no public dataset, so they are his account rather than a finding of ours.

"The cost that gets underestimated is that a tracking break is not proportional to the size of the account, since the value inside an account is not spread evenly. Across 31 Google Ads accounts I audited, covering 9.46 million search term rows, a median of 62% of conversions came from just the top 1% of search terms by spend, and the middle half of accounts sat between 51% and 72%, so the results rest on a very thin layer of history."

The consequence he draws from it is the one that matters, because it describes a loss that hides itself.

"That means a migration that breaks conversion tracking for two weeks damages most of the account's measurable performance while the reporting still looks busy, because the other 99% of terms keep spending as usual and hide the hole."

That is the same structure Sennoune describes from the other end. Her export produced numbers and lost the meaning. His broken tracking loses the meaning and leaves the numbers running. In both cases the dashboard keeps filling.

He also inverts the standard advice about timing.

"The counterintuitive part is that switching during a quiet season, which is what most managers choose to be safe, is often the more expensive option: low volume gives the automated bidding almost nothing to relearn on, so the account keeps paying at targets it can no longer support."

The law that fixes the exit and not the arrival

Answer: The European Union has legislated switching costs directly. Chapter VI of the Data Act requires providers to remove obstacles to leaving, caps the notice and transition periods, and abolishes switching charges outright from 12 January 2027. What it does not do is guarantee the transfer of the thing these three contributors describe losing.

Regulation (EU) 2023/2854, the Data Act, has applied since 12 September 2025. Chapter VI is about switching between data processing services, and it is unusually concrete about money and time.

Article 23 requires providers to remove pre-commercial, commercial, technical, contractual and organisational obstacles that stop a customer terminating the contract, signing with a different provider of the same service type, or porting exportable data and digital assets elsewhere, including to on-premises infrastructure. Article 25 puts the mechanics into the contract itself: a mandatory maximum transitional period of thirty calendar days at 25(2)(a), a notice period for initiating the switch capped at two months at 25(2)(d), and a further retrieval window of at least thirty calendar days after the transition ends at 25(2)(g).

Article 29 handles the price of leaving. From 12 January 2027 providers may impose no switching charges at all. In the window that opened on 11 January 2024 and runs to that date they may charge reduced fees, and Article 29(3) caps those at the costs the provider actually incurs that are directly linked to the switch. Article 29(5) goes further and requires a provider to tell customers, where relevant, when a service is one that involves highly complex or costly switching, or one where switching is impossible without significant interference in the data or the service architecture.

Set that against what the three contributors say the switch cost them, and the fit is partial in a specific way.

The definition of what travels is at Article 2(38). Exportable data means the input and output data, including metadata, generated directly or indirectly by the customer's use of the service, and it excludes anything protected by intellectual property rights or constituting a trade secret of the provider or a third party. Article 25(2)(f) then lets the contract set out an exhaustive list of categories of data specific to the internal functioning of the provider's own service which are exempted on exactly that ground.

Sennoune's spreadsheet of positions is exportable data. Whether the annotations, the change log and the platform's own ranking methodology are is a different question, and the second of those three is the reason her director's question became unanswerable. Ivitskiy's conversion history is data the customer generated, and the learned state of the bidding system that consumed it is not.

Article 2(37) is candid about the ceiling. Functional equivalence means re-establishing, from the customer's exportable data and digital assets, a minimum level of functionality in the new environment. Minimum is the word the drafters chose, and it is doing real work: the obligation is to get the customer running again, not to reconstitute what the customer had.

So the strongest switching regime any jurisdiction has written prices the exit at zero from January 2027, bounds the transition at thirty days, and still leaves the cost these three practitioners actually paid sitting where it was. Songirkar's cleansing bill is work on the customer's own records and no provider owes it. Sennoune's lost baseline is partly the vendor's methodology. Ivitskiy's damage happens in the gap while the new system learns.

The invoice is the part that is regulated. The interval is the part that costs money, and nothing in the law shortens it.