A regional general contractor signs a six-figure enterprise agreement for an AI estimating platform, rolls it out to every group at once, and eighteen months later the spend is sunk: preconstruction loves it, field operations never opened it, the design-build studio fought it, and the self-perform concrete crews could not feed it the data it needed. The money was not wasted because the tool was bad. It was wasted because the firm treated five different businesses as one, and asked all of them to adopt the same capability on the same day regardless of whether they were ready or whether the win was worth chasing. A roadmap would have caught this. A roadmap sequences AI across the firm by readiness and by margin, so the dollars go first to the unit that is ready and where the win is large, and last to the unit that is neither. This lesson builds that roadmap. You walk out with a phased 12, 24, and 36-month plan, organized by business unit, with named investment milestones, that tells your principals where the money goes next and why, so the firm stops boiling the ocean and starts sequencing the wins.

The Expensive All-At-Once Rollout

The failure above is the default failure of firm-level AI, and it is expensive in a specific way. The license fee is the visible cost, but it is the smallest one. The larger cost is the organizational scar: a unit that was forced to adopt a tool it was not ready for now believes AI does not work, and the next initiative, the one that would have landed, meets a workforce that has already decided. The firm spent money to make its own people skeptical. That is the most expensive line item, because it does not show up on the invoice and it compounds across the next three initiatives.

The root cause is a category error: treating the firm as a single buyer with a single readiness when it is a portfolio of businesses, each with its own data, its own workflows, its own people, and its own margin structure. The preconstruction group lives in estimates and proposals. The field operations group lives in daily logs, RFIs, and schedules. The design studio lives in models and drawing sets. The self-perform trades live in production rates and crew data. The service and facilities arm lives in work orders and asset histories. These are not the same business wearing five hats. They are five businesses that happen to share a logo, and an AI capability that transforms one of them may be irrelevant or premature in another.

So the all-at-once rollout fails because it ignores the two variables that should govern every AI investment decision: how ready is this unit to absorb this capability, and how large is the margin win if it lands. The readiness audit, the lesson immediately before this one, gave you the first variable unit by unit. The roadmap builds on it and answers the sequencing question the audit could not: given finite money and attention, which unit goes first, which goes second, and which waits, so the firm captures the ready, high-margin wins before it spends a dollar on the unready ones.

The Roadmap as a Capital Plan

The controlling analogy is one every principal already understands: the capital plan. A firm with finite capital does not buy every piece of equipment it might someday use on the same day. It sequences the purchases by which asset earns its keep soonest and most certainly: the excavator the firm is utilizing at ninety percent goes before the specialty rig that books two weeks a year, even if the specialty rig is more exciting. The firm phases the spend, names the milestones, and revisits the plan as conditions change. An AI roadmap is a capital plan for capability rather than for iron, and it obeys the same discipline.

Under this analogy, each business unit is a place you can deploy capital, and each AI capability is an asset with an expected return and a readiness-to-deploy. The readiness is whether the unit can actually operate the asset: does it have the data the capability needs, the workflows the capability plugs into, and the people who will run it. The margin impact is the return: how much the win is worth if the capability lands, in recovered margin, won work, or avoided loss. A capital plan that ignored utilization and return would buy the wrong equipment; a roadmap that ignores readiness and margin funds the wrong unit. The two variables together set the sequence, exactly as utilization and return set the order of equipment purchases.

The analogy also fixes the cardinal discipline of this lesson: do not boil the ocean, sequence the wins. No firm buys its entire equipment fleet in one quarter, and no firm should deploy AI to every unit at once. The capital plan is phased because capital and absorption are both finite, and the AI roadmap is phased for the same reasons: money is finite, and a unit can only absorb so much change at once before the change itself becomes the bottleneck. The roadmap's job is to phase the deployment so each unit gets the capability it is ready for, when it is ready, in the order that captures the most margin soonest.

Five Units, Five Readiness Profiles, Five Fits

The reason a roadmap cannot be a single line is that the five units differ on all three readiness dimensions and on AI fit, so each demands a different first capability and a different pace. Preconstruction is typically the readiest and the highest-fit unit, which is why it is so often the firm's first win. Its data is structured (historical bids, unit costs, takeoff quantities), its workflows are bounded (estimate, qualify, propose), and the AI fit is direct: generative drafting for proposals, computer vision and generative quantity takeoff for estimating, predictive ML for bid-or-no-bid scoring. The margin win is large and fast because estimating accuracy and proposal throughput convert directly into won work and protected margin. Precon is usually where the roadmap starts.

Field operations is the highest-volume unit and often the largest margin pool, but its readiness is mixed. The data exists (daily logs, RFIs, submittals, photos, schedules) but it is unstructured and scattered, and the workflows are numerous and fast-moving. The AI fit is real but the absorption is harder: generative AI for RFI and submittal drafting, computer vision for progress capture and safety, predictive ML for schedule risk. The margin win is large because field operations is where projects are won or lost, but the readiness work (getting the data into usable shape, and getting field people to trust the output) is heavier, so field ops usually phases in after precon, with the data-readiness investment named explicitly.

