New Relic’s India Observability Data Shows Maturity and the World’s Costliest Outages
India leads the world on OpenTelemetry and AI agent monitoring. It also reports the highest downtime bill in the survey. Both are true, and six tools explain the gap.
New Relic has published an India edition of its 2026 Observability Forecast, drawing on 500 Indian respondents from a global survey of 2,575 IT and engineering leaders across 24 countries, with fieldwork in April and May 2026. The headline finding is a paradox the report describes more than it resolves. Indian organisations lead the world on the technical markers of observability maturity and simultaneously report the highest downtime costs in the survey. Both things appear to be true, and the explanation sits in a number buried in the middle of the document.
The paradox
The maturity markers are strong. Forty-two per cent of Indian respondents report full OpenTelemetry standardisation against a 22 per cent global average, nearly double. Fifty-six per cent actively monitor AI agents in production against 35 per cent globally. AIOps adoption runs at 54 per cent against 36 per cent, and machine learning model monitoring at 57 per cent against 38 per cent. On every one of those measures India sits well ahead of the rest of the surveyed world.
The outcome markers run the other way. Indian enterprises report an average of 3 million dollars in losses per high-impact outage, the highest in the survey. Sixty-one per cent experience such outages weekly or more, and 31 per cent experience them multiple times a day, which the report describes as a tenfold increase on 2025. Time to detect a high-impact outage has stretched from 41 minutes to 55 minutes, and mean time to resolve now exceeds an hour.
What an hour of downtime costs
The cost figures carry the headline and they deserve examination. New Relic puts hourly downtime cost at 2.43 million dollars, which works out to about 40,500 dollars per minute, and the 3 million dollar average per outage is consistent with a mean time to resolve of slightly over an hour. The arithmetic reconciles, which is more than can be said for many vendor cost estimates.
What the release does not explain is how anyone calculates the cost. Lost revenue, service credit penalties, remediation labour and reputational damage are different quantities, and they do not carry the same weight for a bank, a retailer and a logistics company. A single average across 500 respondents in 12 industries is a blunt instrument, and the 3 million figure is best read as an order of magnitude rather than a measurement.
Six tools explain the gap
The mechanism connecting India’s tooling maturity to its poor outcomes is fragmentation, and the report supplies the number. Indian organisations run an average of six monitoring tools. Standardised telemetry therefore sits in silos, which is why only 20 per cent of Indian leaders say AI-assisted observability has significantly improved event correlation, against 27 per cent globally.
That is the crux of the report. OpenTelemetry solves the problem of producing consistent data. It does not solve the problem of analysing that data across tools that share no query layer. A company can standardise fully on OpenTelemetry, run six tools, and still find that no single system can trace a request from the frontend through the service mesh to the database. The data arrives in the right format and sits in the wrong places.
The developer cost lands here too. Indian respondents report losing half of developer time to system disruptions, up from 35 per cent a year earlier and close to double the 30 per cent global median. That figure points at the same fragmentation: engineers spend their hours reconstructing context that a unified platform would have held together.
Industry Variance
Industry variance is where the single average is weakest. A bank’s hour of downtime involves regulatory notification, transaction reversals and customer compensation that a media company’s hour does not. The release reports 12 industries in the sample but gives no breakdown, so a reader running a hospital or an airline has no way to locate their own operation against the 3 million dollar figure. That is the most useful thing the report could publish, and it does not publish it. The one comparison that would settle the central argument is also absent. If six tools cause the correlation failure, then organisations running fewer tools should report better correlation, and the release gives no such cross-tabulation. That single analysis would move the fragmentation thesis from plausible to demonstrated, and it is the most obvious thing to compute from data the survey already holds in its own responses.
The security gap
One finding deserves separate attention. Thirty-two per cent of Indian organisations cite security failures as a frequent cause of downtime, nine percentage points above the global average and, by the report’s own description, the market’s widest operational gap. That is a different failure mode from the third-party cloud outages and software updates the report lists as primary triggers, because a security incident is adversarial rather than accidental.
It also suggests the observability and security toolchains do not talk to each other, which is the fragmentation problem wearing different clothes. A team that can see latency but not intrusion, or intrusion but not the service dependency behind it, is running two half-pictures of the same estate.
The AI agent advantage
The most interesting finding is the one running counter to the rest. India leads the world on AI agent monitoring, 56 per cent against 35 per cent, and the report connects that to returns. The share of Indian enterprises realising 3 to 5 times return on observability nearly doubled year on year, from 19 to 35 per cent, and globally teams monitoring production AI agents are more than twice as likely to achieve 3x or better returns than those running unmonitored systems.
The numbers behind the claim are consistent across three measures, which lends it some weight. AIOps at 54 per cent against 36 globally, machine learning model monitoring at 57 per cent against 38, and production agent monitoring at 56 per cent against 35 all point the same direction. India’s enterprises have instrumented the AI layer faster than their global peers, and the report’s explanation is that the same hyper-growth infrastructure driving AI adoption also forced the tooling investment.
Whether that is causation or selection remains unclear. A company that monitors its AI agents in production is likely to monitor everything else too, so the return may reflect general operational discipline rather than agent monitoring specifically. The report’s framing implies the former while its evidence supports the latter.
Where the comparisons need care
Several findings rest on year-on-year comparisons the release does not substantiate. A tenfold increase in multiple-daily outages implies 3.1 per cent of respondents in 2025, and no explanation accompanies a change of that magnitude in a single year. The jump in developer time lost, from 35 to 50 per cent, is equally abrupt. Either Indian operations deteriorated sharply in twelve months or the two samples differ in composition, and the release does not address which.
The survey also comes from a vendor selling the remedy. That does not make the findings wrong, and the methodology is disclosed in more detail than most vendor research offers, naming Enterprise Technology Research, the sample, the 24 countries, the 12 industries and the April and May 2026 fieldwork window. It does mean the framing deserves the same scrutiny as the figures.
The report’s practical recommendation is consolidation, and the data supports it more strongly than the release makes explicit. Six tools per organisation, standardised telemetry that cannot be correlated, half of developer time lost and the highest outage cost in the survey are four symptoms of one condition. The argument that a unified platform fixes it suits the vendor commercially, and it also happens to be where the numbers point, which is not the same thing as being wrong about it.
Substantial Cohort
One note on the sample. Five hundred Indian respondents from a global 2,575 is a substantial cohort and the largest single country group, which makes the India-specific figures more reliable than most regional cuts of a global survey. It remains a population of people who answer surveys about observability, and the report does not say how anyone recruited them.
The report also lands awkwardly against the consolidation argument it makes. Indian enterprises say they prioritise AI and machine learning features in vendor selection at 39 per cent against cost at 25 per cent, which is a rational preference when returns are strong and a dangerous one when it drives tool accumulation. A buyer who chooses on capability adds a seventh tool before removing any of the first six, and the report does not connect its own buyer-preference finding to its own fragmentation finding.
What to watch
Four markers. Whether the six-tool average falls, because that single figure predicts the rest and it is where consolidation either happens or does not. Whether New Relic publishes the correlation analysis behind the AI agent return claim, since that is the report’s most actionable assertion and its least evidenced. And, hether time to detect stretches past 55 minutes, because detection is where standardised telemetry should help most and is not. And whether the security failure share moves, because at 32 per cent and nine points above global it is the widest gap in the dataset and the least discussed.