Don't Build the Model. Be the Reason It Works.

Every consulting firm with a website has an AI practice page now. Fewer of them have a real answer to the harder question underneath it: build the capability yourself, or plug into someone else’s. That choice gets treated like a branding decision. It isn’t. It’s a strategy decision with a five-part test attached, and most firms are failing at least two of the five parts without realizing it.

The test is the 5C framework: Company, Customers, Competitors, Collaborators, Context. It was built for exactly this kind of question, the one where “everyone else is doing it” is doing the thinking that a real strategic read should be doing instead. Run a typical mid-size or boutique consulting firm through all five, and the honest verdict is usually the same: going all-in on building an AI practice is not a sustainable bet. Becoming a services partner to the firms actually building AI is.

Company: What You’d Actually Be Competing With

Start with the plainest question: what does “building an AI practice” cost at a level that’s actually competitive, and do you have it? Accenture answered that question with a $3 billion, three-year commitment to its Data & AI practice, aimed at doubling its AI-focused workforce from 40,000 to 80,000 people, roughly a tenth of the entire firm.¹² McKinsey’s QuantumBlack runs around 5,000 specialists. BCG X runs around 3,000 engineers. Bain equipped its entire eighteen-thousand-person team with AI tooling.⁵

That’s the entry price at the top of the market. Most consulting firms, including most firms who’d describe themselves as serious players in their industry, are not writing $3 billion checks or fielding five-thousand-person AI units. Measured against that Company reality, “build our own AI capability” usually means entering a fight several years and several orders of magnitude behind, funded out of a budget that was never sized for it.

Customers: What They’re Actually Buying

Clients watching competitors deploy AI want proof and speed, not a briefing on your firm’s hiring plan. The argument that still holds up for consulting’s relevance, even as the underlying models keep improving, is what one AI-focused consultant calls the “last mile”: AI vendors sell models and platforms, but turning that capability into a working outcome inside a specific business still requires process redesign, data integration, and change management that a model alone doesn’t do.⁷ That’s the layer clients are actually paying for. Whether the model underneath it came from your own lab or from a partner’s platform is, to most clients, invisible and beside the point.

That reframes the whole question. If the client’s real purchase is judgment applied to someone else’s model, then building your own model is solving a problem the client didn’t ask you to solve.

Competitors: A Fight You Don’t Need to Enter

The Competitors read is where “build it ourselves” stops looking like ambition and starts looking like a bad matchup. The competitive set for a from-scratch AI build isn’t just other consulting firms. It’s the hyperscalers and frontier labs themselves, who are increasingly selling directly to enterprises, and it’s the handful of consulting giants who already have multi-year, multi-billion-dollar heads starts. Trying to out-build both at once, on a fraction of the capital, isn’t a strategy. It’s a way to lose slowly on two fronts instead of one.

Collaborators: What the Giants Actually Do

Here’s the detail that should carry the most weight in this whole analysis: even the firms with the capital to build extensively don’t rely on building alone. Bain became the first major consulting firm to strike a direct alliance with OpenAI, announced in February 2023 and deepened repeatedly since, including a stake in OpenAI’s enterprise deployment venture.³⁴ BCG holds parallel alliances with both OpenAI and Anthropic. Deloitte’s delivery ties span Google Cloud, AWS, Anthropic, Nvidia, and ServiceNow. McKinsey lists over a thousand ecosystem partners, including AWS, Google Cloud, Microsoft, and Salesforce.⁵

Every one of those firms is layering partnerships on top of its internal build, not instead of it, and in several cases the vendor alliance is the more visible and more publicized move. If the firms with the deepest pockets in the industry still choose to partner extensively rather than build everything themselves, that’s not a footnote. It’s the answer. A smaller firm trying to skip straight to “we built our own” is trying to leapfrog a step the biggest, best-capitalized players in the business haven’t chosen to skip.

Worth naming honestly here too: these alliances create their own bias risk. A firm that both advises on AI strategy and implements a specific vendor’s platform has an incentive to recommend the platform it profits from reselling, a tension independent advisors have pointed out directly.⁸⁶ That’s a real cost of the partner path, not a reason to avoid it, but a reason to be transparent with clients about where the recommendation is coming from.

