A regional GC with a strong balance sheet decided to get serious about AI, and in year one they spent big: a six-figure enterprise platform license, a new Director of VDC and AI, a 4D scheduling subscription across every project, and a vision-capture program on the whole portfolio. The launch was impressive, the executive team was proud, and the first wins were real. Then year two arrived with the renewals, and year three arrived with nothing left to sustain it: the capital had been spent as if AI were a one-time purchase, the operating budget had no line to carry the subscriptions and the people, the Director left because there was no second-year plan to keep them, and the platform that cost a fortune to stand up quietly lapsed because the firm had funded the launch and not the program. They had treated a multi-year program as a single expensive year, and the program died of underfunding in the year it should have been compounding. This lesson is about the leader's job that prevents that outcome: building the multi-year AI investment plan that sequences capital, operating expense, talent, and partnerships against the roadmap and the firm's readiness, an honest, risk-adjusted projection rather than a hype-funded launch, ending in the named artifact, the five-year AI investment plan.

AI Is a Program, Not a Purchase

The first and load-bearing idea is that AI for an AEC firm is not a one-time purchase but a multi-year program, and that difference separates the firm that compounds from the firm that lapses. A purchase is funded once: you buy the thing, stand it up, and you are done. A program is funded continuously: it has a launch, but also renewals, a second cohort of people to train, a third year of tools that mature and replace the first year's, and a steady operating cost that does not end. The firm that funds AI as a purchase pays for the launch and starves the program, exactly the failure the lead describes: the big year-one spend with nothing left to sustain it by year three, because the money went into standing up the platform and none was reserved for carrying it.

The controlling analogy is the building itself. A building is not a capital project that ends at the ribbon-cutting; it is a capital project followed by an operating life. The capital builds it, but the operating budget runs it: the utilities, the maintenance, the staff, the renewals of the roof and the chillers across the building's life. A firm that built a hospital and funded only construction, leaving no operating budget, would have an empty expensive shell, and that is precisely what the year-one AI firm built. AI is the same: the capital stands up the tools and infrastructure, but the operating budget runs the program across its life, and the value, like the building's, accrues over the operating life, not at the launch. So the leader's first job is to stop thinking of AI as a thing to buy and start funding it as a program across years, with both the capital that builds it and the operating budget that runs it.

The Four Dimensions of the Investment

A multi-year AI investment has four dimensions, and the plan must address all four because underfunding any one starves the whole. The four are capital (the tools and infrastructure), operating expense (the subscriptions and the people), talent (whether to hire, build, or partner for the capability), and partnerships (the vendors, universities, and joint-venture partners that extend what the firm can do alone). These are not a menu but the four legs of the program, and a plan that funds the tools but not the people, or the people but not the partnerships, has a leg missing, which is why the year-one firm collapsed: it funded the capital launch and the first hire, but not the operating expense to carry them or the talent strategy to retain them.

Capital is the spend that builds the capability: the enterprise platform license stood up, the vision-capture hardware, the compute or data infrastructure, the integration work that connects the AI tools to the existing stack (Procore, ACC, P6, the ERP). It is often the visible headline spend, the number the year-one firm was proud of, but it is the smallest part of the multi-year cost because it is the one-time stand-up, not the recurring run. Operating expense is the recurring run: the subscriptions that renew every year (the platform, the 4D scheduling, the vision capture, the contract-review tool), and the people who operate the program, the salaries that recur and grow as the program scales. Talent is how the firm gets the capability, by hiring, building from the existing team, or partnering, each with a different cost and multi-year profile. And partnerships are the relationships that extend reach: the vendor relationships that shape the roadmap, the university relationships that feed the talent pipeline, and the JV partner relationships that share the AI investment on shared pursuits. The plan must sequence investment across all four.

