How is AI reshaping enterprise consulting and IT services, why do so many enterprise AI pilots fail to reach production, and which companies are best positioned to win by combining reusable platforms, forward-deployed engineers and accountability for outcomes—particularly in regu

AI is shifting enterprise consulting from selling teams and hours toward deploying repeatable software into a customer’s actual workflow, with engineers responsible for making it work. The strongest position is not “AI model provider” or “consultant” alone, but a firm that can combine a reusable operating layer, hands-on integration and a measurable production outcome. Recent partnerships built around forward-deployed engineers suggest that both software and services companies see that deployment gap as an opportunity.4

• Why pilots stall: A demonstration can run on selected data; production must connect to legacy systems, respect permissions, survive changing inputs, produce auditable decisions and fit how employees work. Projects also need an accountable owner and a baseline against which to measure savings or revenue. A widely repeated “95%” figure concerns measurable financial return in a particular study; it should not be treated as a universal production-failure rate.6

• Where the winners may be: Palantir is the clearest established example for complex, governed large-enterprise deployments: its reported U.S. commercial revenue grew 133% year over year in Q1 2026, though growth alone does not prove every deployment’s ROI.1 ServiceNow is well placed where AI can be added to an existing workflow and subscription relationship; its Q4 2025 subscription revenue grew 21% year over year.2 Accenture and other major integrators have the customer access, migration capacity and industry specialists to deliver at scale, but must turn repeated work into reusable assets rather than simply staffing more engineers. Accenture’s Google Cloud initiative explicitly combines forward-deployed engineers with industry expertise.4 For the mid-market, the most promising model is likely narrower: prebuilt connectors, a few repeatable industry workflows, fixed-scope implementation and managed operation. There is insufficient evidence here to name a definitive mid-market winner.

• Why regulated and legacy-heavy work matters: Integration, security controls, audit trails and change management make these deployments harder than a chatbot pilot—but also create room for a provider willing to own the working system, not just deliver a recommendation. Migration and regulatory-reporting workflows are examples of the outcome-oriented work being targeted.11

• Economics and lock-in: In headcount-based services, revenue generally requires more billable labor. A genuine platform can spread connector, governance and workflow-development costs across customers, so later deployments require less bespoke effort and can earn better incremental margins. That is a hypothesis to test, not an automatic consequence of calling revenue “software”: C3 AI, for example, reported a 32% GAAP gross margin in its July 2026 quarter despite its enterprise-AI software positioning.5 Switching costs rise when a system reliably runs important workflows and accumulates integrations, permissions and operating history; customers should distinguish that operational value from dependence created merely by custom code or difficult data export.

• The decisive metrics: Ask for (1) the share of pilots live after 6 and 12 months, with a clear definition of “live”; (2) independently verified customer outcomes against a pre-deployment baseline; (3) recurring product revenue and retention separate from implementation fees; (4) gross margin after all deployment engineers, cloud and support costs; and (5) implementation time and engineer-hours per customer, tracked across successive deployments. A platform should show faster, cheaper subsequent launches and expansion without proportional growth in delivery headcount. If each new contract still needs a fresh custom team, it is primarily consulting—even if the invoice says “platform.”