Published September 15, 2026
Two large and rapidly growing sources of avoidable cost — and the three places the loop can be cut
Part I: The Tax
1
The double payment
Companies have learnt to negotiate the price of almost everything. Media rates, software licences, agency retainers, implementation fees, freight, headcount — each has an owner, a benchmark and somebody measured on bringing it down. Yet two large and rapidly growing sources of avoidable cost pass through most organisations every year with remarkably little argument.
The first is what a company pays to reach customers it already has. The second is what it pays to rebuild software foundations it has already bought. Both are entirely visible. Neither is usually seen for what it is, and in most organisations neither has been put in front of a board as a number.
The pattern underneath both sounds implausible when stated plainly. Companies repeatedly pay for value they have already created. They acquire a customer, earn an identity, accumulate a history and establish permission to make contact — then, when attention weakens, pay an outside platform to put the same person back in view. They buy software application by application, each arriving with its own copy of the same foundations — then pay to integrate the copies and hire people to keep them aligned.
The two payments look nothing alike. One sits in marketing and is priced by auction; the other sits in technology and is priced by licence. Economically they are the same act: paying a second time for an asset that should have compounded after the first payment. A customer relationship ought to become more valuable with every interaction. Software machinery ought to make the next capability cheaper to produce. Instead both are repurchased.
The word tax is used here in its economic sense rather than its statutory one: a recurring toll that rises with the level of business activity, is treated as unavoidable, and is rarely examined as a category. Nothing in the argument depends on the word. What matters is that both payments behave that way, and that neither has a name. The Reacquisition Tax is what a business pays to reach people it already knows. The App-Stack Tax is what it pays to rebuild foundations it has already funded.
Figure 1. Two payments leaving the business for assets it already holds.
2
Paying to reach people you already know
The mechanism is so ordinary that it takes an effort to see at all.
A customer bought something eighteen months ago. Their address, their order history and, in most cases, a valid permission to contact them are all in the database. At some point they stopped opening what the company sent, and the company — noticing the gap in revenue rather than the gap in attention — buys them back through a platform which holds the same address the company already has, at a price set by an auction the company also pays to enter.
The precision that matters here is easy to lose, and losing it is exactly how the tax stays invisible. The business has not lost the customer’s identity. It has lost their attention. Those are different failures with different remedies and different costs, and treating the second as though it were the first is what turns a relationship problem into a media budget. The identity was never lost. The attention was — and the tax is what a company pays somebody else to supply the attention it stopped earning.
Platforms are exceptionally good at finding intent, and that competence is what makes the tax feel like ordinary growth. The platform can tell when somebody is browsing, comparing or returning to the category, and it reports the resulting conversion with confidence. The dashboard shows a new customer acquired at an acceptable cost. The dashboard is not wrong; it is answering a narrower question than the one that matters. The same person can be labelled new by the channel and known by the company at the same moment. The channel takes the credit and the profit and loss statement absorbs the duplication.
This is not an argument that paid acquisition is waste. A business with no prior relationship to a buyer has to pay somebody to create one, and that money is well spent. The avoidable portion is narrower: the part of the acquisition budget landing on people already in the customer file, still contactable, and reachable at a marginal delivery cost close to nothing — if only they were still paying attention.
Two honest caveats. Being present in a database is not the same as being contactable: permissions lapse, addresses decay and deliverability is earned rather than assumed. And low marginal delivery cost is not zero cost — the message still has to be worth opening, which takes data, judgement and people. The claim is not that owned contact is free. It is that it is an order of magnitude cheaper than renting the same person back, and that the gap is rarely calculated.
3
Paying for the same foundations eight times
A mid-sized consumer business runs somewhere between eight and twenty applications to serve its customers. Each of them, underneath the part that makes it distinctive, contains the same things: a model of the customer, a model of the product, a record of orders, a consent state, a workflow engine, a reporting layer.
Every one of those foundations was built by the vendor, and every one is charged for. The buyer pays for the same substructure once per supplier, and uses only the thin layer on top where the suppliers differ from each other.
