Your organization’s project archive is full. Are you any wiser for it?

What is your data strategy after the project ends?

You have already paid for years of design information. How much of it can your organisation use without reopening the original project?

Ask whether that information is safe, and someone will show you a backup schedule. Ask what it tells you about the last five offices you designed, and you may be shown the backup schedule again, slightly more firmly.

Retention is legitimate work, and the people doing it have done exactly what they were asked. But after another hundred projects, what has become easier for the organisation to answer? “We have another hundred folders” accurately describes the growth. But it is not an account of what the organisation has learned.

That gap matters more now than it did five years ago. Any AI initiative that depends on project history will only be as reliable as the evidence underneath it, and much of your best evidence sits in an archive designed to keep it exactly as it is.

Archive the model. Activate the information. A useful strategy starts by deciding which questions your completed work should help answer, then establishing what it takes to answer them reliably. Here are three worth asking.

1. What did we deliver on the last five projects like this?

A workplace team is preparing its next fit-out and wants to compare space programmes, quantities and layout decisions across five similar projects. Without a reuse process, that means locating files, opening models, extracting data and finding the colleague who remembers the exceptions. The colleague is free on Thursday. The brief is due on Wednesday.

I worked on a single masterplan for 18 years. For at least a decade of that, through design into construction, I became the design guidelines personified. I remembered not just what we had decided, but which competing objectives we had been trying to reconcile. I would like this understood as a tribute to professional judgement, although “was there at the time” undermines the grandeur somewhat. That experience is part of why this matters to me.

A drawing shows what was decided. It rarely shows why, which alternatives were rejected, or what made an exception acceptable. Being the person who can supply that context is useful. Depending on one person to supply it is a risk: it leaves the organisation with a knowledge system that has to be invited to meetings, and may not always be available.

With the relevant information extracted and comparable, the team could review all five projects before writing the brief. The colleague could spend Thursday explaining why decisions were made, not where the files are.

What to measure: The time it takes today to assemble and verify that comparison, including searches and handoffs. Then test the same question against extracted information.

2. Which past refurbishments should shape our next capital plan?

An owner preparing a refurbishment programme wants to find buildings with similar uses, layouts and systems, then compare previous interventions, actual costs and subsequent performance. Those records usually exist, scattered across project archives, cost reports and facilities systems. So assembling the evidence becomes a project in itself before anyone can assess it.

A reuse strategy puts recorded building characteristics and past interventions alongside current condition surveys and operational data. The planning team can then identify relevant precedents, test assumptions and decide where fresh investigation is needed. This is an established way to approach portfolio decisions: the Department for Education uses condition data, verified with additional information, to prioritise England’s school rebuilding programme. Department for Education, programme guidance.

To be clear, old BIM models alone cannot rank refurbishment needs. A completed model records the building as it was understood at handover. Since then, a meeting room may have become a store, the store may have become a meeting room, and both may still be called “Meeting Room” because changing the booking system was harder than moving the wall. The question for your archive is what baseline can be recovered and connected to current evidence.

3. What happened on comparable jobs, and what should that mean for this estimate?

A cost consultant preparing a proposal wants to match previous costs to comparable design quantities and scope. Without a repeatable process, that means model extractions, cost-report searches and checks on what each project actually included. Often an allowance has to be set before the comparison is finished.

Connecting design information to estimates, final costs and recorded changes gives the consultant evidence to challenge that allowance or investigate an unusual quantity. HM Treasury’s Green Book supports the underlying comparison: using historical forecast errors and comparable projects to adjust for optimism bias. HM Treasury, The Green Book (2026)

The model supplies the design parameters. Financial records and scope changes supply the rest. The wall has many properties, but recollection of the final account is rarely one of them.

Why AI raises the stakes

All three questions share a pattern: the answer exists, but it is locked inside files that only open in the software that made them, interpreted by people who were there. That pattern is also the main obstacle to useful AI in the built environment.

An AI assistant asked “what did we deliver on the last five projects like this?” can only be as good as the information it can reach and the definitions it can trust. Point it only at a folder of native files and you risk confident answers to the wrong question. Give it extracted, mapped, source-linked project data and it can do the searching, so your experts can do the judging.

