A public body makes a sourcing decision once and then defends it for years, to auditors, to the bidders who did not win, and to the staff who live inside whatever was signed. That defence rarely turns on how well the recommendation was argued. It turns on whether the evidence behind it was gathered on purpose, by people who knew which question it would eventually have to answer.
A large European public-sector organisation preparing to re-tender its outsourced IT service desk wrote its aim as procuring a first rate desk, putting quality ahead of price in the sentence that governed everything after it. Then it split that aim into six objectives. Every one of the six fed a single artefact, a detailed technical requirement, and that requirement was the document that would go to the market. The shape is the finding. The tender was to be the output of six lines of inquiry rather than the place where an argument got composed.
The challenge
The six were specific. A detailed analysis of the running contract and its associated agreements, together with the service levels and performance indicators it already carried. Input drawn from the business areas and reconciled into a statement of what the future service had to do, including the service levels it would be held to, with the findings reported back and summarised. Contract provisions drawn from what the supplier market would actually offer, covering terms and conditions, penalties, and motivation built on those same service levels. A supply market analysis with recommendations attached to it. Options for folding additional support services into the same contract. And strategies for moving the service to a new provider with the risk kept low.
Four of those gather evidence. Two convert evidence into contract language. Keeping the two kinds apart is the discipline, because the ordinary failure is to run all six as one act of drafting, where the requirement is written from what the IT function already believes and the evidence assembled afterwards to support it. A document built that way reads well and cannot be defended a line at a time. Ask which piece of evidence produced clause nine and the answer is a meeting.
The structure also fixes what happens when one stream comes back thin. If the market work is weak, the contract terms drawn from it are the weak part, and anyone reading the file can see which part that is. A single fused document hides that. Six streams into one does not.
The record we hold is the kick-off agreement and the plan it fixed, so treat the rest of this as method rather than result. The specification, the recommendations and the supplier evaluation are what the plan committed to produce.
The approach
The most instructive thing in the plan is the phase that produced nothing. Seven phases were laid across twenty-five working days and six of them named a key output. The second, planning and data gathering, carried a dash in that column, and its stated objective was to collect the inputs the later phases needed in order to reduce project risk. A deliberate blank against a funded phase is an unusual act of honesty in a schedule that tight, and it is the phase most likely to be quietly deleted, because it is the only one that gives the client nothing to hold at the end of the week.
That phase was scoped to gather five classes of evidence: the contract and its associated agreements, financial data, performance reports, organisational structures and profiles, and processes. The fourth of those is worth noticing. Who reports to whom, and which roles exist, was treated as procurement evidence rather than as background, because the shape of the people is what decides whether a transition is survivable.
Four of the five classes are unstructured document sets, and reading them is the slowest part of a study built this way. That is the part of this work that has genuinely changed. Retrieval-augmented pipelines over a contract set, a finance archive and years of performance reporting return sourced extracts in hours rather than weeks, provided someone has specified what a good extract looks like and holds an evaluation set to check it against. The Platform work we do starts there, with the ingestion and the checks, because a pipeline that reads a contract stack without a way to prove it read them correctly is a demonstration rather than evidence.
The current-state diagnosis was specified as a single comparison across two axes: the service levels and performance indicators as agreed, set beside the same measures as achieved. Most performance reporting shows achievement without putting the commitment next to it, which is why so many re-tenders open with an argument about whether the incumbent underperformed. The comparison fitted a twenty-five day plan only because both sides of it already sat in one system of record, where service level management ran alongside incident, problem, change and configuration. In an estate where commitments live in a signed document and results live somewhere else, that same task is a reconciliation project rather than a review. No performance figures for this client appear here; the method is what transfers.
Future-state requirements were to come from a workshop framed around the issues and the options for changing the service, rather than around a list of wants, with up to twenty-five business users consulted in groups over video where travel could not be justified. That ceiling was a calendar decision, and it caps how deep any conversation goes. The market stream was scoped as secondary research across four topics: outsourcing models, contract terms, penalties, and motivation built on service levels.
The seam in an otherwise deliberate structure sits right there. Among fifteen written assumptions was one holding that no more than three suppliers would respond and be evaluated. It was fixed at kick-off, before the market work began, and secondary desk research cannot test it. The number was an estimate of evaluation effort, and it reads like a judgement about how competitive the market is. Nothing in the method could tell the two apart.
The outcome
The plan closed the way it opened, on structure rather than on effort. Four progress reviews were set at exactly seven-day intervals. One draft per deliverable, with a week for the client to respond. Sign-off was its own phase with its own named output, reached through two presentation events, a draft for review and a final presentation to the stakeholders who had to accept it. That final presentation occupied the last working-day column on the chart, with nothing behind it.
What this record evidences is that plan, the agreed inputs and obligations behind it, and the method each stream was to follow. It does not evidence a signed contract, an appointed supplier or a service that improved. A second deliverable was conditional on an extension that had not yet been confirmed.
Two of the six streams would be run differently now. Market analysis at this speed had to be desk research, which is what left the three-supplier assumption untestable; agentic market scanning can enumerate and qualify a real bidder pool cheaply enough that the assumption becomes a finding instead. Requirements no longer have to stop at a sample of twenty-five either, because ticket histories, survey text and interview transcripts across every site can be read in full. The organisational stream does not move. Reading who will resist a transition is still a person's job.
There is a new stream that did not exist then. Any supplier bidding for a desk like this one now prices automation into the offer, which puts three questions inside the specification that were not previously there: what the provider automates, what level of autonomy it runs at and who approves each step, and who owns the automation and the knowledge base it was trained on when the term ends. A public buyer in Europe carries obligations under the AI Act for the systems it procures, and those obligations are cheaper to write into a requirement than to retrofit into a signed contract. Our Consult engagements start by putting that stream beside the other six rather than appending it as an annex.
Six streams into one document is slower to organise and far easier to defend. The alternative is a decision that reads persuasively and cannot be taken apart, which is the same as a decision nobody can stand behind once the people who made it have moved on.