The subscriptions are the visible part and, in most companies, the smallest. The true bill is eight subscriptions, plus the integrations between them, plus the work of deciding which system is right when two disagree, plus the people employed to hold the arrangement together. The last of these is often the largest and is the least likely to appear in any assessment of what the software costs, because it is filed under headcount rather than under software.
The largest consequence may not be the bill at all. It is the loss of organisational speed. A simple change becomes a cross-application project. A new workflow waits on three internal teams and four suppliers. A question about a single customer produces five answers, and somebody has to decide which one the business will act on. The stack delivers a great many capabilities while steadily reducing the company’s ability to behave as one system — and that cost appears on no invoice at all.
The same qualification applies as before, and it needs to be operational rather than gestural, because otherwise this reads as an argument against buying the best available tool — which it is not. Two questions, applied to any application at renewal.
— Does it contribute a distinctive capability the business could not otherwise obtain?
— What share of its cost is that capability, as against rebuilding identity, catalogue, workflow, permissions and reporting underneath it?
An application that passes the first is specialised software and worth its price. An application that fails the second is mostly a copy of foundations already funded, sold at the price of the feature on top. The App-Stack Tax is the second category. Nothing here argues against the first.
4
Visible as cost, invisible as waste
Neither tax is hidden. Both are entirely present on the profit and loss statement, in categories any finance function can read: acquisition cost, marketplace commission, software subscriptions, implementation, integration, operations headcount. Anybody can see the money. What nobody can see is that a portion of it did not need to be spent.
Figure 2. The taxes are not concealed. They are distributed across categories that each look reasonable in isolation, and each of which already has a sponsor.
The finance function is well equipped to ask whether a cost is too high. It has no instrument for asking whether a cost needed to exist. Those are different questions, and only the first has a procedure attached to it. A 12 per cent reduction in the acquisition budget is a negotiation anybody knows how to run. A question about which share of that budget is buying back people the company already knew has no owner, no method and no meeting.
So the position is this. Both taxes are visible as cost and invisible as waste. No budget line is called money we did not need to spend. Nobody’s objective includes reducing it. No supplier has any interest in pointing it out, since both taxes are somebody’s revenue. And because the total is distributed across half a dozen categories that each look defensible on their own, no single review has ever seen the whole of it in one place.
There is an organisational version of the same asymmetry, and it is the more stubborn of the two. The media team is accountable for the return on the next campaign, not for whether that campaign reached people the business should still have been able to reach itself. The application owner is accountable for whether the tool works, not for whether seven other tools already contain its foundations. Every participant can be locally correct while the company is globally wasteful — and because every participant is locally correct, every participant can defend their position without lying, which is why these conversations never resolve.
There is a third reason, and it is the least comfortable of the three. Removing either tax reduces somebody’s budget, vendor estate or organisational standing before it improves the company’s profit. The saving arrives later and lands on the profit and loss statement; the loss arrives immediately and lands on a person. That asymmetry is enough to keep a well-run business paying both taxes indefinitely, and no amount of analysis dissolves it.
Challenging either tax therefore requires something no review is designed to do: change the unit of analysis. Stop asking whether this campaign performed or this application delivered its features, and ask whether the payment needed to exist.
Figure 3. Both questions are reasonable. Only one of them has an owner, a method and a place on an agenda.
Companies scrutinise what has been named. That is the entire argument for naming these.
Key points
— Both taxes are the same act: paying a second time for an asset that should have compounded.
— The identity was never lost. The attention was.
— The software bill is the visible edge; the larger costs are headcount and lost speed.
— Both are visible as cost and invisible as waste, and every participant is locally correct.
— Removing either tax costs somebody their budget before it improves anybody’s profit.
Part II: The Loop
5
Both were rational when they started
It would be convenient to treat all of this as negligence. It is not, and an essay that says so will be admired and not forwarded. Both taxes began as sensible answers to real constraints, and neither was a failure of attention at the time it was incurred.
| The decision |
The constraint that made it right |
What changed |
| Buy the audience |
A customer who had stopped opening was, in practice, unreachable. Nothing in the owned channel could win the attention back, so the attention had to be rented from whoever had assembled it. |
The owned surface became capable of earning attention again, and the effect of an intervention became measurable against a control rather than asserted. |
| Buy the eighth application |
Eight jobs needed doing and nobody sold a ninth thing that did all of them. Assembling a stack was not a preference; it was the only route to the work getting done at all. |
Producing software stopped being the expensive part, which makes a single system covering several jobs buildable at a price a small business can pay. |
Table 1. Both taxes were rational responses to constraints that no longer hold.