The remaining three units stretch the range. Design, in firms that have it, is high-fit for generative design and model-checking but often lower-readiness because the work is bespoke and the verification stakes (the stamp, the design intent) are highest, so it phases carefully. Self-perform trades have rich production data but often the least mature digital workflow, so the readiness investment (capturing crew production cleanly) precedes any AI win. Service and facilities has the most structured recurring data (work orders, asset histories, COBie handoffs) and a strong predictive-ML fit for maintenance, but it is often the smallest margin pool and the lowest organizational priority, so it tends to sit later in the roadmap unless it is a strategic growth bet. Five units, five readiness profiles, five different first moves.

A firm-level AI roadmap is a capital plan for capability: it sequences each business unit by its readiness to deploy and the margin the win is worth, funds the ready high-margin units first and the unready ones last, and phases the spend across 12, 24, and 36 months so the firm captures the wins it can absorb instead of boiling the ocean all at once.

The Sequencing Logic: Readiness Times Margin

The roadmap's sequencing rule is a two-variable sort, and stating it plainly prevents the most common mistake. Each candidate move (a unit plus a capability) gets two scores: its readiness (can the unit deploy this now, derived from the readiness audit) and its margin impact (how large the win is if it lands). The sequence is set by both together, not by either alone. Sequencing by margin alone funds the highest-value win even when the unit cannot absorb it, exactly the all-at-once rollout that scarred the field crews. Sequencing by readiness alone funds the easiest deployment even when the win is trivial, spending scarce attention on a capability that moves no margin.

The product of the two is what orders the phases. A move that is both ready and high-margin (typically precon estimating) is a phase-one win: it lands fast and pays off large, so it goes first and funds confidence for everything after. A move that is high-margin but not yet ready (often field operations) is a phase-one readiness investment and a phase-two deployment: you spend the first phase making the unit ready (the data work) so the second phase can capture the large win. A move that is ready but low-margin is a cheap early win you can take opportunistically but should not prioritize over a high-margin readiness investment. A move that is neither ready nor high-margin waits, and naming it as deliberately deferred is itself a roadmap decision, because an unwritten "later" becomes an accidental "never" or an impulsive "now."

This is why the roadmap is built on the readiness audit and not instead of it. The audit produced the readiness score per unit; the roadmap multiplies it by the margin estimate and sorts, then lays the sorted moves across the three time horizons. The discipline it enforces is that every funded move can name both why it is ready and why the win is worth the spend, and every deferred move can name which variable is missing (the readiness it lacks or the margin it would not move), so the principals choose on the two variables that matter rather than on the loudest internal advocate or the flashiest demo.

The Three Horizons: 12, 24, and 36 Months

The three horizons are not arbitrary calendar slices; each phase has a distinct strategic job, and naming the job keeps the milestones honest. The 12-month horizon is the proof phase: it funds the ready, high-margin wins (usually precon, often one field-ops quick win) and the first readiness investments for the high-margin-but-not-yet-ready units. Its purpose is twofold: capture real margin early, and produce the internal proof that AI works here, the political capital that funds the later phases. A 12-month plan that funds only readiness work and no wins starves the firm of proof; one that funds only wins and no readiness investment leaves the second horizon with nothing teed up.

The 24-month horizon is the scale phase: it deploys the capabilities whose readiness the first phase built (field operations moving from data-readiness to live deployment), extends the proven phase-one wins deeper into their units, and begins the readiness work for the harder units (design, self-perform). The 36-month horizon is the maturity phase: it brings the slower, higher-stakes, or lower-priority units into the program (design model-checking at scale, self-perform production AI, service and facilities predictive maintenance) and shifts the firm from deploying point capabilities to integrating them across units, where the precon estimate feeds the field plan feeds the service handoff. Each horizon's milestones should name the unit, the capability, the investment, and the expected margin, so the roadmap is a set of commitments and not a wish list.

The horizons also encode the absorption limit. A unit that took a phase-one win is busy absorbing it through the first year and should not be handed a second major capability in the same window, so the roadmap staggers each unit's moves across horizons rather than stacking them. This is the capital-plan discipline again: you do not deploy three new rigs to one crew in one quarter, and you do not deploy three new AI capabilities to one unit in one year, because the unit can only absorb so much before the change itself becomes the constraint and the later capability fails for lack of attention rather than lack of merit.

Naming the Investment Milestones

A roadmap that says "improve estimating with AI in year one" is a slogan, not a plan, because no one can fund a slogan or hold it to account. A milestone is named when it specifies four things: the unit (precon), the capability (generative quantity takeoff plus bid-scoring), the investment (the platform, the integration work, the training, named with the order of magnitude of spend and the owner), and the expected margin outcome (faster, more accurate estimates converting to a measurable bid-win-rate or margin-protection target). A named milestone can be approved, budgeted, assigned, and measured; a slogan can only be admired.

