An Insurer Says It Spends Up to 1.5% of Premium on Technology. Should That Be on Your Scorecard?

On 3 August 2026, Business Standard reported ICICI Lombard's MD and CEO Sanjeev Mantri saying the insurer invests up to 1.5 percent of premium in technology. Until an executive says a number out loud, insurer technology spend sits in the same category as culture and appetite: everybody has a view, nobody can cite anything. Once it is quoted, it becomes a line a corporate buyer can put on a panel scorecard and ask every other insurer to fill in.

The timing sharpened it. Eleven days earlier, on 23 July 2026, McKinsey published How AI will reshape the economics of insurance: A CEO's guide to strategy. ANI News summarised the argument on 31 July as AI helping insurers cut onboarding costs and speed up claims processing, and Big News Network followed on 1 August with the same finding framed as the industry moving towards scale and specialisation. So the market got a consultancy thesis about AI changing insurance unit economics, and an Indian insurer putting a rupee-denominated commitment behind it, in the same fortnight.

For a risk manager running an annual panel review, that combination is tempting and slightly dangerous. Tempting because a percentage of premium is comparable across insurers in a way that "they seem digitally advanced" never was. Dangerous because the thing you actually buy from an insurer is not spend. It is claims behaviour, service turnaround and wording discipline. This piece is about how to convert the first into tests of the second, and where the two come apart.

Start with what the figure is. It is an input measure. It says an insurer has decided to route a defined share of its topline into systems, data and people who build them. It says nothing about what those systems do, which line of business they serve, or whether any of it reaches the desk handling your fire claim.

Three specific gaps sit inside a headline technology-spend number:

• Allocation is invisible. A large share of insurer technology investment in India goes into distribution and retail motor and health, because that is where the transaction volume is. Commercial property, marine and liability claims are lower volume and higher complexity, and they are usually last in the queue. A high spend ratio tells you nothing about which side of that line your account sits on.
• Capitalisation choices differ. What one insurer books as technology investment another books as operating expense, and core system migrations, cloud contracts and licence renewals can all live inside the number. Two insurers reporting different percentages may be running comparable programmes with different accounting.
• Spend is not outcome. A core system replacement absorbs enormous budget and can, in its first two years, make service worse. The buyer feels the migration, not the roadmap.

Marketing sits on the same fault line. Adgully reported on 4 August 2026 that ICICI Lombard describes its playbook as moving from GenAI campaigns to hyper-personalisation. That is a real capability and a real investment. It also has no bearing on how long an endorsement takes on your industrial all-risks policy. Keep the two conversations apart when an insurer presents them together.

The useful move is to replace the spend question with a small set of outcome questions that an insurer either can or cannot answer with account-level data. Ask for the last four quarters, for accounts of your size and industry, not portfolio averages.

1. Claims turnaround by line and by claim size. Average and median days from intimation to settlement, split by fire, marine, engineering and liability, and banded by claim value. Averages hide the tail, so ask for the median and the ninetieth percentile together. A single large loss can flatter or wreck a mean.
2. Document extraction accuracy on survey reports. If the insurer runs document intelligence on surveyor reports and estimates, ask what proportion of fields are extracted without human correction, and what the correction rate looks like on handwritten or scanned reports. This is where AI touches commercial claims most directly, and where accuracy claims are rarely audited.
3. Endorsement cycle time. Days from a broker-submitted endorsement request to the issued endorsement, split by type: sum insured increases, location additions, named-insured changes. Endorsements are the highest-frequency service event on a commercial programme and the one buyers complain about most.
4. Cashless authorisation turnaround. For group health, pre-authorisation and discharge authorisation performance against the regulatory windows, at account level. The mechanics of how automation hits those windows, and how it fails, are covered in our piece on straight-through cashless claims.
5. First-response time on a large loss. Hours from intimation above a defined threshold to a named claims manager making contact and a surveyor being appointed. On a serious property or business interruption loss, the first forty-eight hours determine the quality of everything that follows.
6. Reopened and disputed claim rate. How often a settled claim is reopened, and how many claims go to grievance or Ombudsman. This is the quality counterweight to every speed metric above.

Each of these is a number an insurer with real systems can produce in a week. An insurer without them will offer a portfolio average, a case study, or a demonstration. That answer is itself data.

The corpus evidence on claims outcomes points in an uncomfortable direction for anyone tempted to score insurers on spend. Two divergences matter.