That is a different accusation from carelessness, and considerably easier to act on, because it asks nothing of anybody’s judgement in 2019. It only asks whether the constraint still holds. A cost becomes waste when the alternative changes. The old decision does not become irrational in retrospect; the environment moves around it. Which is also why both taxes have grown rather than shrunk — a decision that was correct when taken is rarely reopened, and rising costs inside it are read as inflation rather than as evidence.
6
How the two taxes feed each other
Each tax is self-reinforcing on its own. Weaker owned attention deepens dependence on platforms, which leaves less reason to invest in the owned channel, which weakens the attention further. More applications create more integration and more reconciliation, which consumes the capacity that might have gone into consolidating them.
The more interesting relationship is the one between them, and it is the reason these belong in one argument rather than two.
Figure 4. The connection runs in both directions, and the second direction is the one that keeps the loop closed.
Read clockwise from the top left. Foundations built separately in each system mean there is no single record of the customer. Without a single record, targeting on owned channels is weak — the business knows what somebody bought in one system and what they browsed in another and cannot reliably join the two. Weak targeting on owned channels is precisely why reach gets rented from platforms which, having observed the same person across thousands of contexts, know more about them than the brand whose product they bought.
That much is familiar. The step that closes the loop is the one at the bottom left, and it is worth sitting with. A company with a large reacquisition budget has, in effect, purchased a way of not fixing its foundations. Reach can be bought. Attention can be bought. So the incomplete customer record never becomes urgent, the consolidation project never reaches the top of the list, and the money that would have paid for it is already committed to the auction.
Stated as an accounting relationship rather than a diagram: the Reacquisition Tax subsidises the App-Stack Tax, and the App-Stack Tax manufactures more Reacquisition Tax. The business pays platforms because its software cannot assemble the customer truth it already possesses, and postpones fixing the software because platforms can always be paid to find the customer again.
They are not two parallel problems that happen to appear in the same business. They are one problem arriving as two invoices.
Key points
— A cost becomes waste when the alternative changes. The environment moves around the old decision.
— Fragmented foundations produce an incomplete record, which is why owned targeting is weak.
— A budget large enough to buy reach removes the urgency to earn it — and that closes the loop.
— One problem, two invoices.
Part III: Three Cuts
7 One loop does not need two programmes
The diagnosis carries an implication that is easy to miss and worth stating on its own. If the two taxes form a single loop, they do not require two transformation programmes. A loop weakens wherever it is broken. So the question facing a business is not how do we fix all of this but which single cut can we make.
One precision, because the loop invites an overstatement worth avoiding. The three cuts do not each solve both taxes. They act at different points and do different work. Measurement drains neither tax; it manufactures the pressure the loop removed. Rebuilding owned attention reduces the Reacquisition Tax directly, and withdraws the subsidy that lets an unfixed stack stay unfixed. Consolidating the foundations reduces the App-Stack Tax directly, and improves the conditions under which owned growth is possible at all. Each cut weakens the loop; none of them is a substitute for the others.
Figure 5. Three places the loop can be broken. Each acts at a different point, and they differ in cost, speed and who has to agree.
| The cut |
What it reduces |
What it costs |
When it pays |
| 1 Measure it |
Neither tax directly. It produces the number that gives the waste an owner and restores the pressure the loop removed. |
A fortnight of somebody’s time. No system changes. No budget. |
Immediately, as pressure — which is the missing ingredient. |
| 2 Rebuild owned attention |
The Reacquisition Tax directly, and it withdraws the subsidy protecting the unfixed stack. |
Modest, and payable out of the tax it reduces. |
One to two quarters, measurable against a control. |
| 3 Consolidate the foundations |
The App-Stack Tax directly, and it improves the conditions for owned growth. |
Migration effort, now an order of magnitude below what it was. |
Slowest and most durable. Compounds thereafter. |
Table 2. The three cuts, in the order most businesses can take them.
