NowCast by Thurro continues to deliver disciplined forecast accuracy 

The Q3 FY2026 earnings season is over. A large number of listed companies have reported results, and investors and analysts are sifting through the numbers to understand what changed and what it means for the quarters ahead. 

By the time results are released, however, the quarter is already over and the wait for the next one begins. Until then, investors operate in an information gap, piecing together press releases, management commentary, industry data, and brokerage research to gauge real-time performance. It is, at best, an imperfect and fragmented exercise.

Investors seek continuous insight into how their investments are performing and rely on analysts for updates on the macroeconomy, sectors, and companies. Roadshows, research notes, and thematic reports provide context—but ongoing visibility into underlying performance remains essential for informed decision-making.

That is where NowCast by Thurro comes into play. A cutting-edge technique of predicting the immediate present and near-future performance of companies using vast amounts of publicly available data, sophisticated modelling, and advanced algorithms, NowCast by Thurro was built to fill this information vacuum.

In the past, our track record has been strong: more than two out of three NowCasts came in within ±5% of actual reported revenues, with an average error close to zero. It is, however, in Q3 FY2026 that NowCast by Thurro demonstrated the maturity of its process, with every estimate except one landing within the tolerance band. 

Let the numbers speak. In Q3 FY2026, NowCast by Thurro estimated the revenues of 21 listed companies across a wide range of sectors: exchanges, asset management, automobiles, logistics, internet commerce, food services, energy, and financial services, covering around INR 1.6 trillion in quarterly revenue. BSE and Hyundai both came in at an almost perfect 0.0% deviation. Even for large, complex businesses—Coal India with actual quarterly revenue of INR 349 billion, and Maruti Suzuki at INR 475 billion—the NowCast estimates were within 0.7% and 1.6%, respectively. Ten out of 21 companies (47%) were within a ±1% error band, and 15 of 21 (71%) were within ±2%. Only CONCOR, at 7.9% deviation, fell meaningfully outside the tolerance range.

NowCast by Thurro relies on hard data, sophisticated modeling, and advanced algorithms. This means that not only is the average error close to zero, but the dispersion of the error is also very low. Analysts tracking certain companies and stocks can reportedly be more hopeful and positive about the outcomes. An average of ~0% error across forecasts over time indicates that there is no emotional bias in the Nowcast. Short-term trajectories of a company’s performance can change rather quickly because of dynamic regulations, natural expansion, altered competitive dynamics, or other cross-industry changes. It is important, from the perspective of investors, to understand these changes for two reasons: (a) understand how the stock price action could react to these changes, and (b) be better prepared to ask the management deeper questions on the changes taking place in the business dynamics of the company and the sector.

Typically, analysts have been trained to look at financial results and accompanying commentary to gauge how a company is performing. NowCast by Thurro draws upon publicly available digital information on the companies. In each sector, we track what we call “core alternative data”—the operational data that serves as a leading indicator of revenue.

These “core alternative data” are then paired together in a model with relevant other data points like pricing, inflation, macroeconomic indicators, etc., to generate the NowCasts. For all the NowCasts by Thurro, there is sophisticated modelling, powered by advanced algorithms, that takes into account the core alternative data and meshes it with many other macroeconomic and sectoral indicators to develop a robust model for predicting the revenues of a company.

The key insight is that there is no one-size-fits-all model. Each company’s NowCast is a bespoke model built on the core alternative data specific to its business.

The Q3 FY2026 numbers did not happen by accident. Over the past several quarters, the Thurro data team has iterated significantly on the underlying models and, more importantly, on the processes that surround them. Here are the key learnings. 

1. Sanity checks are as important as the model itself: This is perhaps the single biggest learning. The NowCast models are automated. They ingest operational data, run statistical models, and produce estimates. But automated does not mean autonomous. Every quarter-end, the data team runs a rigorous sanity check on every input variable: Are the sales volumes in the right order of magnitude? Has there been a sudden unexplained spike or drop? Is the data consistent with the source?

2. Monthly fine-tuning tightens the quarterly estimate: For several companies, the NowCast now operates on a monthly cadence within the quarter. Some data sources provide daily or weekly feeds. Exchange trading volumes, for instance, are available daily. In such cases, the model is fine-tuned as each month closes, incorporating the actual data for the month that has passed and recalibrating the estimate for the remainder of the quarter. By the time the quarter ends, the NowCast has had two rounds of intra-quarter refinement, significantly reducing the residual error.

3. Data type and magnitude errors are the silent killers: A recurring learning has been that the most damaging errors are related to data. The team has built checks specifically for this: range validation (is this number within the expected bounds for this variable?), magnitude validation (is this roughly in line with historical values?), and spike detection (has there been a sudden change that needs to be investigated?). These checks are run before the model is allowed to produce a NowCast, acting as a gatekeeper on data quality.

4. Each company needs a bespoke approach: Different sectors—and indeed different companies within the same sector—require different data taxonomies and model architectures. When the core alternative data are well-identified and the data pipelines are clean, the model is remarkably accurate. When they are not, the error creeps up, which is also what makes the sanity check process and the choice of variables so critical. The NowCast is ultimately a function of two things: the right data, and the right process around that data.

The NowCast universe has grown significantly from 14 companies to 22 over the last year.  

The process is selective: a company is added only when we can identify reliable “core  alternative data” for it and the model achieves acceptable accuracy in backtesting. Not every company’s operational data lends itself to accurate nowcasting; the success rate for new additions is about 50–60%. At the current pace, we are adding 10 to 15 new companies per year. Currently, the pipeline has 13 additional companies in various stages of development. We expect 7 to 8 of these to reach the required accuracy threshold, which would take the active NowCast universe to approximately 30 companies by mid-2026. The sectors we are expanding into include energy, retail, and additional financial services names.

The ambition remains what we had stated earlier: to take NowCast by Thurro to at least 50 companies, covering a wide swathe of sectors and a significant share of the listed corporate universe in India. This set is expected to cover companies from auto, energy, retail, financial services, and many more. Watch this space.

The journey from a proof of concept covering a handful of companies to a production-grade system nowcasting 22 companies with sub-5% average error has been one of continuous iteration. The models have gotten better, but more importantly, the processes around them—data validation, monthly calibration, source verification—have matured. The Q3 FY2026 results are the strongest validation yet that this approach works.

Not only are the NowCast by Thurro unique on the street, they also offer you sharp insights on the company’s performance almost as soon as the quarter ends, sometimes even before. By the end of the quarter, users of the platform have a significant head start on the key trends of the industry and individual companies. For investors, this means better-informed decisions, sharper questions for management, and a meaningful edge in understanding the short-term trajectory of businesses. We have used a similar framework for Thurro’s food and beverage inflation NowCast—which tracked official MOSPI prints closely through 2025—offering investors early visibility into inflation trends before official releases.

View disclaimer