This is why the post-project data strategy is now an IT leadership question, not only a records-management one. The organisations that will get value from AI over the next three years are not necessarily those with the biggest archives. They are the ones that have made their archives answerable.

What makes reuse worth paying for?

Start with two numbers: how often the question recurs, and how much effort answering it takes today. An archive of 10,000 projects is impressive on a slide. It may also contain 10,000 projects nobody needs to ask about this year. The useful questions are which decision recurs, who needs the answer, and what getting it currently involves.

One customer gave us a challenge that belongs in every efficiency pitch. A 20% efficiency gain, they said, would not justify investment on its own. Free up one day in five, and it never arrives as a vacant Friday. It turns up in fragments and is absorbed into more meetings and more ambitious tea-making.

Time saved is not the outcome. Name the consequence instead; perhaps it’s a comparison finished before the brief is fixed, an estimate checked before submission, or a specialist freed to review another project. Then name who will make it happen.

An illustrative business case

Suppose teams answer 12 historical information questions a month, each taking three hours to search, extract and check. Customers describe this task as taking anything from an hour to a working day, so three hours is a middle case, not an average. If a checked answer took 15 minutes instead, that would free up 396 staff hours a year.

Assume a loaded staff cost of $70 an hour, that half the released time becomes productive capacity, and a first-year cost of $5,000 covering setup, extraction, mapping, checking, tools and maintenance. Frequency then decides the case:

Capacity value = questions × 12 months × 2.75 hours × 50% × $70. Illustrative assumptions, not customer results.

Two assumptions carry the weight. The 50% conversion rate is a figure to test, not a standard: if none of the released time becomes useful work, the capacity case falls away. And capacity is not cash. The UK Government Efficiency Framework treats staff time released by an IT system as a non-cash benefit unless spending actually falls. HM Treasury and Government Finance Function, Government Efficiency Framework, section 4

For a CIO, the more important point is that this is one question type. The case scales with the number of recurring questions worth answering, not with the size of the archive. Finding those questions, and testing the assumptions against your real workflow, is where the work begins.

Is our historical data actually usable?

Some questions will be answerable, some will require additional records, and some will expose gaps worth fixing. You can discover which is which without waiting for the archive to be perfect. An archive built across years, teams and changing standards will be inconsistent. One project calls a field “Room Type”, another “Space Function”, a third just “Type”. There was probably a meeting about this. There may have been several, each producing the correct answer for the people in the room. You have inherited all the correct answers.

A naming convention that changed halfway through the decade does not erase what was recorded under it. The quantities, spatial relationships, materials and decisions are still there. The practical approach:

• Start with one defined question and a manageable sample of projects.
• Establish what the fields mean and reconcile only the differences that matter to that question.
• Keep original values alongside mapped ones, expose uncertainty, and preserve the link to source and version.

IStructE’s AI implementation guidance raises the same practical concerns about historical datasets: ownership, access and data quality. Comparability also decays with time. Research on case-based cost estimation identifies changing prices, construction methods and economic conditions as reasons historical cases need ongoing maintenance. Xiao, Improving the long-term use of case-based reasoning model in early construction cost estimation (2021)

The alternative is to make the whole archive uniform first. That turns every new data standard into another reason the organisation can't consult old projects, and makes it seem as though the organisation began accumulating knowledge last Tuesday.

Make tomorrow’s information easier to use

While you extract value from completed work, change what enters the archive next. Decide which future questions matter, define the information they need, and validate it at agreed points during delivery, while project teams can still explain inconsistencies and close gaps.

The folder passed inspection; the answer remained under investigation. “We’ll have better data in future” is a reasonable ambition. Repeated every year, it lets the future keep its excellent record of never arriving.

Where to start

You already retain the work. The old projects still contain information you paid for, including everything from before last Tuesday. The strategic choice is whether that archive stays a compliance cost or becomes something your organisation, and its AI, can learn from.

Pick one recurring question worth answering. Establish what your existing information can support. Measure whether the answer justifies the effort. Then fix the inputs so the next answer is cheaper.

Archive the model. Activate the information.

Book a workshop. In one session we will help you identify the question, select a sample of completed projects, and build a business case from your actual workflow and numbers rather than ours. Bring a few projects and, if you can, the colleague who remembers everything. We can work around Thursday. The aim is to preserve what they know without keeping the whole person for another eighteen years.