Every carrier and MGA leadership team is feeling the same pressure right now.

Boards expect an AI strategy. Competitors are announcing pilots. Investors want to see movement, not a plan to make a plan. Standing still feels like falling behind, even when nobody can quite say what the operation actually needs.

That pressure pushes a lot of teams toward the same decision. Buy a tool first. Figure out how it fits the workflow later.

OIP Insurtech sees what happens after that decision gets made, and the result rarely matches what leadership expected when they signed the contract.

The order matters more than most people realize. Buying technology before understanding the operation underneath it is a predictable way to end up with a disappointing return on an AI investment, no matter how good the tool actually is.

Why the Tool Comes First

The instinct to buy the tool first makes sense. It isn’t careless, and it isn’t a sign that leadership hasn’t thought things through.

AI vendors move fast. Sales cycles are short, demos are polished, and the value on screen looks immediate and obvious. A vendor walks in, shows a clean submission getting processed in seconds, and the room sees exactly what they came to see.

There’s also a real pressure to show progress. A signed contract feels like progress in a way that an internal audit never does. One produces a press release. The other produces a spreadsheet nobody outside the operations team wants to read.

Board and investor pressure make this worse. Having an AI strategy has become its own kind of credential. Something leadership can point to in a meeting regardless of whether it’s actually working. The optics of “we adopted AI” tend to move faster than the substance of whether that adoption was built on solid ground.

There’s also a quiet assumption baked into a lot of these purchases. The tool itself will reveal what’s wrong with the workflow once it’s running. Buy it, deploy it, and the gaps will surface on their own.

In practice, that’s rarely how it plays out. OIP Insurtech sees this pattern across client conversations regularly, and it isn’t a criticism of any single decision. It’s a sequencing problem that’s easy to fall into under pressure to move fast.

What Actually Happens When AI Lands on an Unmapped Workflow

The pattern OIP Insurtech sees after implementation tends to follow the same shape, regardless of the vendor or the specific tool.

The technology performs well on the clean, predictable submissions it was built and trained to handle. Standard formats, complete data, straightforward risk types. On those, the results often match what the demo promised.

Everything else is where the trouble starts. Non-standard submissions, inconsistent formats, document types the tool wasn’t tuned for, these make up a larger share of real volume than most leadership teams expect going in. The pilot phase usually runs on curated test documents that don’t represent the full mess of an actual production inbox, so the gap only becomes visible once real volume hits the system.

Manual workarounds appear almost immediately once that gap shows up. Someone on the team starts handling the exceptions by hand, then starts handling more exceptions, and within a few months the “automated” workflow has a parallel manual process running alongside it that nobody officially designed.

At that point, it becomes genuinely difficult to tell what’s actually wrong. Is the tool underperforming? Or was the underlying workflow simply never restructured to support the kind of automation that was purchased? Most teams can’t answer that question with confidence, because nobody measured the workflow before the tool arrived.

The end result is a technology cost added on top of an operation that’s still carrying its original inefficiencies. Leadership now has a new line item, a layer of added complexity, and a workflow that looks automated on paper but still depends heavily on manual intervention underneath. McKinsey’s research on AI implementation in financial services has repeatedly pointed to this exact pattern, where the technology itself is rarely the reason an initiative underperforms.

The Question No One Asks Before Signing the Contract

There’s a question that almost never comes up during AI vendor evaluation, and it’s the one that matters most: do we actually know where our operational inefficiency lives right now?

Most leadership teams can sense there’s a problem somewhere. Turnaround is slower than it should be. Headcount keeps growing faster than premium. Submissions are slipping through the cracks during peak periods. But sensing a problem and being able to quantify it are two very different things, and very few operations have a real baseline to work from.

Without that baseline, there’s no way to know which parts of the workflow are actually good candidates for automation and which parts need to be fixed first.

Some types of work tend to automate well:

• Submission intake and data extraction
• Document classification and normalization
• Basic compliance checks
• Any task that follows defined logic and produces predictable outputs

Other types of work behave very differently, and don’t benefit from automation the same way:

• Risk assessment and underwriting judgment calls
• Broker relationship decisions
• Complex, non-standard exceptions that require context

Trying to force the second category into an automated pipeline usually creates more problems than it solves.

This is the step that gets skipped under pressure to move fast. It feels like a delay when leadership wants to show momentum, and it doesn’t come with a vendor demo or a press release attached to it.

The asymmetry here is significant. Skipping this step doesn’t save time in the long run, it just moves the cost downstream. The wasted technology spend, the manual workarounds, the rework required to fix a deployment that landed on the wrong workflow, all of it ends up costing far more than the measurement step would have cost upfront.

What Measuring the Operation First Actually Looks Like

The Workflow Intelligence Diagnostic exists to answer the question raised in Section 3, before a technology decision gets made rather than after.

It’s an eight week engagement that maps real workflows through live shadowing and stakeholder interviews, quantifies where time and cost are actually being lost, and produces a clear view of what’s genuinely ready for automation and what needs to be fixed first.

The engagement runs across four phases:

• Weeks one and two focus on observation: workflow shadowing, system walkthroughs, and data intake to understand how the operation actually runs day to day
• Weeks three and four quantify the key metrics that matter, including time to quote, quote-to-bind ratio, underwriting capacity, and document handling performance
• Weeks five and six produce maturity scoring, a benchmark gap analysis, and AI readiness mapping across the workflow
• The final two weeks deliver an executive readout with a clear, sequenced decision path for leadership

The output gives leadership something concrete to act on, including:

• A friction profile showing exactly where time and capacity are being lost
• A submission economics breakdown that quantifies cost per quoted submission
• An AI readiness map identifying which parts of the workflow are genuinely ready for automation
• A transformation roadmap sequenced by impact, not just by what’s easiest to implement first

Engagements to date have surfaced more than 20 percent efficiency opportunities in core workflows, with some clients reporting returns of up to five times the cost of the engagement within the first year.

This is the step that belongs before a vendor contract gets signed, not after. An AI investment that lands on a workflow that’s actually been mapped and measured performs differently than one that lands on a guess.

Conclusion

The order of operations matters more than the technology itself.

Buying AI before understanding the workflow underneath it is a common mistake, and an understandable one given the pressure most leadership teams are under to show progress quickly. But it’s also a predictable way to end up with a disappointing result, regardless of how capable the tool actually is.

Measuring the operation first changes what an AI investment actually is. Instead of a bet made on a demo and a sales pitch, it becomes a decision backed by data about where the inefficiency actually lives and what’s genuinely ready to be automated.

Before the next technology decision gets made, it’s worth knowing exactly where the operation stands today. That clarity is what makes every investment that follows more likely to deliver what it promised.

Contact the OIP Insurtech team to learn more.