Quality has to survive Monday morning

AI can furnish an FM tender with the language of excellence. The harder job is showing what will happen, why it will work and who will be better off when it does.

Facilities management has an unusually unforgiving relationship with reality. A beautifully described maintenance regime is of limited consolation when the heating fails, while the phrase “seamless service delivery” becomes positively provocative when somebody is standing outside a locked building holding the keys that were supposed to work. Eventually, every promise in the tender has to encounter a Monday morning.

That is where Quality begins. In FM, it means buildings that are safe, available and fit for their users, supported by people, systems and decisions capable of keeping them that way. Government’s FM standard encompasses operational availability, compliance, asset condition and a productive, sanitary, secure environment.[1] The tender must explain how those outcomes will survive mobilisation, absence, competing priorities and the discovery that the asset register contains several works of imaginative fiction.

Social Value belongs in this argument, but cannot carry it alone. Employment opportunities, stronger supply chains and environmental improvements can distinguish an offer when they address the contract’s priorities. They do not excuse an implausible staffing model. Under the central-government Social Value Model, in-scope procurements generally give Social Value at least 10% of the overall score, with specified provisions for alternative evaluation methods; this is not a universal rule for every FM buyer.[2] The tender’s actual criteria decide what earns marks.

When everyone sounds excellent

Quality is difficult to create because it requires resources and choices. More resilient cover must come from somewhere; training requires time as well as enthusiasm; maintenance needs reliable information before software can do anything especially intelligent with it. A bid can promise all three in a sentence, while delivery teams must reconcile them with the price.

Generative AI makes the sentence easier. Used carefully, it can help organise evidence and improve an explanation. Fed generic material, it can also produce a remarkably assured account of a service that nobody has properly designed. The danger is that persuasive fluency becomes easier to obtain than operational substance.

There is some evidence for the homogenisation concern: a 2024 experiment found that AI-supported short stories became more similar to one another, even while individual creativity ratings improved.[3] That was a storytelling study, not an FM tender evaluation. Our inference is narrower: when bidders draw on similar tools and familiar language, buyers may have to work harder to distinguish the underlying offers.

An evaluator confronting six polished accounts of proactive, collaborative excellence still needs defensible reasons to score one above another. Government’s AI procurement guidance recognises the need for proportionate checks on suppliers’ capability and the accuracy of their submissions.[4] Suspicion of smooth prose is no substitute for assessment, but neither is admiration of it. The useful response is to make the operational commitments easier to inspect.

The buyer should be able to follow a promise from the paragraph to the person delivering it, and from the invoice to the benefit received.

Give the promise somewhere to stand

At AAT, we approach this through HEARD: Hear and Highlight the need, Empathise with the people affected, Articulate the solution, Reinforce it with proof and Demonstrate the return. It is a discipline for developing the answer, rather than another diagram the evaluator must admire before being allowed to discover the point.

We begin with the specification, scoring descriptors, estate information and conversations with the people who will deliver the service. What is failing? Who experiences the consequences? Which assumptions need testing? Empathy here means understanding the pressure on a school business manager handling another heating complaint, or an estates team attempting to distinguish completed maintenance from reassuringly green dashboard entries.

Consider a hypothetical estate suffering repeated heating faults. “Predictive maintenance” provides an attractive heading, but leaves most of the work unexplained. A credible answer would identify the assets to be surveyed, how criticality determines priorities, who investigates recurring faults, what response cover exists and how the buyer sees whether failures are falling. It would also explain what happens when the first intervention does not work.

A related Social Value commitment might create supported routes into engineering employment. That needs defined opportunities, suitable supervision, a delivery partner, funding, milestones and retention measures. An apprentice is a person requiring a viable role, not a convenient unit of benevolence to scatter across the response.

These commitments must fit the contract and distinguish additional benefit from obligations already being purchased. Under the central-government model, assessment is qualitative: evaluators examine the method and implementation plan against the criteria, rather than scoring promised volumes alone.[2] General corporate virtue cannot substitute for the specific work proposed here.

HEARD must also respect the question. We map the argument to every required element and scoring descriptor, making commitments easy to locate within the word limit. A splendid answer to a neighbouring question remains a disappointing use of everybody’s afternoon.

