The technology buyer of tomorrow

A conversation about how AI might change the way insurers buy software, and the picture it left me with.

A few days ago Michael Price, an IT sales recruiter at Journey Business Group, posted on LinkedIn about how buyers are changing, and the back and forth that followed in the comments turned to how AI might change the way businesses buy software altogether. By the end of it I had a picture of a buying process for core insurance software that looks almost nothing like the one we run today, and I’ve been turning it over since.

This is that picture. Parts of it exist already and the rest is a few years away. It’s worth thinking about if you buy insurance technology, or sell it.

How we buy today

Buying a core insurance platform follows a familiar route. You write a requirements document, usually a few hundred questions long, and send it to a longlist. You sit through demos built to show each platform at its best and take references from customers the vendor has chosen for you. Then you make a decision you’ll live with for the next decade, largely on trust.

It’s rarely one person’s call either. Forrester reckons the typical buying decision now involves 13 internal stakeholders,¹ which is a lot of people leaning on the same few demos.

The process has two weaknesses that nobody really solves. The first is that you rarely see the system do your work. You see it do a version of work the vendor has prepared, on data the vendor has cleaned, presented by someone who is very good at presenting.

The second is easier to miss. The requirements are usually a description of the system you already have, because when you ask a team what they need, most will describe their current workflow with the annoying bits taken out. You end up buying a newer version of the past.

Step one: an AI that understands your business

Now imagine an AI model that understands how your business works better than you do.

That sounds far-fetched until you look at what’s already possible. Process mining and task mining tools record how work actually flows through a business, from system logs and from the screens people work on, and then analyse it. When I was at Automation Anywhere, this kind of process discovery was already part of the automation story. The technology to watch how work gets done and find where it slows down is real.

Take that further and picture a single model that has mapped every process in your business and analysed how every user works. It knows where your teams lose the most time, which tasks cause the most rework and every role people need to fulfil in a day. Think of it as a custom GPT for your own operation.

Step two: give it the keys to every vendor

Instead of watching a demo, the model gets back-end access to every platform on your shortlist and does the work itself. It builds your actual products inside each system and recreates the workflows your teams rely on, including the reports your board reads. Then it reports back in plain terms:

• These two systems can’t do what you need.
• This one can, with caveats.
• This one handles your process 20% faster.
• That one costs 30% less to run.
• This one is the strongest overall, but it’s missing a piece of core functionality you depend on.

You’d have all of it in one sitting, built from how your business really operates. You wouldn’t need to trust a claim, because you’d have watched it work.

It also deals with the second weakness. Because the model has studied where your current processes break, it won’t simply rebuild them. It can suggest how you should be working and test each platform against that instead.

If that sounds fanciful, people are already pointing AI at the hardest end of insurance systems. McKinsey has written about AI agents doing core modernisation work in days that would take a specialist months.² Building a product inside a vendor’s platform isn’t such a leap from there.

Step three: bring in the CFO

Now plug in the finance data: headcount, salaries, who works on what and how long each task takes. Add years of quote volumes and strike rates.

At that point the evaluation turns into a business case. Suppose most of your cost sits in claims. A platform that looks weaker on paper might cut claims handling time far enough to beat the obvious favourite. Or suppose your data shows you can’t handle twice your current quote volume. A system that lets the same team process more quotes, at your historical strike rate, now has a revenue figure attached to it.

The model could come back with a number for each platform, showing that this one returns X over five years and that one returns Y, and break it down by function so you could see where each platform saves money and where it makes it.

Step four: buy the best of each

This is where the logic leads somewhere vendors may like less. If the model can score each platform function by function, why buy all of it from one supplier?

It might recommend policy administration from one vendor and claims from another. It has already tested the integrations and could write the code that connects them. It could even adapt the software itself, although that reopens the old build-versus-buy argument and the question of who maintains it afterwards.

Suite vendors, ourselves included, have always sold partly on the cost and risk of integration. If a model has already tested the joins and written the code, much of that argument goes. What’s left is the value of one record, one contract and one supplier to call when something breaks on a renewal day. That value is real, and I’m not convinced every insurer would want a patchwork of suppliers, however well chosen. But the model could put a price on that convenience too. Some insurers will look at the figure and decide it’s worth paying. Others will find it’s worth a good deal less than the premium they’ve been paying for it, and that is the conversation suite vendors should be preparing for now.

Let me make it concrete

Picture an MGA writing commercial van insurance for two capacity providers, with three platforms on its shortlist and a renewal season it can’t afford to disrupt.

In my version of the future, the model spends a few weeks learning how that MGA really quotes, binds and handles mid-term changes. It builds the van product, the rating and the referral rules inside all three platforms. Then it replays last year’s business through each one and sets the results against what the team actually costs to run.

The report that comes back reads very differently from a scoring matrix built on the vendors’ own answers. It says something like: this platform produces the bordereaux your capacity providers want without anyone rekeying a thing, and that one needs a workaround every month. A demo would never have told you that.

What changes for buyers

The biggest change is where trust comes from. Today it comes from people, meaning the salesperson and the reference customer. Tomorrow more of it comes from evidence about how each platform would handle your own business.

Buyers would also need to get comfortable with letting a model see everything. That raises real questions about data security and access, and about how far you trust the model’s own judgement. Models make mistakes, and a confident wrong answer about a ten-year platform decision is an expensive one.

What changes for vendors

For vendors, the demo loses its power. A polished presentation counts for little if the buyer’s AI has already built the product in your platform and found the gaps.

What gains power is everything the model touches. Open APIs and good documentation become sales assets, and configuration a machine can understand matters as much as configuration a person can. Vendors who hide their weaknesses behind good demos will struggle.

That cuts both ways for us. A model with back-end access to our platform would find our gaps too, and it would find them faster than any prospect does today. I’d still take that trade. A gap a model finds before a sale is far cheaper to fix than one a customer finds after it, and I’d rather be judged by something that has used the system than by whoever gives the slickest presentation.

What changes for salespeople

I wrote recently that complex-sale salespeople would be among the last roles AI changes, and I still think people will buy from people for a while yet. But the job shifts towards helping a buying committee make sense of what the model found and deciding what to do about it.

There’s an early hint of that already. Gartner found that 69% of B2B buyers would rather check what AI has told them with a salesperson.³ The best salespeople have always done that anyway.

How far away is this?

Probably a few years. The process mapping exists today. The part where a model configures products inside a vendor’s platform is further off, and buyers would need to trust the result.

Gartner is bolder than me and predicts that by 2028, 90% of B2B buying will be AI agent intermediated.⁴ Insurance tends to move more carefully than that, and I suspect the first versions will be messier than the picture I’ve painted here.

If I were buying today

None of this exists in one neat package yet, but if I were running a platform selection tomorrow I’d borrow from it anyway. I’d map how the work really flows before writing a single requirement, and I’d write those requirements as outcomes rather than features. I’d ask every shortlisted vendor for their API documentation and notice how readily it arrived. I’d ask the finalists to build one of my real products instead of showing me theirs.

And I’d watch very closely how each of them handled the bordereaux.

Show References Hide References

1. Forrester. The State Of Business Buying, 2026. 2026.
2. McKinsey & Company. Can agentic AI (finally) modernize core technologies in insurance?. 2026.
3. Gartner. Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights. 2026.
4. Gartner. Gartner Unveils Top Predictions for IT Organizations and Users in 2026 and Beyond. 2025.