The first is between speed and fairness. An automated claims pipeline optimises what it is measured on. If the measured thing is turnaround, the fastest route to a good statistic is a quick query that stops the clock or a quick partial settlement that closes the file. A settled-fast claim and a settled-fully claim are different events, and only one of them shows up in a turnaround report. Any scorecard that carries claims speed without carrying decline rate, partial-settlement rate and reopened rate alongside it will reward the wrong behaviour.

The second is between published ratios and actual outcomes. Settlement ratio is the metric buyers reach for, and it is close to useless on its own because the regulator does not collect the reasons behind repudiations. We have written separately on scoring a group health panel without relying on the settlement ratio, and the same limitation applies to commercial lines. A technology-spend figure, layered on top of a settlement ratio, gives you two weak signals rather than one strong one.

There is also a line-of-business divergence that Indian insurers rarely volunteer. The visible AI results sit in motor and health, where claim volumes justify the model-building, while property and liability claims stay substantially manual. We have examined that split in detail in this analysis of where Indian insurers' AI wins actually sit. If your programme is a manufacturing property and machinery breakdown account, an insurer's headline AI story is largely describing work done for someone else.

A panel review scores insurers on things that change your outcome. Financial security, wording appetite, claims record, service, and pricing discipline. Technology belongs in that structure, but as a service determinant, not as its own headline category.

A practical weighting for a commercial programme:

• Security and solvency: unchanged, and it stays the gate. Nothing else matters if the balance sheet does not. Empanelment methodology is set out in our note on insurer empanelment and security rating.
• Claims outcome quality: decline rate, partial-settlement rate, reopened rate, Ombudsman referrals. Highest weight of the service group.
• Service turnaround: the six tests above, measured on your account, second highest.
• Technology capability: scored only through those measured outcomes, plus two capability questions that do have direct buyer impact, which are whether the insurer can issue policy documents and endorsements through a broker-facing system and whether it can give you claims data in a structured export rather than a PDF.
• Declared technology spend: zero weight as a score. Record it, quote it back when an insurer's service numbers are weak, and use it in the stewardship conversation.

That last point is where the ICICI Lombard number earns its place. An insurer that publicly commits 1.5 percent of premium to technology has given you a fair question at the annual review: this is what you spend, here is what our endorsement cycle time looks like, what changed. That is a legitimate use of a public statement. Scoring it as if it were a service level is not.

Set the technology discussion inside the stewardship meeting rather than the placement meeting, because in placement it becomes a pitch and in stewardship it becomes accountable.

Bring your own data first. Your broker holds the record of every endorsement request, every claim intimation and every escalation on the account. Build the actual timeline from that record before you ask the insurer for theirs, because the two rarely match and the gap is the most useful thing in the meeting. An insurer reporting a seven-day endorsement cycle against your log showing nineteen days is a specific, fixable conversation.

Ask what changed rather than what exists. "What did you deploy in the last twelve months that affected commercial property claims, and what moved as a result" is a question a real programme can answer with a before and after. A capability list cannot.

Ask who owns model decisions. If document extraction or claims triage runs on models, ask who signs off changes, how errors on commercial documents get caught, and what happens when the model is wrong on a survey report. On complex claims, a wrong extraction that nobody catches becomes a wrong reserve, then a wrong settlement discussion.

Put the two or three service commitments that actually matter into the placement documentation rather than the meeting minutes. A promised endorsement turnaround that lives only in a slide does not survive a change of relationship manager.

Finally, close the loop annually. Re-run the same six measurements every year with the same definitions, so you build a trend rather than a snapshot. Insurers improve and regress, and the only way to see either is a consistent measurement you own.

The McKinsey thesis that AI reshapes insurance economics through onboarding cost and claims cycle time is probably right at the industry level, and over several years it will change which insurers can price competitively on which risks. None of that helps you choose a panel next quarter. What helps is a set of measurements specific enough that an insurer cannot answer them with a narrative.

A sequence for the next sixty days:

1. Pull twelve months of your own service record from your broker: endorsement requests with dates, claim intimations with dates, escalations, and outcomes. This is your baseline and it costs nothing to assemble.
2. Send the same six-question data request to every insurer on the panel and to any you are considering adding, with a fixed format and a deadline. Uniform questions produce comparable answers.
3. Score claims outcome quality and service turnaround separately, and refuse to blend them. A fast insurer with a poor decline record should not be able to average its way to a good score.
4. Record declared technology spend in a notes column, with zero weighting, and use it to press for commitments in the stewardship review.
5. Write the two or three commitments you extract into the placement file so they bind next year.

The insurer that reports 1.5 percent of premium going into technology may well be the best service partner on your panel. The number is not what proves it. Your endorsement log, your large-loss first-response times and your reopened-claim rate are.