Following a numerical result into the customer’s decision

Mandate

Make evaluation, feasibility and time resolution support the declared planning use

Outcomes

• Withheld recommendations that violated a stated hard constraint
• Required time resolution to preserve relevant demand peaks
• Kept unresolved scenarios in the product release account

Decisions in context

Five heads lead fifty-two research and engineering colleagues within ₹9.8 crore annual operating expense, alongside a distinct ₹1.7 crore capital programme. I report to the CEO without a statutory seat. Quality head George Qezvalri can challenge the product claim independently of the development team’s preferred demonstration. The constraint review required us to admit that four apparently complete recommendations did not solve the agreed problem. The time-resolution work then showed that a correct hourly total could conceal the shorter peak relevant to the decision. I fund the correction and make the unresolved part affect release scope. Customers retain their professional and operating approvals, and our product does not autonomously dispatch equipment. My 0.25 percent beneficial interest through the employee vehicle also belongs in the assessment of any proposed outside appointment, rather than remain hidden behind the technical title.

Reason for the move

Current role; considering energy-software technology leadership and a prepared first board contribution

Concurrent appointment

Executive employment without a statutory directorship

Mandate

Connect model performance claims with inputs and conditions actually available to users

Outcomes

• Withdrew a forecast claim I had approved
• Reconstructed evaluation around issue-time information
• Retained unsuccessful historical windows in the revised account

Decisions in context

Thirty-eight colleagues worked within ₹6.6 crore annual research and engineering expense. I was responsible for translating analytical results into product recommendations, which made the 2019 forecast error particularly consequential. I had approved a demonstration whose evaluation used later weather observations in eighty-six of 120 historical windows. The data could be accurate and still inappropriate for the question of what an earlier forecaster could have known. I withdrew the claim and funded reconstruction of the input boundary. The revised exercise left twenty-eight windows outside the declared tolerance, and we kept those failures visible rather than turn the corrected process into another premature success story. The experience changed my review of research: before examining the score, I wanted to understand which inputs the model had been allowed to use and the actual decision that the evaluation represented.

Reason for the move

Appointed CTO with wider product and technical accountability

Concurrent appointment

No statutory appointment

Mandate

Turn analytical methods into products whose assumptions and output status users can understand

Outcomes

• Made model assumptions part of product acceptance
• Distinguished planning output from operating authority
• Connected engineering with product and quality review

Decisions in context

I moved from analytical assignments into a continuing product with twenty-two colleagues and ₹3.4 crore annual engineering expense. Users outside the modelling team needed to understand the assumptions behind an answer and the status of a recommendation. A technically valid computation could still be unsuitable if the product hid the distinction between a supported plan and an exploratory result. I worked with product and quality colleagues to make those differences visible. The software remained decision support, with customers and qualified owners retaining operational approval. That boundary was clear in principle, but the later forecast error showed that I had not examined the evaluation conditions with equal care. Research leadership gave me a chance to strengthen that account, while also making me responsible for a demonstration whose apparent success I would eventually have to withdraw.

Reason for the move

Promoted to lead research and evaluation across the product portfolio

Mandate

Make the problem definition and constraint meaning explicit before comparing model answers

Outcomes

• Clarified differences between preferences and hard limits
• Examined evaluation conditions with client analysts
• Preserved the client’s authority over practical decisions

Decisions in context

The modelling-lead appointment brought ten colleagues and ₹1.1 crore annual delivery expense. Client programme budgets and practical approvals remained outside our delegation. I learned that disagreements about an answer often began with different definitions of the problem. A user could call something a constraint while another treated it as a preference that might be traded for a better result. An evaluation could also reward performance under conditions unlike the intended use. Those issues required a conversation about meaning before another calculation. I wanted to follow that responsibility into a product used repeatedly by people who had not attended the original modelling discussion. Joining Drevlqari made the assumptions and the status of the output a continuing engineering concern, rather than something explained once in an analytical handover and then left to the client to preserve.

Reason for the move

Joined an energy-software product company to own the continuing user proposition

Mandate

Trace numerical results through the assumptions and examples used to examine them

Outcomes

• Implemented models under technical supervision
• Checked outputs against defined examples
• Learned the distinction between calculation and decision authority

Decisions in context

I joined in June 1999 after completing computer-engineering studies. Six people formed the simulation group, and I implemented models under a technical lead. The work had no authority over a customer’s equipment or practical operating choices. I learned to trace outputs through the assumptions and examples used in development, and I enjoyed the clarity of a reproducible result. That satisfaction could also be misleading: a model might consistently solve a problem that differed from the one a user intended. Team leadership later made me responsible for explaining those differences, while product work required them to remain visible beyond a single discussion. The early habit of following a number back through its assumptions remains valuable, but my later forecast mistake taught me to include the timing and availability of the inputs in that trace.

Reason for the move

Promoted to modelling leadership