You can't estimate a prompt
Story points ran on self-estimates. When you're one prompt away from a working feature, what exactly are you estimating? Most teams already dropped them.
→Navigara · Product·AI ROI
Prove your AI ROI.
Measure engineering before and after AI. Then tie the speed to the roadmap, and the roadmap to revenue.
Trusted by engineering teams
NKey metrics · trending to month-end
LivePerformance · YoY
+128%
Target +90% · aheadtarget
Delivered with AI
80%
Share of ETV shippedtarget
Roadmap alignment
40%
Target 55% · behindtarget
Cost per ETV
$200
Target $250 · efficienttarget
01The pain
Everyone bought the tools. Nobody can show what they bought.
4 months
Uber’s entire 2026 AI coding budget, gone. Nothing shippable to show for it.
“That link is not there yet.”
More tokens. More spend. No stat that proves anything reached a customer.
02The real problem
You can’t measure ROI on AI until you measure engineering.
If a KPI for engineering existed, you’d already be using it to prove AI tooling ROI.
"Faster" is not a metric
The pitch for AI tools is "make engineering faster." Faster at what? Coding, reviewing, lines generated? None of that reaches a customer.
03What to measure
Tie AI to the roadmap, not to revenue.
You can’t tie engineering straight to revenue. As a former CTO, I got tired of guessing at that gap. So measure the next best thing: your roadmap delivery.
EngineeringRevenue
no direct link
EngineeringRoadmap deliveryRevenue
the path you can measure
04The unit
Why invent a unit? To see the gain, without the bias.
Performance (ETV)HeadcountIndexed · pre-AI = 100
ETV · Engineering Throughput Value
Scored straight from commit history by LLMs, ML, and algorithms, through the lens of a senior engineer. It exists to do two things vibe-based reporting can’t.
01 · Show the YoY gain
Performance, normalized for headcount, against the year before AI. Here it’s up 128% while the team grew far less.
02 · Drop the bias
No surveys, no self-report, no story points. One unbiased read of what the code actually shows.
05The method
Four questions turn commits into an ROI statement.
The first three are the math. The fourth is the one teams skip.
Q1How much faster are we thanks to AI?Performance change
Q2Faster at what?Work mix
Q3What did the performance gain cost?Cost per throughput
Q4Was any of it on the roadmap?The part everyone skips
Step 1Performance change
Measure this year against the year before AI.
No estimates. Real performance now against the year before AI. Every team is faster, and the gain is never uniform.
April 2025April 2026Avg developer performance · ETV
Backend core
BE team
+16%
Data team
Devs who handle databases
+55%
Externals
Outsourced agency devs
+7%
SFO
West US team
+33%
Step 2Faster at what
Split the performance, or the number lies.
80% means nothing until you know the category. Bug fixes? Maintaining AI slop? Or actual features?
Growth 62%
Maintenance 26%
12%
Growth· New valueMaintenance· Keeping it runningFixes· Bugs & slop
The test:good adoption keeps the bug and maintenance share flat. If those buckets balloon, you’re paying the tools to clean up after the tools.
Across 676 OSS engineers
Maintenance dropped 10 points. Growth rose.
Step 3Cost per throughput
Tie the performance to the spend.
$25k
tooling spend last month
7 AI tools, one month
ETV
shipped to production
the 128% gain, scored
$200
per ETV shipped
efficient token spend
Read it as a unit price. Cheap per ETV means the spend is working. Expensive per ETV means it isn’t.
Step 4Roadmap alignment
The part everyone skips.
Was any of this on the roadmap? Or did the team finally build the thing they always wanted to, now that they can? Connect ETV to Jira and you can see it.
45% unaligned isn’t automatically bad. But look at what they’re building. Usually it’s a process problem, not a people problem.
40%
Roadmap work
15%
No roadmap link
45%
Unaligned
→In the product
See what your AI spend actually built.
Key objectives rank by delivered ETV. Open one and you see the features your token spend shipped, and whether the work was growth, maintenance, or fixes.
Your AI bill stops being a number and starts being a feature list.
NObjectives · Q1 2026
Ranked by ETVObjectiveAI cost
01Revamp checkout experience
420
74% growth18% maint8% fixes
02AI-assisted dev tooling
310
66% growth22% maint12% fixes
03Platform reliability
268
41% growth47% maint12% fixes
04Payments rework
194
58% growth27% maint15% fixes
Total ETV1,192
06The payoff
Put it all together and you get faster roadmap delivery.
Roadmap objectivePlannedShippedSoonerAI cost
Revamp checkout experience3 mo
Platform reliability2 mo
AI-assisted dev tooling4 mo
Payments rework2 mo
Whole roadmap~3 mo
Three months of roadmap, pulled forward.That’s revenue Uber couldn’t find.
→The deliverable
Read unbiased reporting about your engineering.
Weekly, monthly, and quarterly reports on what your teams are actually doing and what it cost. One page the board reads in two minutes.
Export to PDF and send
Navigara · Engineering report
Q1 2026
Performance+128%target +90%
vs. year before AI
vs. year before AI
Roadmap alignment40%target 55%
-6pp QoQ
-6pp QoQ
Cost per ETV$200target $250
-31% QoQ
-31% QoQ
Objectives that drove the quarter
01Revamp checkout experience420 ETV
02AI-assisted dev tooling310 ETV
03Platform reliability268 ETV
ProofMeasured in the open
This isn’t theory. The repos already show it.
676 engineers · 6 big-tech orgs · 5 quarters. Lines of code never saw it. ETV did.
→The same data does more
Same unit. Four more plays.
ETV reads your commits once. That read pays off well past the ROI number.
Vendor control
Do outsourced teams earn their rate?
Benchmark agency teams in the same ETV rate as your own. The premium only makes sense next to the output it actually buys.
360 reviews
Reviews grounded in what shipped.
The same data, read per team and per person. Outcome-based, not activity. Evidence to back the engineer the dashboards never credited.
CapEx & OpEx
Capitalize software, defensibly.
Audit-ready CapEx and OpEx, derived straight from the code. No spreadsheets, no estimates.
Explore CapEx & OpExSlack & Teams
Ask in Slack. Get an answer.
Natural-language questions across every metric, team, and timeframe. Cited answers, in real time.
→The honest part
Same measurement. Two very different stories.
That walkthrough assumed adoption is working. You’d be surprised how often we see the opposite.
Working
- Roadmap delivery up
- Bug and maintenance flat
- Spend efficient per ETV
- Speed turns into revenue
Not working
- High token burn
- Roadmap delivery flat
- Fixes and slop climbing
- Spend with nothing to show
Tell me which one your team is.
→Stop guessing
Lead engineering with data.
Your team is performing. Time to prove it.