Businesses want autonomous AI agents, but fragmented data, outdated systems and years of technical debt are blocking deployment.
The next major AI spending wave may not be about buying better models. It may be about repairing everything those models need to work with.
Capgemini recently raised its 2026 revenue-growth forecast after reporting €12.08 billion in first-half revenue. The company attributed stronger demand partly to organizations modernizing their applications, data platforms and core infrastructure in preparation for larger AI deployments. Its second-quarter bookings rose 9.2% to €6.55 billion. See Capgemini’s updated 2026 outlook.
That is not quite the frictionless AI revolution executives were promised. Before companies can deploy intelligent agents across their operations, many must first clean up decades of disconnected software, inconsistent information and undocumented processes.
Companies Have AI Ambitions but Legacy Foundations
Capgemini CEO Aiman Ezzat described the trend as a multiyear modernization cycle. Companies may want AI agents capable of completing complex tasks, but their underlying systems were often built by different vendors, during different technology eras, for departments that rarely shared information.
According to Reuters’ report on Capgemini’s outlook, legacy applications, fragmented data and accumulated technical debt are now among the largest obstacles to deploying AI at scale.
An AI agent may be capable of reading an email, comparing a contract, checking inventory and drafting a customer response. That workflow still fails when customer records are duplicated, inventory data is delayed, contract terms are stored in personal folders or employees follow procedures that differ from the official documentation.
The model is not the weakest link. The business environment is.
AI Exposes Problems Traditional Software Could Ignore
Conventional software can operate within narrow boundaries. Accounting software processes transactions in its own database. Scheduling software manages appointments. A document system stores whatever employees upload.
AI attempts to work across those boundaries. It retrieves information, interprets context and combines material from several sources. That makes existing information problems much harder to ignore.
A company may discover that:
- Three departments maintain different customer lists.
- Nobody knows which policy is current.
- Important procedures exist only in email.
- Employees use undocumented workarounds.
- Former staff still own critical files.
- Systems use incompatible terminology.
- Managers routinely override official processes.
These conditions may have existed for years. Experienced employees compensated for them by knowing where to look, whom to ask and which rule to ignore. AI cannot reliably reproduce that judgment when the context was never documented.
Connecting an assistant to disorganized information does not create intelligence. It creates faster access to organizational confusion.
Modernization Is More Than Replacing Old Software
The obvious response is to replace legacy applications. That may be necessary, but technology replacement alone does not solve the problem.
A modern cloud platform can still contain duplicate documents, inconsistent permissions and unreliable data. A newly migrated SharePoint environment can become the same old shared drive with nicer buttons.
True AI readiness requires several forms of cleanup:
Data cleanup: Customer, product, financial and operational records must be accurate enough to support automated decisions.
Application integration: Systems must exchange information without employees manually copying it between spreadsheets and emails.
Knowledge organization: Policies, procedures and guidance must have clear owners, approval status and review dates.
Process clarification: The company must document how work actually happens, including exceptions and escalation points.
Permission design: AI must retrieve only the information each user is authorized to access.
This is why Capgemini’s results matter beyond the consulting industry. They suggest that AI demand is pulling forward work companies postponed for years. The shiny agent arrives, looks under the floorboards and finds 14 spreadsheets named “FINAL.”
Small Businesses Face the Same Problem at a Smaller Scale
A small company may not have a 40-year-old mainframe, but it can still have substantial technical and knowledge debt.
Its customer information may be divided among accounting software, a CRM, employee inboxes and someone’s personal spreadsheet. Procedures may differ depending on who is working. Important decisions may be buried in Teams messages. One employee may understand how the entire operation fits together.
Small businesses should not respond by launching a company-wide modernization program. They should select one operational area where poor information creates measurable friction.
A practical starting point might be employee onboarding, estimate preparation, customer intake, service troubleshooting or internal policy questions. The company can then identify the systems, documents, people and decisions required to complete that process correctly.
The Maisy knowledge-hub implementation model follows this principle by preserving existing Microsoft 365 documents as living sources while adding a governed knowledge layer employees can search conversationally. The objective is not to replace every system. It is to make approved information usable across the workflow.
Clean Before You Automate
An organization preparing for AI should begin by identifying authoritative sources. Each important question or decision should point toward one approved document, system record or business owner.
Duplicates should be archived. Outdated material should be separated from current guidance. Missing procedures should be captured from employees. Permissions should be tested by role. Real employee questions should be used to validate the results.
This work is less dramatic than demonstrating an autonomous agent, but it determines whether that agent will be useful. Maisy’s structured discovery and cleanup process places information inventory, conflict resolution, ownership and security before AI enablement for exactly this reason.
Capgemini’s growth suggests the AI market is entering a less glamorous but more serious phase. Businesses are moving beyond experiments and confronting the condition of their actual systems.
The companies that benefit most from AI may not be those that adopt agents first. They may be those willing to clean up the machinery before asking the robot to drive it.
