I make quality the competitive advantage.

Senior Manager, Accenture — AI in Quality Engineering. Based in Burlington, Ontario.
20+ years turning testing from bottleneck into accelerator.

Selected work

Tools and frameworks I've built — each one solving a problem that manual effort couldn't scale past. Drawn the way they're designed: on paper, in lines and nodes, before a line of code.

Experience

Senior Manager at Accenture, in Custom Application Engineering — QE transformation, test automation, and performance engineering for enterprise clients.

Earlier in my career: delivery and performance-engineering roles across North America and India.

275% automation ROI $1.5M licensing savings 0.02% defect leakage

Certified — AWS Cloud Practitioner · Azure AZ-900 · ISTQB Foundation · Power Platform PL-900 · AppDynamics Sales Engineer · Splunk Power User.

Recognition — CQTA Practitioner of the Year 2026.

Education — metallurgical engineering, Punjab Engineering College, Chandigarh.

Writing

Writing is how I keep up — and how I test ideas in the open. It all lives under one name: AI × QE. Not vendor theory, but my own tools, workflows, and experiments — and what actually held up when I put them to work.

AI × QE — the newsletter

My weekly LinkedIn newsletter on AI in quality engineering: new tooling, agentic workflows, and the evidence behind the claims — “Quality Accuracy.” Subscribe to get each edition as it lands.

Tool — The AI × QE ROI Calculator

The companion to Edition Nº 3, “The ROI of AI in Enterprise QE.” Every vendor pitch shows one ROI number; this calculator shows two — the pitch-deck figure, and the same figure calibrated by your evaluation-measured acceptance rate. Run your own team's numbers and see where the claim survives.

Edition Nº 1 — “30 AI keywords you hear in meetings, explained like you're five”

The first edition: thirty AI terms that come up in every meeting — attention, MoE, quantization, LoRA, RLHF, evals and more — each explained in plain language, with an everyday object to picture it by. For everyone who nods along in the meeting and googles it after.

Next edition — Wed Oct 28: spec-driven development

Edition Nº 2: why the blueprint matters more than the prompt — traceability, independent verification, and evidence as a requirement, drawn from the spec-driven framework in my work above.

The weekly LinkedIn series

Between editions, shorter posts on my profile — one idea at a time, argued in the open: what AI actually changes in testing, and where the hype breaks.

About

I'm a hands-on engineering leader: I can discuss an automation framework, investigate a delivery bottleneck, build a working prototype, and explain the investment to a client — in the same meeting.

Two decades across banking, payments, insurance, retail, and public-sector systems taught me one thing: quality holds when it's engineered in, not inspected in. These days that means applied AI — agents, spec-driven development, and evidence a planning meeting can rely on.

Home is Burlington, Ontario, where I live with my wife and our two boys. The work is important; it isn't the whole story.

Role
Senior Manager, Accenture — Custom Application Engineering
Focus
AI in Quality Engineering — QE transformation, test automation, and performance engineering
Based in
Burlington, Ontario — home is my wife, our two boys, and the crew in the photos below
← Back to the site

The AI-in-QE ROI Calculator

No percentage lifts. This is a testing program, the way you actually run it.
So many test cases a quarter, so many hours to write them. For every activity in the testing lifecycle, enter the work you do and the hours it takes today — then the hours it takes with AI, split honestly into generation time and the calibration & evaluation time that makes the output worth trusting. The roll-up is what an executive actually needs: hours, dollars, and capacity, with the evaluation effort on the bill instead of under it.

Your program

Costs first, then the work itself, activity by activity. The starting numbers describe one example program — 1,500 test cases a quarter at half an hour each is 750 hours of test design, before AI enters the room. They are a worked example, not anyone's real figures: replace them with your own program's.

Salary, benefits, and overhead, blended across the team — one hour of QE, fully loaded.

Licences, platform, and usage charges for the whole program — usually the only cost in the pitch deck.

Presets

Presets rewrite only the generation and evaluation columns below — your volumes and today's hours stay yours. The vendor pitch zeroes the evaluation column entirely. Watch what that does to the ROI: that gap is the whole argument of Edition Nº 3.

Strategy & planning

unit: test plans & strategies

—

Test design

unit: test cases

—

Automation scripting

unit: automated scripts

—

Test data & environments

unit: data sets / env builds

—

Execution

unit: test runs & cycles

—

Defect analysis & reporting

unit: defects triaged & reported

—

Regression maintenance

unit: regression tests maintained

—

The roll-up — what the executive sees

—

ROI on the tooling spend, with evaluation hours counted as a real cost — the number you can defend in the budget meeting

Hours today / quarter
—
Hours with AI / quarter
—
of which evaluation
—
Hours saved / quarter
—
Effort change
—
Value of saved hours / qtr
—
AI tooling cost / qtr
—
Net / year
—
Capacity freed
—

Capacity freed is hours your team gets back — for coverage, exploratory testing, and the backlog. It is not a headcount line, and it is not presented as one.

—
—

Method & sources

The method, in plain words

For each activity, the maths is the maths of a program plan — volumes and hours, nothing more exotic:

Hours today = units per quarter × hours per unit
Hours with AI = units × (generation hours + evaluation hours)
Value = hours saved × loaded hourly rate
Net per year = (value − tooling cost) × 4 quarters
ROI = net per year ÷ (tooling cost × 4)

The scope is deliberately narrow: effort only. Escaped-defect savings, licence consolidation, and attrition are not counted — in either direction. If the effort case does not stand on its own, softer numbers should not be asked to carry it.

How to get your evaluation number

  1. Pick one activity — test design is a good start — and take 20 recent units of real work: actual test cases, actual scripts.
  2. Run the AI on them, and hand the output to the person who would have to sign it off in production.
  3. Time the sign-off end to end — reading, checking against the requirement, correcting, finishing what the AI got wrong. The average is your calibration & evaluation hours per unit.
  4. Re-measure when anything changes — model, prompts, or workflow. Evaluation time is a property of a setup, not of AI in general.

If nobody in the program can state the evaluation time, the program does not have an ROI number yet. It has a hope with a licence attached.

What the independent evidence says

  • METR, 2025 — a randomized controlled trial. Experienced open-source developers using AI tools took about 19% longer on real tasks, while believing they had been about 20% faster. Perception is not measurement — that gap is exactly why the evaluation column exists.
  • Cui, Demirer, Jaffe, Musolff, Peng & Salz, 2025 — field experiments with 4,867 developers across three firms. AI assistance raised completed tasks by roughly 26% on average, with the largest gains among the least experienced developers. Real value — measured in output, not in headcount removed.
  • Stanford AI Index, 2025. 78% of organizations reported using AI in 2024, up from 55% in 2023. Adoption is broad; program-level ROI, measured activity by activity, is still rarely published.
  • US Bureau of Labor Statistics — Employer Costs for Employee Compensation. Benefits add a substantial loading on top of wages, which is why this calculator prices a loaded hour rather than a salary hour.
  • Boehm; NIST, 2002. The later a defect is found, the more it costs — and inadequate testing infrastructure was estimated to cost the US economy tens of billions of dollars a year. QE hours are not overhead to be minimized; priced honestly, they are the product.

Vendor and consulting-firm ROI claims are treated as claims, not evidence — including, in the preset above, the shape of a typical vendor pitch.

Where this comes from This calculator is the working companion to AI × QE Edition Nº 3, “The ROI of AI in Enterprise QE” — the data-backed edition on what AI adoption is actually returning in enterprise testing: capacity and avoided failure cost, measured honestly — not the headcount story. It lands in the newsletter after Edition Nº 2.

Subscribe to AI × QE ↗