Bring evidence that can withstand a telephone call

Proof needs to support both the result and our ability to repeat it. For maintenance, that might mean dated work-order records, comparable asset populations, repeat-fault trends, compliance audits and client-validated performance. A percentage needs its denominator, baseline and measurement period; otherwise “a 40% improvement” is merely an impressive number searching for something to have improved.

For the employment commitment, we need evidence of starts, completions and sustained employment, alongside an honest account of our contribution. A partnership agreement can establish capacity to deliver; attendance records demonstrate activity; neither automatically proves a lasting outcome. Relevant case studies should explain the circumstances and limitations, with client verification where permission allows.

Past success also needs connecting to future delivery. We need an approved resource plan, named responsibility, budget, reporting arrangements and corrective action if milestones slip. Certificates may establish useful assurance about management systems, but cannot, by themselves, establish that this mobilisation is achievable. Where the evidence is missing, the answer is to qualify the claim or develop the solution, rather than ask the prose to become more confident.

Follow the money without inventing it

The financial argument should be equally inspectable. Consider an illustrative energy intervention against an agreed, normalised annual baseline of £500,000. An evidenced 8% reduction would represent £40,000 gross annual savings. If additional recurring delivery costs were £10,000, that leaves £30,000 before any gain-share payment, assuming no upfront capital cost. An agreed 70:30 buyer/supplier split of that net saving would leave £21,000 with the buyer and £9,000 with the supplier.

Those figures are a worked example, not a forecast or an AAT case study. A real offer needs an agreed measurement method, adjustments for weather, occupancy and tariffs, treatment of implementation costs, and clarity about who funds improvements and carries underperformance risk. Gain-share only helps where the procurement and contract permit it, with the baseline, verification, payment timing and protection of service standards settled. Otherwise it is an argument about arithmetic postponed until after award.

Other benefits need their own accounting. Fewer breakdowns may avoid costs, while released staff time may create capacity without reducing the payroll. A Social Value financial proxy estimates wider benefit; it is not cash available to the buyer, and should not be added to budget savings as though both were spare money. We should also avoid counting the same improvement twice under different headings.

Where monetisation would be contrived, a measurable service or social outcome is more honest. We can show fewer repeat failures, reduced disruption or sustained employment without attaching a pound sign to every human advantage. Those benefits strengthen a submission where they meet the published criteria; an unsolicited spreadsheet does not create an extra scoring category.[5]

The writer’s contribution is to make this chain visible, while challenging its weak connections before the buyer has to. We listen for the real need, help articulate a deliverable response, establish the evidence and explain the benefit in terms the reader can evaluate. Quality earns its marks through that relationship between promise and substance, then earns its reputation when the building opens on Monday and the people inside can get on with their day.

Sources and editorial notes

Sources checked 23 September 2026. This section is supporting material, excluded from the article word count. Numbered references can become ordinary hyperlinks in the published version.

Government Facilities Management Standard 001: Management and Services, particularly sections 1.1–1.2. This government-estate standard informs the explanation; individual FM contracts have their own requirements.
PPN 002: Guide to using the Social Value Model, particularly sections 3.2–3.4 and the FAQ on corporate responsibility evidence. The central-government policy is not presented as universal across local government, private procurement or every UK jurisdiction. Under Price per Quality Point, the minimum is applied to the quality score; the guide also describes a limited market-maturity exception. Quantities are contractual commitments, but the model assesses the quality of the proposed method and plan.
Doshi and Hauser, 2024: Generative AI enhances individual creativity but reduces the collective diversity of novel content. The research concerns short-story writing. The application to FM tender differentiation is explicitly an editorial inference, not a demonstrated procurement finding.
PPN 017: Improving transparency of AI use in procurement, particularly paragraphs 9–10. AI use is not prohibited; the guidance discusses proportionate due diligence.
Procurement Act guidance: Assessing Competitive Tenders, particularly paragraphs 9–19 on criteria, published assessment methods and relative importance.

Personalisation: Replace the hypothetical heating example with a permission-cleared experience if you have one. The article claims no AAT delivery track record, guaranteed tender score or realised financial saving. Its “we” describes the editorial and bid-development approach. The financial illustration assumes no upfront capital expenditure and is net of additional recurring delivery costs before gain-share; real investment costs must be included in the actual appraisal.

Illustration: Constructivist estate forms with a shared geometric tree canopy and visible branching foundations: service quality, community benefit and evidence supporting the same place. Vermilion, near-black, warm ivory and restrained print grain, matching Issue 02.