Context: Which Way the Curve Is Bending

Zoom out to the macro forces nobody controls. Frontier model capability keeps getting cheaper and more commoditized on a roughly annual cycle, which means whatever proprietary edge a build effort produces this year is closer to a baseline feature by next year. Model training itself keeps getting more capital-intensive at the frontier, which raises the ante for anyone trying to compete at that layer specifically. Specialized AI talent remains expensive and scarce relative to how consulting firms have traditionally staffed and billed. And the regulatory environment around AI is still forming, which makes committing years of capital to a specific in-house build a bet on rules that haven’t been written yet.

Every one of those forces points the same direction: toward renting capability from whoever is absorbing that capital intensity and regulatory uncertainty, and toward building your differentiation in the layer above it, not in the model itself.

The Cost of Getting This Call Wrong

Everything above is a strategic argument. The stakes underneath it are also a straightforward cost problem, and the data on that is now specific enough to quantify.

RAND Corporation interviewed 65 experienced data scientists and engineers across company sizes and industries and found that more than 80 percent of AI projects fail outright, roughly twice the failure rate of non-AI IT projects.⁹ Gartner separately predicted that at least 30 percent of generative AI projects would be abandoned after proof-of-concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs, and unclear business value as the recurring reasons.¹⁰ Neither number is about AI failing as a technology. Both are about organizations misjudging what they could actually execute, which is exactly the assessment this article is arguing through.

The build-versus-partner split inside that failure rate is the number that matters most for this specific decision. MIT’s NANDA initiative studied more than 300 enterprise AI deployments in 2025 and found that tools sourced through external partnerships reached deployment about 67 percent of the time, compared to roughly 33 percent for tools built in-house, twice the success rate.¹¹ That’s not a rounding difference. It’s the same 5C conclusion showing up as a hard number: betting on your own build isn’t just the more capital-intensive path, it’s the statistically worse bet on its own terms.

The opportunity cost compounds the direct loss. A survey of 700 senior business leaders found that U.S. organizations lose an average of 2.4 percent of annual revenue on AI initiatives that fail to deliver expected value, and fewer than a third of those organizations treat killing an underperforming AI project as a normal, timely decision. Nearly half only pull the plug after significant time and money are already spent.¹² Every quarter a firm spends staffing and funding a build that was never going to out-compete the vendors already ahead of it is a quarter not spent on the implementation relationships and delivery capability, the actual partner-side capacity, that the same data says wins twice as often. Misjudging this call doesn’t just cost the money sunk into the build. It costs the position the firm could have had if it had assessed correctly from the start.

The Actual Strategy: Support the Race, Don’t Try to Win It

Put the five C’s together, cost data included, and the picture is consistent, not just convenient. Company says you likely don’t have the capital to build competitively. Customers say they’re buying outcomes, not model ownership. Competitors say the build lane is already occupied by players with a multi-year, multi-billion-dollar lead. Collaborators show that even those players choose to partner extensively rather than build alone. Context says the economics of owning frontier AI keep getting worse, not better, for anyone outside the very top of the market.

None of that is an argument against AI. It’s an argument against the specific bet of trying to become an AI vendor when your firm is, and has always been, a services business. The available and genuinely sustainable move is to become the best possible implementation partner to the vendors who are actually racing to build the underlying technology: the process redesign, the change management, the industry-specific judgment, the last mile that turns someone else’s model into a working result for a specific client. That’s not a consolation prize. It’s the same role systems integrators have played in every previous platform wave, and it’s a role with real, durable demand precisely because the vendors racing to build frontier models have no interest in doing it themselves.

Key Takeaways

Key Frameworks:

• 5C Analysis (Company, Customers, Competitors, Collaborators, Context) – a strategic lens for deciding whether to build a capability or partner for it, rather than defaulting to whichever one looks more impressive on a slide.

Try It: Write one honest sentence for each of the five C’s about your own firm’s AI position, no hedging. If three or more of those sentences point toward “we can’t out-build this,” that’s your answer, and it’s time to go looking for the vendor partnership that plays to what your firm actually does well.