AIDP holder spotlight
Amar Ramdhave, AIDP
AI Delivery & Quality Engineering Leader
Independent Advisor (formerly Director of Quality Engineering, Coupa Software)
“AI rarely fails on the model. It fails on the decisions around it.”
Five questions
In his own words
Answers as given to ADaSci, lightly edited for length.
I advise organisations on delivering AI that holds up in production, with a focus on quality and evaluation. That means helping teams frame where AI creates real value, choose between building, buying or partnering, and put evaluation and quality gates around LLM and agentic systems so leaders can trust what they scale. Before this, I spent nine years at Coupa Software, most recently as Director of Quality Engineering, leading 32 engineers across eight product domains. There I led a transformation on two fronts: technical, by introducing LLM-based test generation and synthetic validation, and people, by growing engineers' skills and confidence to work with AI. Both taught me that adoption depends as much on people as on technology.
I've led AI-driven change from the quality side for years, and I wanted an independent benchmark for the hardest part to prove: whether AI actually delivers. Most credentials test whether you can build AI or govern it. AIDP tests whether you can make the right delivery calls under real constraints: what to pursue, how to pursue it, and when to stop. With no official prep course by design, the founding sitting was an honest test of whether my experience held up against a rigorous standard. Alongside my CAITL in AI transformation, it completes the picture on the delivery side.
Adoption starts long before deployment. First, a value case owned by the business and stated before any technology choice. Second, an honest feasibility and data check, with the discipline to choose buy, partner, or no AI when that's the better answer. Third, design for the people who will use it: fit it into existing workflows, measure outcomes rather than activity, and treat the gap between pilot and daily use as the real project. Finally, leadership has to keep reviewing every AI investment and move money toward what the evidence supports. Most AI that never gets adopted was never framed around a real problem in the first place.
In one organisation-wide AI rollout, adoption was pushed from the top and measured as activity: sessions per day, call counts, who wasn't logging in. The pipelines weren't mature, retrieval wasn't tuned, and LLM test-case generation for each epic had no proper instructions behind it. The result was heavy hallucination, efficiency of only 50–60%, and strong engineers who felt they were gaming a score. My team reset: we set outcome targets first, ran the work as a proof of concept, and used direct model access to start from real use cases. The lesson: lead from outcomes, not activity. Today I run adoption as an agile loop, where a central team ships proven tools with guardrails, gathers feedback, and scales only what adds value.
Sit it if you're accountable for AI outcomes, not just AI output. It rewards judgment built from real delivery, weighing trade-offs between options that all look reasonable. You can't cram for it, so the best preparation is reflecting honestly on the programmes you've led, especially the ones that didn't land. Read the Competency Blueprint closely, because it describes the role precisely. Go in thinking like the person who owns the business result, not the person building the model or checking compliance. If that's the job you do, or want to do, AIDP is a credible way to show it.
Also a founding holder
Aditya Vidwans, AIDPDelivery Manager · BackbaseRead his answers →