AI, Straight-Through Claims and the One-Hour Clock: How IRDAI's Cashless TAT Rules Are Forcing Automation in Commercial Group Health

On 29 May 2024 IRDAI issued its Master Circular on Health Insurance Business, folding 55 earlier circulars into one document that still governs the market in 2026. Buried in it are four timelines that quietly re-engineered how every health insurer and third-party administrator (TPA) runs claims. Cashless pre-authorisation must be decided within one hour of receiving a complete request. Final discharge authorisation must follow within three hours. Where a hospital charges extra because the insurer ran past three hours, that cost is borne by the insurer from shareholders' funds, not the policyholder. Final settlement of a claim must happen within a defined window after complete documents are received, with delay interest payable.

For a corporate buyer of group medical cover (GMC), this is not regulatory trivia. It is the single biggest driver of how your employees experience the policy at the hospital counter. The General Insurance Council's Cashless Everywhere push, live since early 2024, widened the expectation further by allowing cashless treatment even at non-network hospitals. So the volume of cashless requests an insurer must clear inside one hour keeps rising while the deadline stays fixed.

IRDAI's own commentary has cited that a large share of pre-authorisations, in the region of 87 percent, are now cleared within the hour, and roughly 97 percent of discharges within three hours. Those numbers are only reachable because adjudication is no longer fully manual. The maths is unforgiving: a mid-size insurer handling tens of thousands of cashless requests a day cannot put a human doctor on each one inside sixty minutes. Straight-through processing (STP), driven by rules engines and increasingly by machine-learning models, is what closes that gap. The regulation did not mandate AI. It mandated a speed that only automation can sustain.

When a network or non-network hospital uploads a pre-authorisation request, the insurer or TPA system now runs a sequence that used to sit on a medical officer's desk. Optical character recognition and document intelligence read the discharge summary, the diagnosis, the line-item estimate and the treating doctor's notes. A rules layer checks eligibility: is the member active, is the ailment within the waiting period, is the procedure covered, does the room rent fall inside the policy sub-limit, are there proportionate-deduction triggers. A scoring model then routes the request into one of three lanes.

The first lane is clean auto-approval. A cataract day-care procedure on a long-tenured member, well inside sub-limits, with a recognised hospital tariff, can be approved by the machine in minutes. The second lane is auto-query, where a field is missing or a tariff looks off, and the system fires a templated request back to the hospital. The third lane is human referral, where the model flags ambiguity, a high amount, a fraud signal or a non-standard procedure for a medical officer to decide.

The point a corporate risk manager should hold onto is this. STP is excellent at the clean middle of the distribution, which is most claims by count. It compresses the one-hour and three-hour windows into something the insurer can actually hit. The risk lives in the tails. A model tuned to protect the TAT statistic will lean towards fast disposal, and fast disposal of an ambiguous case is either an over-cautious query that delays a sick employee or a quiet partial decline the member only discovers when the hospital asks them to settle the balance themselves. That tension between speed and accuracy is the whole story of AI in cashless claims, and it is exactly where your service-level agreement (SLA) has to bite.

There is a real, defensible upside here, and brokers should not be cynical about it. Before the TAT regime, the most common complaint on a corporate GMC was the discharge wait: an employee medically cleared to leave but stuck for hours while the TPA cleared the final bill. The three-hour discharge rule, backed by the shareholder-funds penalty, has materially shortened that wait at insurers that invested in automation. That is a tangible employee-experience win you can point to at renewal.

STP also reduces a particular kind of arbitrary inconsistency. A rules engine applies the same room-rent logic, the same sub-limit, the same proportionate deduction to every claim, every time. Human adjudicators, working under volume pressure, drift. They approve generously on a quiet morning and tighten up when the queue is long. Deterministic automation removes that lottery, which is genuinely fairer to members even when the answer is a deduction they dislike.

There is a data dividend too. Once claims flow through structured pipelines, the insurer can give you cleaner management information: cashless versus reimbursement split, average pre-auth approval time, query rates by hospital, top procedures by cost. A risk manager who asks for this in the right format can spot a problem hospital, a leaking benefit or a wellness intervention worth funding.

At renewal, ask the insurer for its actual TAT performance on your specific account, not the market average. The headline 87 percent and 97 percent figures are portfolio-wide. Your population, your network mix and your top hospitals may sit well below that, and only account-level data tells you.

Where automation helps most is high-frequency, low-ambiguity care: planned surgeries at network hospitals, standard maternity, common day-care procedures. For a workforce skewed towards these, faster cashless is close to pure benefit, and a well-run STP programme is worth paying a slightly firmer premium to access.

Automation does not remove claims disputes. It changes their shape and, dangerously, their visibility. The failure modes below are the ones that surface as employee escalations after a fast cashless decision.