Capital and Opex: The Build-Versus-Run Split

The most common and most expensive mistake in AEC AI funding is conflating capital and operating expense, treating the whole program as a capital project that ends, when the larger and more durable cost is the operating run that does not. The leader must split the two explicitly, because they are funded, governed, and reported differently. Capital is the build: the one-time stand-up cost of the tools and infrastructure, funded from the capital budget, often as a project with a defined end. Operating expense is the run: the recurring annual cost of the subscriptions and people, funded from the operating budget, every year, for as long as the program runs.

The build-versus-run split matters because the run is the larger multi-year number and the one that kills underfunded programs. A platform that costs a meaningful sum to stand up costs a comparable or larger sum every year to license, and the people who operate it cost their salaries every year, so over a five-year horizon the operating expense dwarfs the capital even though the capital is the headline. The year-one firm funded the headline build and left the operating expense unfunded, so the program had no money to run, the structural reason it lapsed. The discipline is to reserve the operating budget before spending the capital, because a capital build with no operating budget to run it is the empty hospital, worse than no build at all because it spent the capital and got no operating life. So the leader's discipline is to fund the run, not just the build, and to size the operating budget for the full multi-year horizon, not just the launch year, because the run is where the program lives or dies.

AI is a building, not a ribbon-cutting: the capital builds it, but the operating budget runs it across its life, and the program's value accrues over the operating life, not at the launch. The firm that funds the build and starves the run has built an expensive shell, the empty hospital with no operating budget, which is why the multi-year plan must reserve the operating expense before it spends the capital.

Talent: Hire, Build, or Partner

The talent dimension is how the firm gets the AI capability it needs, and the leader has three options, each with a distinct multi-year cost and risk profile: hire the capability (bring in a Director of VDC and AI, a data lead, an AI compliance lead from outside), build it from the existing team (train the PMs, supers, estimators, and VDC coordinators the firm already has into AI-augmented operators), or partner for it (use a vendor's professional services, a consultancy, or a university relationship to supply the capability the firm does not hold). Most firms need a mix sequenced against the roadmap and the readiness, because the right answer changes over the multi-year horizon.

Hiring buys the capability fast but carries the multi-year cost of the salary and the key-person risk the year-one firm hit: a single Director hired with no second-year plan to retain or support them leaves, and the capability leaves with them. Building develops the capability durably but slowly, because training the existing team into AI-augmented operators takes time and a training budget and competes with billable work, so it is the slower, cheaper, more durable path that pays off in the later years. Partnering supplies the capability without the headcount, using a vendor's services or a university relationship, which is fast and flexible but does not build the firm's own muscle, so it is the bridge while the firm builds, not the permanent answer. The multi-year talent plan typically partners and hires early to get capability fast, then builds the internal capability over the middle years so the firm holds the muscle by the later years, sequencing the three so the firm is not permanently dependent on the partner or the single hire, and the capability compounds into the team rather than walking out the door.

Partnerships: Vendors, Universities, and JV Partners

The fourth dimension, partnerships, is the one leaders most often leave out of the investment plan, and it is the one that extends what the firm can do beyond what it can fund alone. There are three kinds of partnership in the AEC AI program, each serving a different purpose. Vendor partnerships are the relationships with the tool providers (Autodesk, Procore, ALICE, Augmenta, Document Crunch under Trimble, OpenSpace), which at the enterprise level are not just purchases but relationships that shape the roadmap: a firm that partners deeply with a vendor influences the product, gets early access, and shares the integration cost, while a firm that merely subscribes takes what it is given. University partnerships are the relationships with the schools and trade programs (the ACE Mentor pipeline, the ABC and AGC Edge programs, the DeWalt and ABC Central Florida training pilot) that feed the talent pipeline and supply research capability the firm cannot staff alone. JV partnerships are the relationships with the other firms on shared pursuits, where the AI investment can be shared across the joint venture on a large pursuit, so neither partner carries the whole cost alone.