The rest of this essay takes them in turn. The first requires nothing from any supplier, which is deliberate: a diagnosis that can only be acted on by buying something is a diagnosis nobody should trust.
8
Cut one: measure it
Both taxes can be sized approximately in an afternoon, and the reason this is so rarely done is not difficulty.
| The question |
How to answer it |
What a bad answer looks like |
| What share of paid conversions were already known to the business? |
Take a recent period of paid conversions and match the identities against the customer file as it stood before the paid interaction. Separate people new to the business from previous buyers, subscribers, app users and known prospects. |
Nobody has run the match. In most companies the figure is not disputed — it has never been produced. |
| Of those, how many had received a useful owned message in the weeks before? |
Not another promotion — a message informed by that person’s state, context and prior relationship. This separates paid recovery that was unavoidable from attention that was allowed to decay. |
The last owned contact was a discount sent to the whole list, and it went out after the paid conversion. |
| How many systems hold their own version of the customer, the product and the transaction? |
Count every system that stores or reconstructs identity, catalogue, orders, consent, events or workflow state — including applications that merely keep a local copy. Then count the integrations built to keep them agreeing. |
The number of integrations exceeds the number of systems, and somebody’s job title exists because of it. |
Table 3. Three questions, all answerable from data the business already holds.
None of this will produce an audited number. Identity matching misses some people and overstates others; attribution windows complicate the story. That is acceptable, because the first objective is not precision. It is to find out whether the machine reporting new customers contains a material known-customer component, and whether the business has bought eight applications or eight incomplete versions of itself.
This cut looks like the weakest of the three and is usually the most important, for a reason that comes straight out of the loop. The loop closes at no pressure to fix the foundations. Measurement is the only cut that manufactures the missing pressure — and it is almost never chosen, because nobody is promoted for producing a number that makes their own budget look wasteful. Somebody senior has to ask for it, be willing to receive an uncomfortable answer, and then act on it against the budgets it implicates. Visibility creates pressure. It does not by itself defeat the people who benefit from the arrangement being invisible.
9
Cut two: rebuild owned attention
The second cut attacks the point where weak owned targeting hands the customer to a platform. If the owned channel earns a response again, recovery stops requiring an auction — and the reacquisition budget stops subsidising the unfixed stack.
Three things have to change, and they are true whoever supplies them. Email is the natural place to start, being the one channel a business owns outright and the one it has most thoroughly worn out.
First, the message has to be current when it is read rather than when it was sent, assembled from the state of the customer and the catalogue at the moment of opening. An email written on Tuesday and read on Friday is presently a Tuesday email; it does not have to be. Second, the useful action has to be completable inside the message rather than three redirects away from it, because that is where most intent is lost. Third — and most brands have none of this at all — some proportion of contact has to exist to maintain the relationship rather than to sell. A channel used only for selling teaches people not to open it, and that lesson, once learnt, is what the reacquisition budget is later paying to unlearn.
Then accountability, which matters more than any supplier’s method. Whoever does this work should be answerable for a customer state rather than for a volume of campaigns, and the effect should be established against a randomised control group, running concurrently, receiving the brand’s pre-agreed baseline treatment — the existing programme, not nothing at all. A control that receives nothing inflates the apparent result and deserves the resistance it will get from anybody asked to deprive their own customers to prove a supplier’s point. The comparison that means anything is against what the business would otherwise have done.
Most businesses will do this work themselves, on a platform they already own, and carry the result themselves. That is the ordinary case and it remains the right one. A transfer of accountability is for the pools a company’s own operation has not closed, and there are usually two of them: customers who are still engaged but whose valuable action has stalled, and customers who have gone dark and are being re-bought through paid channels. The first is larger than most people expect and resolves in weeks; the second is the one visible in the media budget.
What we have built for this
The doctrine is NeoMarketing, and its first commercial forms are Progency Finish and Progency Recover — a managed mandate over one of those declared pools, with accountability for the state of those customers rather than for the campaigns sent to them.
