Contents · 7 min read
01 / 06 · Speed floor
Why is quote speed no longer enough?
In short
Faster quotes can help you keep broker placements, but competitors can improve their speed too. You need to match the new pace before you can build an edge beyond it.
You see the problem when a broker gets a useful quote from a rival while your file is still waiting. AI-native firms are already binding risk in half the time and at lower cost, so matching that pace keeps you in the placement but does not give your leadership team an edge.
That pressure is real, so speeding up your queue is a sound first move. It can cut delay and help your team answer brokers while the risk is still live.
But if you only automate your existing queue, you automate the same queue that every rival can automate. Once the market catches up, faster work stops setting you apart.
Your company still needs speed because brokers have felt what fast service is like. The next issue is whether the model behind that quote can defend your position.
02 / 06 · Bought tools
Can the pricing model defend your position?
In short
A model that your competitors can also buy will not protect your place for long. Your advantage must come from data and judgment that belong to your firm.
A good pricing model can improve the work, and your underwriters should use it. The problem is that the same vendor may sell much the same tool across the market.
The insurance technology briefing draws the key line. A vendor model can become common, while your own loss data and the reasons behind expert overrides remain yours.
Your book contains risks you chose and claims that later proved those choices right or wrong. It also contains moments when an underwriter saw that clean data hid a poor risk.
That knowledge often sits in the heads of senior people, so it can leave when they do. A model alone cannot save what your team never records.
The next step is to turn what your own book teaches you into a better choice.
03 / 06 · Your book
When does your own book become an advantage?
In short
Your book becomes useful when each risk choice is tied to what happened later. You also need the underwriter's reason whenever the model's answer was changed.
Start with one risk your team can follow from the first review through any later claim. Keep the model's answer and the final decision in the same record.
Record the first choice
Save the risk facts your team used and the price the model gave.
Capture the change
If an underwriter changes the answer, record the reason in plain words.
Add the result
Link the decision to the claims that follow, including a result with no claim.
Review the next choice
Check whether the old reason should change how your team views a similar risk.
This matters because a senior underwriter may spot when the data gives the wrong view. The insurance CEO briefing warns that poor risk selection can stay hidden until losses appear much later.
A long record does not prove that your firm is learning. The record helps only when the people who see the outcome can reach the people making the next choice.
04 / 06 · Shared data
How does the result improve the next decision?
In short
Underwriting and claims must work from shared facts, because the claim tests the earlier choice. Without that link, your team collects more data but learns little from it.
Your underwriter sees the risk before it is bound, while your claims team sees how it behaves later. Both views are needed to judge whether an override was sound.
A faster feedback cycle can change how your team selects risk. It can also change price when the same pattern appears again.
Have the two teams review one set of linked cases each month. Ask where an override helped and where it hurt. Then change the next review rule only when the result supports it.
That process can make your team's judgment more consistent, but it also gives AI more weight in a key decision. The next step is to make clear who remains in charge.
05 / 06 · Human control
Who remains accountable for the decision?
In short
Name one owner for every AI use that affects a key underwriting choice. A named person must be able to stop or override it, and your firm must keep a record of what happened.
Human review is useful only when the person has real power. A sign-off added after the system has acted does not give your firm control.
The EU AI Act checklist sets out a plain rule for serious uses. Name the person who can step in and state the exact point where they can stop or change the result.
Use that rule in your underwriting work:
Have the business owner name the decision
State which risk choice the AI will shape and why the use is worth the risk.
Have the chief underwriter set the limits
Write when a person must review the answer and what the system may never decide alone.
Give a named underwriter override power
Make sure that person can stop the decision before your firm is bound.
Have risk or compliance keep the record
Save the model's answer and the human choice, with the reason for any change.
Have the owner review the result
Check later claims and decide whether the use should continue.
This keeps judgment with your firm while AI prepares a better view. It also gives your board evidence that the work is controlled, which AI strategy for boards treats as a management duty.
With ownership set, you can test whether one use creates learning or merely saves time.
06 / 06 · First use
Where should your team start?
In short
Start with one key underwriting decision that you can track through to claims. Expand only when the result helps your team make the next decision better.
Choose a decision with a clear owner and a result your claims team can trace. Keep the first scope small enough for your leaders to inspect each choice.
Run this test before you expand:
Speed
Did the time from submission to quote fall enough to match the market?
Judgment
Did your team record why a person changed the model's answer?
Result
Did later claims show whether that change helped risk selection or price?
Learning
Did that result alter the next decision in a clear and approved way?
If only the first answer is yes, your team improved the queue. That gain is useful, but it is not yet hard for a rival to copy.
If all four answers are yes, repeat the test with more cases before adding another use. More cases do not automatically mean better decisions.
The essay on human judgment makes the wider point: automation saves work, while AI can also help your best people apply their judgment. Your test should show which result you are getting.
That proof gives your board the basis for the next decision on underwriting AI.
The next underwriting decision
Your management team now has to decide whether its first underwriting use will merely shorten quote times or create a controlled learning loop from risk selection through claims and back to the next decision. GRAIL's insurance briefing puts that choice in the wider context of what AI changes for an insurance CEO, including why faster service is becoming expected while your own loss data and expert judgment can remain harder to copy. Use it to frame the decision before your team expands beyond the first test.
What AI changes in insurance, written for the CEO. Free to read, no form.
Questions CEOs ask about this
Can an insurance company use AI to build a defensible underwriting advantage rather than just process applications faster?
AI can create an underwriting advantage when your firm links each risk decision to later claims and uses the result to improve the next choice. You may assume a faster quote or a stronger bought model will set you apart, but competitors can match both. Your decision is whether to improve the queue or build a record that makes your underwriting judgment harder to copy.
Which proprietary signals can improve risk selection?
Start with loss results from risks you bound and the reasons behind each expert override. Broker links can add useful context when your firm owns the rights and can connect the signal to a clear decision.
How do we separate genuine learning advantage from model complexity?
Ask whether a past result changed the next risk choice or price in a clear way. If the model grew more complex but your team's next decision stayed the same, you have not shown a learning gain.
Can faster feedback change product design and pricing?
Use claims to change risk selection and price when they reveal a repeated pattern. Keep product design outside this first test.
What governance is required for consequential decisions?
Name one business owner and the person who can stop or override the AI use. Keep the model's answer and the final decision, then monitor later results through an audit trail.
Which underwriting decisions should remain human?
A person should retain decisions that require judgment about misleading data or an unusual risk. Your chief underwriter should set those limits and approve any change as the evidence grows.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- Insurance CEO briefing: why speed is rising and expert underwriting judgment remains hard to copy. (accessed October 29, 2026)
- Insurance technology briefing: why owned loss data and override reasons matter more than a vendor model. (accessed October 29, 2026)
- Insurance commercial director briefing: why underwriting results and sales results need shared operating data. (accessed October 29, 2026)
- EU AI Act checklist: how to name an owner and give a person power to stop or override a key AI use. (accessed October 29, 2026)
- AI strategy for boards: why the board needs a clear position on AI and its control. (accessed October 29, 2026)
- The Only Asset AI Can't Build: why helping people apply judgment can create more value than removing their work. (accessed October 29, 2026)
Next scheduled review: February 13, 2027