Partnerships belong in the investment plan because they change the cost and the reach of the program. A vendor partnership can reduce the capital and operating cost through enterprise terms and shared integration, and shape the roadmap so the firm's tools mature toward its needs rather than away from them. A university partnership reduces the talent cost by feeding the pipeline and supplying research the firm would otherwise hire for, addressing the build-talent dimension from outside. A JV partnership shares the AI investment on a pursuit, so the firm can field an AI-enabled bid on a large project without carrying the full investment alone, the leverage that wins the pursuit the firm could not afford to enable by itself. So the plan must name the partnerships, not leave them implicit, because they make the four-dimensional investment affordable and reachable; the firm that funds capital, opex, and talent but ignores partnerships pays more and reaches less. The multi-year plan sequences the partnerships alongside the spend they reduce.

Sequencing Against the Roadmap and Readiness

The four dimensions are not funded all at once; they are sequenced over the multi-year horizon against the roadmap and the firm's readiness, where this lesson ties back to the L4 roadmap and business-case lessons. The roadmap (built in the L4 roadmap lesson) sets the order in which the firm pursues AI capability by business unit, and the readiness audit (the L4 readiness lesson) sets how fast the firm can absorb each step, so the investment plan funds the roadmap in the order the readiness allows, not faster. A firm that funds the whole program in year one, ahead of its readiness, builds the expensive shell because it has tools and people the organization is not ready to use: the year-one firm spent at the pace of its ambition, not its readiness, and the spend outran the organization's ability to absorb it.

Sequencing against the roadmap means the investment follows the prioritized use cases, funding the high-margin-impact ones first (the ranked priorities from the L4 prioritization lesson) so the early investment pays back and funds the later, rather than spreading the year-one capital across everything and getting return on nothing. Sequencing against the readiness means funding the next step only when the firm is ready to use the last one, so tools and people are deployed into a ready organization rather than an unready one. The two together produce a phased plan: the early years fund the highest-priority, highest-readiness use cases (often estimating and administrative work, where the AGC 2026 AI Construction Outlook found adoption concentrated, at 23% and 45% respectively), proving the return and building readiness; the middle years fund the next tier as readiness grows and the early returns fund them; the later years fund the ambitious, lower-readiness use cases the firm is by then ready for. The investment compounds, each year's return and readiness funding and enabling the next, rather than a single front-loaded spend that outruns the organization and lapses, the discipline that separates the compounding firm from the year-one firm.

The Honest, Risk-Adjusted Projection

The multi-year plan is a projection of cost and return over five years, and the leader's discipline is to make it honest and risk-adjusted rather than a hype-funded projection that promises more than it can deliver. A hype projection assumes every tool works as the vendor demo promised, every use case pays back on schedule, and every year's return funds the next, the projection that gets the year-one budget approved and the year-three program defunded when the returns do not materialize. An honest projection assumes some tools underperform, some use cases pay back slower than hoped, and the returns are a range rather than a point, the projection that survives the inevitable shortfall because it budgeted for it.

Risk-adjusting means three things. First, the returns are projected as a range with a probability, not a single number, so the plan reports the risk-adjusted return (the L4 business-case discipline applied to the multi-year horizon) rather than the optimistic point that will not be hit. Second, the plan reserves a contingency for the tools that underperform and the use cases that pay back slower, so a shortfall in one area does not defund the whole program, the reserve the year-one firm did not have. Third, the plan stages the spend so the later years' investment is conditioned on the earlier years' results, with go and no-go gates (the operationalization discipline from the L5 pilot lesson) so the firm escalates the spend on what works and stops it on what does not, rather than committing the whole five years' spend up front on year-one optimism. The honest projection gets the program through the year-three valley where the hype projection dies, because it told the truth about the risk, reserved for the shortfall, and staged the spend against the results, so the program is funded by what it actually delivers. The leader's discipline is honesty over hype, because the hype gets the launch funded and the honesty gets the program sustained.