Naming the investment, not just the capability, is what makes the roadmap a capital plan rather than a strategy deck. Every milestone carries a cost: the license or platform fee, the integration and data-preparation work, the training and change management, and the internal owner's time. For the high-margin-but-not-yet-ready units, the first named milestone is often a pure readiness investment with no immediate margin (the field-operations data-cleanup project that earns nothing this year but makes next year's deployment possible), and naming it with its own owner and budget keeps it from being quietly skipped because it does not show a return in its own phase. The roadmap names the unglamorous readiness milestones with the same rigor as the win milestones, because skipping them is what breaks the later phase.

Finally, each milestone names its owner, because an unowned milestone drifts. The principal owns the roadmap, but each milestone is assigned to a unit leader accountable for the deployment and the outcome, converting the roadmap from a firm-level aspiration into a set of personal commitments with names attached. This is also how the roadmap earns its review cadence: at each horizon boundary, the owners report against their named milestones, the readiness and margin assumptions are re-scored against what actually happened, and the next phase is re-sequenced on current data rather than on the original guess, so the roadmap stays a living capital plan rather than a one-time document that ages into irrelevance.

The Applied Problem: Build Your Firm's 12/24/36-Month Roadmap

Here is the exercise. Produce your firm's AI roadmap by business unit: a phased 12, 24, and 36-month plan with named investment milestones. Start from the readiness audit you produced in the prior lesson, which gave you a readiness profile for each unit you operate (some subset of preconstruction, field operations, design, self-perform, and service and facilities). For each unit, add the second variable: estimate the margin impact of the highest-fit AI capability for that unit, in the terms your firm tracks (won-work, recovered or protected margin, avoided loss). You now have a readiness score and a margin estimate per candidate move.

Sort the candidate moves by readiness times margin, then lay them across the three horizons using the phase jobs: the 12-month proof phase funds the ready high-margin wins (usually precon) plus the first readiness investments for the high-margin-but-not-yet-ready units (usually field operations); the 24-month scale phase deploys the readiness you built and extends the proven wins; the 36-month maturity phase brings in the slower or lower-priority units and begins integrating capabilities across units. Stagger each unit's moves so no unit is handed two major capabilities in one window, respecting the absorption limit. For every funded move, write the named milestone: unit, capability, investment (platform, integration, training, with order-of-magnitude spend and owner), and expected margin outcome. For every deferred move, name which variable is missing so the deferral is a decision and not an oversight.

The deliverable is the roadmap itself: a one-page-per-horizon plan, organized by business unit, with named investment milestones a principal can take to the partners and fund. The lasting product is a firm that sequences its AI spend by readiness and margin instead of buying one enterprise platform for everyone and scarring the units that were not ready. The principal who masters this stops boiling the ocean and starts sequencing the wins, funding the ready high-margin unit first, investing in readiness where the win is worth it, and deferring the unready low-margin moves on purpose, so every dollar of AI spend lands on a unit that can absorb it and a win worth chasing, and the next initiative meets a workforce that has seen AI work rather than one that has already decided it does not.

Key Takeaways

• The default firm-level AI failure is the all-at-once rollout: one enterprise platform forced on every unit, which wastes the license and, more expensively, scars the units that were not ready, making the next initiative meet a skeptical workforce. The roadmap prevents this by sequencing instead of broadcasting.
• A firm is not one buyer but a portfolio of businesses: preconstruction, field operations, design, self-perform, and service and facilities each have different data, workflows, people, and margin structures, so each demands a different first AI capability and a different pace.
• The controlling analogy is the capital plan: an AI roadmap sequences each unit by its readiness to deploy (data, workflows, people) and the margin the win is worth, exactly as an equipment plan sequences by utilization and return, and phased because both money and absorption are finite.
• The sequencing rule is readiness times margin: a ready high-margin move (often precon estimating) goes first; a high-margin-but-not-yet-ready move (often field operations) gets a phase-one readiness investment and a phase-two deployment; a ready low-margin move is an opportunistic cheap win; a neither move is deliberately deferred. Using either variable alone is the common mistake.
• The three horizons each have a job: the 12-month proof phase captures ready wins and earns the internal proof that funds later phases; the 24-month scale phase deploys the readiness built earlier and extends proven wins; the 36-month maturity phase brings in the slower units and integrates capabilities across units.
• The absorption limit staggers each unit's moves across horizons: a unit absorbing a phase-one win should not be handed a second major capability the same year, because the change itself becomes the constraint and the later capability would fail for lack of attention rather than merit.
• A milestone is named only when it specifies the unit, the capability, the investment (platform, integration, training, with order-of-magnitude spend and an owner), and the expected margin outcome, so it can be approved, budgeted, assigned, and measured; pure readiness investments are named with the same rigor as win milestones so they are not quietly skipped.
• The roadmap is a living capital plan: each milestone has an owner who reports at the horizon boundaries, where readiness and margin are re-scored against what actually happened and the next phase is re-sequenced on current data, so the plan stays current rather than aging into a one-time deck.