The commercial arrangement is deliberately awkward for us. A baseline is agreed before anything starts, a randomised control runs alongside on that baseline, and payment is a share of the verified difference. No lift, no payment. A supplier arguing that a brand pays too much to re-buy its own customers should be willing to be paid only when it demonstrably reduces that bill. Otherwise the argument is a different invoice with better rhetoric.
What this does not do:
- It does not replace paid acquisition. It makes paid the fallback rather than the first reflex.
- It does not recover everybody. Some customers have left, and the control group is how you find out which.
- It does not work without that control. A claim of lift with no comparison is a claim about the weather.
- It does not fix the foundations. It buys the time that fixing them requires.
10
Cut three: consolidate the foundations
The third cut attacks the origin of the loop: the same substructure, funded once per supplier. The generic move is to stop paying for foundations by the vendor and start paying for them once — one customer record, one catalogue, one order history, one consent ledger, one workflow engine — with the distinctive part of each application sitting on top of that rather than beside it.
The renewal test from section three is how this becomes actionable without a transformation programme. Applications that contribute distinctive capability stay. Applications whose cost is mostly a private rebuild of shared foundations are candidates for consolidation, and there are usually more of them than anybody expects.
The prize is not a smaller software bill, though that follows. It is the removal of hand-offs that exist only because separate applications were built. Search intent should change the next email. A support issue should suppress a promotion. A declared preference should alter merchandising the same afternoon. A well-integrated stack can be made to do some of this, with friction, delay and a standing maintenance cost; on shared foundations it is not a project at all, because the hand-off has stopped existing.
What we have built for this
The Software Foundry builds products on one shared core. The first covers eight jobs: import customers, consent and suppression; ingest catalogue, orders and events; capture identities through forms and pop-ups; build useful segments; send campaigns reliably; run the six essential lifecycle flows; report revenue and activity simply; and migrate from the incumbent without risk. The promise is four items long — one bill, eight jobs, a tenth of the price, a safe way out — and the price is published, the same page shown to everyone, with usage metered and nothing negotiated.
Roughly a tenth of the incumbent price is possible because producing software is no longer the expensive part, and because the foundations are built once rather than once per product. It is not a discount and not a promotional tier that later rises. It is a thesis we intend to prove rather than a result we are claiming, and the proof is whether the second product reuses enough of the first.
What this does not do:
- It does not replace the commerce platform. It sits alongside it.
- It does not do reviews, loyalty, helpdesk or subscriptions, and it has no journey canvas.
- It does not do custom work. Standardisation is what the price is made of.
- It is not outcome-priced. Its accountability is that leaving is as easy as arriving.
11 Why the two remedies are one thesis
The two remedies price themselves in opposite directions, and the contrast is not an inconsistency. It is the clearest evidence that both follow from the same shift.
Figure 6. The same shift, read from both ends.
Artificial intelligence is collapsing the cost of producing capability. It is not collapsing the cost of trust, of migration, of judgement, or of accountability for whether a thing worked. The abundant half is getting cheaper and the scarce half is not. When everybody can produce the capability, the position worth holding is being the party willing to be measured on the result.
So software can become dramatically cheaper in the same period that a verified outcome becomes more valuable. One remedy charges a published price that does not move with the customer’s success, because production became cheap. The other charges a share of a verified difference, because accountability did not. A business can take either cut first — or take only the first cut, buy nothing, and still be better off than it was.
What it should not do is keep paying both taxes on the grounds that each individual invoice looks reasonable. They are one problem arriving as two invoices, and the loop closes at the point where the money that would have fixed it has already been committed to the auction. The opportunity is not to automate the existing bills, or to negotiate them down by a further eight per cent. It is to stop paying them for value the business already created — a customer relationship it built once, and foundations it has now funded several times over.
Key points
— One loop, three cuts. Each acts at a different point; none substitutes for the others.
— Measurement is the cut that manufactures the pressure the loop removed, and it requires no supplier.
— Attention work buys the time that fixing the foundations needs; it does not substitute for it.
— One price rises with verified success; the other refuses to rise. Both follow from the same shift.
Assets should compound, not be repurchased.