The buying process nobody puts on the website

In 2004 Diageo agreed a licence for SAP’s enterprise resource planning software, priced by reference to the number of Named Users. Around 2011 and 2012 it built two new systems on a platform supplied by Salesforce.com: Gen2, which managed the work of its sales and service representatives, and Connect, which let customers place and review their own orders instead of ringing a call centre.

Neither system was an ERP system. Both talked to one. On 16 February 2017 Mrs Justice O’Farrell held that the people using Gen2 and Connect were accessing the licensed software indirectly and required Named User licences to do so, noting that the definition in the agreement itself contemplated use or access via a third party system. SAP had claimed £54,503,578 in additional licence and maintenance fees, and the hearing was fixed to deal with liability only and not quantum.

That is the most important British case about how enterprise software actually gets bought, and it is not about buying at all. It is about a definition agreed seven years before the systems that triggered it existed.

The metric is the purchase

Every enterprise agreement contains a unit of entitlement. A named user, a concurrent user, a processor core, a document, a transaction, an employee count, a revenue band. Buyers negotiate the price per unit at length. They rarely negotiate the definition of the unit at all, and the definition is where the money is.

The reason is a mismatch of time horizons. Price is a present-tense argument between two commercial teams. The definition is a claim on an architecture that does not exist yet. When a business later builds an integration, a self-service portal, a data warehouse feed or an automation that talks to the licensed system, it is making a licensing decision, and the people making it are engineers who have never read the agreement.

Three practical consequences follow, and none of them requires a lawyer to implement.

Hold a written statement of the licence metric for every material system, in language an engineer can act on, and put it where architecture decisions are recorded rather than where contracts are stored. Second, add one question to whatever design review your organisation already runs: does this connect to a licensed system, and if so, who or what is counted at the other end. Third, keep your own count. The only defence against a supplier’s audit position is a defensible internal position, produced continuously, not assembled in six weeks under pressure.

An accounting decision from 2021 decides who is allowed to sign

The second force shaping enterprise buying is not commercial at all. It sits in a technical agenda decision that most technology leaders have never read and most of their finance directors have.

At its March 2021 meeting the IFRS Interpretations Committee finalised an agenda decision on configuration or customisation costs in a cloud computing arrangement under IAS 38, and the Board did not object to it the following month. In the fact pattern considered, a customer buys the right to access a supplier’s application software over a contract term. The Committee observed that the customer often would not recognise an intangible asset for configuration or customisation work, because it does not control the software being configured and those activities do not create a resource controlled by the customer that is separate from that software.

If no intangible asset is recognised, the costs are expensed. Where the services are distinct, they are recognised as an expense as the supplier performs them. Where they are not distinct, because they cannot be separated from the right of access itself, they are recognised as an expense across the contract term. The same treatment applies where a third party does the implementation work. A prepayment for services not yet received is an asset, and nothing more.

Now translate that into a boardroom. A traditional on-premises programme could capitalise a large share of its implementation effort and amortise it over years, which spread the hit to reported profit and kept the decision inside a capital envelope. The equivalent software-as-a-service programme puts most of that same effort straight through the profit and loss account, in the years the work is done.

The software is cheaper and the accounting is uglier. That single fact explains a great deal of behaviour that looks irrational from a technology perspective: multi-year phasing that serves no delivery purpose, an unexplained preference for supplier-financed arrangements that bundle implementation into a subscription fee, a residual affection for perpetual licences in capital-intensive sectors, and finance functions that intervene in system selection much earlier than they used to.

It also means the business case for a migration is not really a technology document. It is a profit forecast, and the person who has to defend it reports to the chief financial officer. Any technology leader who prepares a case without establishing the accounting treatment first is preparing to be surprised by the approval route.

The record that private buyers do not keep

There is one respect in which public buyers are now unambiguously better informed than private ones, and it is worth naming because it is unflattering.

Since February 2025, larger public contracts carry published key performance indicators, and performance against them is assessed and published during the contract’s life. That means a public authority shortlisting a supplier can, in principle, look up how that supplier has performed elsewhere in the public sector, from a source the supplier did not write. What that regime does and does not achieve is examined in what the reforms actually changed for suppliers.

No equivalent exists on the private side. A company evaluating a vendor has the vendor’s references, which are curated, the analyst coverage, which is paid for by the vendors being covered, and whatever its own staff have heard. The one genuinely reliable source is its own history with that supplier, and most organisations do not keep it. Implementation overruns, unplanned change requests, support response times against the service credits, the number of escalations required to close a defect: all of that is knowable and almost none of it survives into the next renewal, because the people who lived through it have moved on and nobody wrote it down in a form the next buyer could find.

Keeping a supplier file is unglamorous, costs one person a few hours a quarter, and is the highest-return procurement practice available to a private buyer. It is also the only thing that converts a renewal from a formality into a negotiation.

The regulator has entered the licensing argument

The third force is new, and it changes the risk profile of long agreements.

On 31 March 2026 the Competition and Markets Authority announced a package of actions on business software and cloud services, under which Microsoft and Amazon set out steps on cloud egress fees and interoperability intended to reduce the expense and effort of using more than one provider. The authority said it would keep engaging with both and that its board would review progress in six months. It did not designate either firm in cloud services.

It did something narrower and, for software buyers, more consequential. On 14 May 2026 the CMA launched a strategic market status investigation into Microsoft’s business software ecosystem, examining whether the company holds that status in business software and whether it can use the position to limit customer choice. The scope named productivity software, personal computer and server operating systems, database management systems and security software. A designation decision is due by February 2027, and designation would open the way to conduct requirements or pro-competition interventions.

For anyone signing a five-year agreement in those categories during 2026, that is a diary entry rather than a headline. The terms available to you may be altered by a regulator inside the life of your contract. It is worth asking, before signature, what happens to your pricing and your entitlements if the licensing rules governing a product change by intervention rather than by negotiation, because standard agreements are silent on it and the answer will otherwise be decided by whoever drafted the change control clause.

What the process actually looks like

Strip away the published methodology and the buying process in a large British organisation runs roughly like this. An incumbent system reaches a point of pain. A shortlist forms, heavily influenced by whoever is already in the estate and by the personal history of the two or three people who will run the selection. A formal evaluation is conducted, scored against weighted criteria that were written after the shortlist existed. A preferred supplier emerges. The commercial negotiation then concerns discount and payment profile, occupies most of the elapsed time, and touches the licence metric barely at all. Finance intervenes late, on the accounting treatment. The contract is signed close to a quarter end, because that is when the discount is available.

None of that is scandalous and most of it is unavoidable. The failures are concentrated in the parts nobody owns: the definition of the unit, the treatment of implementation cost, the terms governing exit, and the absence of any institutional memory of how this supplier behaved last time.

Those four are where the ten-year cost of a system is decided, and all four can be improved without changing the process at all. The pricing structure sitting on top of them is moving in its own direction, which is covered at the shift from seats to meters, and the renewal side of the same relationship is examined in the subscription model after the growth era. The rest of this desk’s work is at SaaS and enterprise software.