Whenever I sit down with enterprise leaders who feel stuck on AI, I notice a frustrating false binary.
Either they try to launch complex models directly on top of decades-old, fragmented legacy databases. Or they freeze every pilot for three years while IT attempts a massive enterprise data warehouse overhaul.
Having spent over a decade building production AI systems, I can tell you that neither extreme works. And this is what I explain in this edition of my newsletter.
- Rahul Thota, CEO, Akaike (about me)
Learn about my AI workshop for enterprises & SMBs
I work with Pharma, CPG and other industry leaders to identify which stalled pilots are technically feasible, commercially useful and worth shipping - without disrupting existing workflows.
Think of it like renovating a house. You don't wait until every single nail, paint stroke, and decorative corner is 100% perfect before you move in. And even if you do, problems still pop up once you move in. It’s always better to move in, start living there, and fix the problems as they come up in your daily routine. And guess what? Half the minor flaws you worried about end up being things you are completely fine living with.
The 95% Accuracy Trap
We saw what happens when an organization refuses to “move in” while advising a massive media and streaming platform.
We built an automated query engine on top of their content database so business leads could pull instant views and engagement metrics. During testing, the system was hitting roughly 95% accuracy. For a pilot stage, that was a massive step forward.
But their internal tech lead panicked. He insisted that before any business stakeholder or executive touched the tool, it had to hit 99.9% perfection. So instead of putting it in front of a small cohort of real users to get feedback, the project sat trapped in internal testing loops for months. They fell into analysis paralysis, trying to solve every hypothetical edge case without knowing what the end users actually cared about. The pilot ultimately got dropped.
Move In First, Fix the Plumbing Second
Compare that with how we handled a major consumer goods brand.
Their non-technical business managers were waiting days for IT to manually pull sales reports out of messy legacy ERPs. But instead of pausing everything to restructure years of historical data tables, we deployed a lightweight conversational engine right over their existing setup. On the precautionary side, we handed it to a small cohort of managers, before opening it up company-wide.
As those teams started querying the system daily, their real usage showed us exactly where the true data bottlenecks were. We saw which specific tables needed faster refresh rates, which schemas caused confusion, and which old databases nobody actually cared about. This allowed us to fix the high-priority data problems step-by-step, while completely ignoring the clutter that wasn’t driving business value.
Today, over 150 active business managers query that system in real time!
3 Questions to Break Data Foundation Paralysis
Before you delay your next pilot for a multi-year data cleanup, here are three criteria you can use to stress-test data readiness:
Are you cleaning data based on real usage or hypothetical perfection? My recommendation is, do not try to clean your entire data warehouse at once. Deploy the AI layer to a limited cohort first and let real user queries highlight the 10% of data that actually needs fixing.
Have you put the system in front of a test cohort yet? Internal IT reviews can easily turn into endless testing loops. Put the tool in front of 3 to 5 real business users with clear expectations so you get operational feedback early.
Are you treating cleanup as a prerequisite or a byproduct? The fastest way to fix a broken data model is putting it in front of actual users and watching how they interact with your AI system.
You don’t need a pristine, perfectly sanitized data warehouse to get immediate business value from AI. Build the intelligence layer first, get it into end-user workflows, and let real operational demand dictate what you clean next.
I hope I’ve made my point.
- Rahul Thota, CEO, Akaike (about me)
Learn about my AI workshop for enterprises & SMBs
I work with Pharma, CPG and other industry leaders to identify which stalled pilots are technically feasible, commercially useful and worth shipping - without disrupting existing workflows.