The Applied Problem: Produce the Five-Year AI Investment Plan

Here is the exercise. Produce the five-year AI investment plan for your firm: the multi-year plan that sequences capital, operating expense, talent, and partnerships against the roadmap and the readiness, as an honest, risk-adjusted projection. The plan is the leader's instrument for funding AI as a program rather than a purchase, and it is what prevents the year-one firm's failure.

Produce three things. First, the four-dimensional investment by year: the capital (tools and infrastructure stand-up) split explicitly from the operating expense (the subscriptions and people that recur), with the operating budget reserved for the full five-year horizon and not just the launch; the talent sequence (hire, build, or partner) across the five years, guarding against the key-person risk; and the named partnerships (vendor, university, JV) with the spend each reduces or the reach each extends. Second, the sequencing against the roadmap and the readiness: which use cases fund in which years, in the order the prioritization sets and the readiness allows, so the early returns fund and enable the later investment. Third, the risk-adjustment: the returns as a range with a probability rather than an optimistic point, the contingency reserved for the tools and use cases that underperform, and the go and no-go gates that condition the later years' spend on the earlier years' results.

The deliverable is the five-year AI investment plan, the leader's instrument for funding AI as a multi-year program that compounds rather than a one-time purchase that lapses. This is the third lesson of the AEC transformation playbook, and it ties the L4 roadmap, readiness, prioritization, and business-case lessons into a single funded plan across the four dimensions. The leader who masters it funds the build and the run, sequences the talent so the capability compounds into the team rather than walking out the door, names the partnerships that extend the investment, sequences the spend against the roadmap and the readiness so the early returns fund the later years, and projects candidly so the program survives the year-three valley. That is the only way an AEC AI program becomes the compounding capability the firm intended rather than the expensive shell the year-one firm built, because AI is a building, not a ribbon-cutting, and the leader's job is to fund its operating life, not just its construction.

Key Takeaways

• AI for an AEC firm is a multi-year program, not a one-time purchase: it has a launch but also renewals, second cohorts, and a steady operating cost, so the firm that funds the launch and starves the program builds the expensive shell that lapses, the big year-one spend with nothing left to sustain it by year three.
• The investment has four dimensions that must all be funded: capital (tools and infrastructure), operating expense (subscriptions and people), talent (hire, build, or partner), and partnerships (vendors, universities, JV partners), and underfunding any one starves the whole, which is why the year-one firm's launch collapsed.
• The capital-and-opex split is the most expensive thing to get right: capital is the one-time build, operating expense is the recurring run, and over five years the run dwarfs the build, so the plan must reserve the operating budget before it spends the capital, because a build with no run is the empty hospital with no operating budget.
• The talent dimension is a sequence, not a single choice: hire and partner early to get capability fast, build the internal capability over the middle years so the firm holds the muscle by the later years, sequencing the three to guard against the key-person risk that the year-one firm proved is fatal.
• Partnerships are the leverage dimension leaders most often leave out: vendor partnerships shape the roadmap and share integration cost, university partnerships feed the talent pipeline, and JV partnerships share the AI investment on a pursuit, so the plan names them because they make the four-dimensional investment affordable and reachable.
• The four dimensions are sequenced against the roadmap and the readiness, tying back to the L4 roadmap, prioritization, and readiness lessons: fund the high-priority, high-readiness use cases first so the early returns fund and enable the later investment, rather than a front-loaded spend that outruns the organization's ability to absorb it.
• The projection must be honest and risk-adjusted, not hype-funded: returns as a range with a probability rather than an optimistic point, a contingency reserved for the tools and use cases that underperform, and go and no-go gates that condition the later years' spend on the earlier years' results, because the honest projection survives the year-three valley where the hype projection dies.
• The deliverable is the five-year AI investment plan, the leader's instrument for funding AI as a compounding program rather than a lapsing purchase, funding the build and the run, sequencing the talent so the capability compounds into the team, naming the partnerships, and projecting candidly, because AI is a building, not a ribbon-cutting, and the leader's job is to fund its operating life.