• Silent partial approvals. The system authorises a lower amount than billed because of a room-rent or sub-limit deduction, and the employee learns only at discharge that they owe the balance. The TAT was met. The experience was poor. Insist that any deduction above a threshold triggers a clear, written explanation to the member, not just a reduced sanction figure.
• Over-querying to stop the clock. Some insurers fire an information query to pause the one-hour timer, then take their time. A high query rate on your account, especially repeated queries on the same case, is a tell. Track it.
• Model-driven declines on edge procedures. New treatment protocols, modern day-care techniques and combination procedures can fall outside the model's training, producing reflexive declines or referrals that a human would have approved. These hit your most seriously ill employees hardest.
• Non-network friction under Cashless Everywhere. Cashless at a non-network hospital depends on the insurer agreeing a tariff with that hospital in real time. The AI cannot manufacture a tariff that does not exist, so these requests stall more often. If your workforce is geographically dispersed, this matters.

The broker's job is to convert these failure modes into contract language and into a monthly review cadence, so the failures show up as numbers you can act on rather than as a stream of individual employee complaints to your HR team.

The TAT rules give you regulatory minimums. A good corporate programme contracts for more than the minimum, because the regulation governs the insurer's compliance, not your members' experience. Build the following into the service agreement with the insurer and its TPA.

First, separate speed metrics from quality metrics. Demand reporting on cashless pre-auth approval time and discharge time (the TAT story) and, separately, on first-time approval rate, query rate, partial-approval rate and decline rate. An insurer hitting 95 percent on TAT while declining or heavily deducting on a fifth of cases is not running a good programme.

Second, require human-in-the-loop above defined triggers. Any claim above a rupee threshold, any full decline, and any deduction above a set percentage of the billed amount should carry evidence of a medical officer's review, not a pure machine decision. This is also where IRDAI's broader expectations on model governance and explainability for insurers point, so you are pushing on an open door.

Third, write a member-communication standard. Every partial approval or decline must reach the employee in plain language with the policy clause cited, within the same window as the hospital decision. Most member anger is about surprise, not about money.

Fourth, fix an escalation path with names and timelines that sits above the TPA, so a stuck case at 2 am does not wait for the next working day. Map this to the insurer's grievance machinery and the Bima Bharosa route so your members have a real fallback.

Finally, agree a quarterly model-and-process review where the insurer walks you through query trends, top deduction reasons and any change to the auto-decision logic that affects your population.

Two insurers quoting the same premium on your GMC can offer wildly different lived experiences depending on their claims automation. At placement, treat operational maturity as a rateable factor, not an afterthought. The questions below separate a genuinely automated insurer from one running manual processes behind a TAT-compliance veneer.

1. What share of your cashless requests are decided straight-through without manual touch, and what is the auto-approval versus auto-query split?
2. What is your account-level pre-auth and discharge TAT for books of our size and industry, over the last four quarters?
3. How do you handle non-network cashless under Cashless Everywhere, and what is your real-time tariff agreement success rate?
4. What governance sits over the adjudication model: who signs off changes, how is bias and drift monitored, and how is explainability handled for declines?
5. What does your query rate, partial-approval rate and decline rate look like, and how have they moved since automation went live?

An insurer that answers these crisply, with account-level numbers, is one whose automation is a service asset. An insurer that retreats to portfolio averages and marketing language is asking you to take its claims experience on faith.

On pricing, expect the better-automated insurer to hold or firm its rate rather than discount, because faster, cleaner cashless reduces leakage and grievance load and they know it is worth paying for. That is usually the right trade for a corporate buyer. The cheapest GMC quote often hides a slower, more manual, more dispute-prone claims operation that your HR team pays for in escalations and your employees pay for in discharge waits. Place on total claims experience, evidenced by data, not on headline premium alone, and put the automation commitments you were promised into the binding wording so they survive the relationship manager who sold them.

The TAT regime and the automation it forced are now baseline market reality, not a differentiator you can wait out. The differentiator is whether your specific programme is governed to capture the upside and contain the failure modes. A practical sequence for the next ninety days follows.

Pull your own data first. Ask your current insurer and TPA for twelve months of account-level claims metrics: cashless share, TAT performance, query rate, partial-approval rate, decline rate, delayed-payment interest paid, and top decline and deduction reasons. If they cannot produce this cleanly, that itself is a finding.

Map your member complaints against that data. Most HR teams hold a thread of claims grievances that never reach the broker in structured form. Tie those anecdotes to the metrics and you will usually find one or two hospitals, one or two benefit clauses, or one TPA workflow driving a disproportionate share of pain.

Rewrite the SLA before you market the renewal, using the speed-versus-quality split, the human-in-the-loop triggers, the member-communication standard and the quarterly review described above. Carry that SLA into the market as part of the requirement, so insurers quote against your governance bar rather than the regulatory floor.

Finally, brief your HR and finance stakeholders that faster cashless and fair cashless are different things, and that the programme is being managed for both. An employer that can show employees a quick discharge and a clear, explained decision when the answer is partial has a benefit that retains staff. That is the real prize hidden inside a set of regulatory timelines, and it is won at the SLA and placement table, not at the hospital counter.