contributors
Prashant Lamba
Chief Information Officer, RGP
November 4, 2026 • 3 Min Read
Every leadership team meeting has the same theme: The team pushes to implement AI use cases ASAP; the supplier assures everyone their “autonomous agents” will solve all problems by the end of next quarter, and the decision is made.
When enterprise AI fails, it’s rarely because the model wasn’t smart enough. It’s because we hooked up a 100-gigabit fiber line to an old dial-up modem. We pointed incredible intelligence straight at messy data.
Intelligence is the tool; context determines worth and value
PRASHANT Lamba Chief Information Officer, RGP
AI models are becoming commodities that everyone can purchase. Your moat isn’t a model; it’s your operational context—the unique pricing strategy, client relationships, and internal rules of the system or process only your team understands. You can’t buy your enterprise’s operational context off the shelf: your customer histories, pricing and margin data, contract details, and business logic.
That context lives inside your data architecture. If your core systems don’t talk to each other—if customer records in your CRM disagree with your billing platform, or your ERP still relies on manual reconciliation spreadsheets—adding an AI layer doesn’t create intelligence. It simply creates confusion.
The Three Tests Before Letting AI Touch Your Data
Before getting yet another exciting and fancy tool to connect with your internal data and systems and call it “digital transformation,” answer these three basic questions:
Test 1: Lineage
Where did this data come from, and who updated it last?
What breaks if you skip it: The agent quotes pricing from last year’s draft someone forgot in a shared folder.
What good looks like: Clear ownership, and one undisputed system of record.
Test 2: Permissions
What stops an AI agent from pulling up the sensitive data?
What breaks if you skip it: Semantic search bypasses the messy folder permissions nobody ever cleaned up.
What good looks like: The model only sees what the logged-in user is already cleared to see.
Test 3: Semantic Consistency
Does “active account” mean the same thing to sales, marketing, and finance?
What breaks if you skip it: Every team gets a different answer, all delivered with 100% confidence.
What good looks like: A single, agreed-upon definition in the system before the bot starts answering questions.
No AI model can work efficiently in broken plumbing
PRASHANT Lamba Chief Information Officer, RGP
Many people still see data cleanup as boring admin work. But when you automate bad data, you make expensive mistakes happen fast. Setting up clear permissions and keeping records clean isn’t red tape—it’s how you make sure your tools don’t return bad results.
To get value out of AI technology investments:
- Pick the problem, not the model. Pick one specific, repeatable problem where success is measurable with clear KPIs and ROI-account reconciliation or connecting AI to sales data.
- Assign real owners and treat data as product. If no human is accountable for keeping a dataset clean, keep it out of automated workflows.
- Give people safe and frictionless tools. Make the official tools easy to use. If getting access feels like a nightmare, your team will paste sensitive data into personal ChatGPT tabs on their phones. Give them a frictionless, approved space to experiment so they don’t have to sneak around IT.
The real winners won’t be the companies spending the most on AI vendors. They’ll be the teams willing to do the unglamorous work of making their operational context usable—clarifying which sources are authoritative, who has access, what key terms mean, and who is accountable for keeping that context reliable.