How a $12k AI strategy workshop saved Northwood Insurance $320,000

The brief

Northwood Insurance is a Canadian regional commercial-insurance carrier with ~$180M in annual premiums. Their underwriting director had a budget approved by the board: $400,000 to build a custom AI underwriting platform. The pitch from their preferred vendor: a multi-agent system using GPT-4 to read submission packages, classify risk, and recommend approval/decline.

The vendor proposal landed on the underwriting director's desk on a Thursday. The board wanted a yes/no by the following Friday. He asked an industry friend, "should I sign this?" — and the friend pointed him to us with one piece of advice: "Get a second opinion before you write the check."

We don't usually publish case studies where we tell clients not to build. We wanted to publish this one.

The constraints we walked into

Budget already approved: $400k was committed; the board would be happy if it was spent
Vendor relationship existed: they'd been pitched 4 times over 6 months
Underwriting team enthusiasm: ~3 underwriters had "won" internal politics to get AI on the roadmap
Real pain point: 14-day submission-to-decision time, way too slow for the brokers who fed them
1-week deadline: board meeting on Friday

The engagement: 2-day strategy workshop

Northwood signed a $12,000 fixed-price 2-day strategy workshop on the Monday. We flew a principal engineer to their Mississauga office for Tuesday and Wednesday. Day 1 was discovery — interviewing 4 underwriters, the IT director, the CCO, and the underwriting director, plus a deep read of the vendor proposal. Day 2 was the working session: whiteboard the actual workflow, model the cost-benefit of each candidate AI intervention, and produce a recommendation.

We spent half of day 1 reading the vendor proposal. Two things were immediately obvious:

The proposal solved 80% of the problem with a sledgehammer. The vendor was building bespoke OCR + classification — work that off-the-shelf commercial-insurance intake tools (DataCubes, Convr, Send) already did at a fraction of the cost. The buy-vs-build analysis in the proposal was incomplete.
The actual bottleneck wasn't AI. Day 1 interviews revealed that underwriters were waiting 8 of the 14 days on broker callbacks, not on intake processing. Building an AI to speed up the 6 days you control while the other 8 days are unchanged moves the cycle from 14 days to … about 10. Not 4, as the vendor implied.

The recommendation we delivered

On day 2 we whiteboarded an alternative path:

$45,000 — buy DataCubes or Convr for the intake/OCR/classification layer. Off-the-shelf, vendor-supported, 4-week deployment.
$35,000 — build a custom broker-portal nudge layer. Where underwriters were actually losing days — automated broker reminders, missing-info detection, status-shareable links. This is the part vendors don't build because it's "not AI."
$0 — DON'T build a bespoke AI underwriting agent. Save the remaining $320k. Re-evaluate in 12 months once the intake layer is live and they have real data on what underwriting decisions actually look like at scale.

Total recommended spend: $80,000 over 6 months. Down from $400k. With a higher probability of moving the 14-day metric to single digits — because we were attacking the actual bottleneck.

The delivered output

End of day 2, Northwood walked away with:

A 9-page written report with the workflow analysis, the cost-benefit model, and the alternative recommendation
Three vendor evaluation rubrics for the intake-platform shortlist (DataCubes, Convr, Send)
A scope document for the broker-portal nudge layer — ready to use as an RFP for any agency
A specific go/no-go criteria document for revisiting the bespoke-AI path in Q4 2026

The board meeting on Friday

The underwriting director presented our analysis alongside the vendor proposal. Board approved the alternative path. Net savings vs the original plan: $320,000.

What happened next

Northwood implemented the intake-platform purchase themselves (90% done by week 6, fully live by week 10). They came back to Paisol for the broker-portal build — a fixed-price web development engagement at $35,000, shipped in 10 weeks. The 14-day submission-to-decision metric moved to 6 days in their first quarter post-launch.

They're now in conversations with us about revisiting the bespoke-AI question in late 2026 — with 12 months of real underwriting data and a clearer view of where AI would actually move the needle.

We were 6 weeks from signing the wrong check. Paisol's workshop cost us $12k and saved us $320k — and the only reason I called them was a friend's offhand advice to "get a second opinion." This is the engagement I wish we'd done first instead of fourth.

— Underwriting Director, Northwood Insurance

Why we publish stories like this

About 30% of our AI consulting engagements end with a recommendation not to build something — or to build less than the client originally scoped. That's the value of vendor-neutral consulting. We're willing to tell you no on a small engagement so you trust our yes on a bigger one.

About 60% of consulting engagements at Paisol do eventually convert to a build. Northwood's broker-portal followed that pattern. The 40% that don't convert are clients who run with our written roadmap on their own — or hire someone else. Both fine.

Numbers

Workshop cost: $12,000 (fixed)
Original vendor proposal: $400,000
Alternative plan total: $80,000
Net savings: $320,000
Follow-on broker-portal build (10 weeks later): $35,000
14-day → 6-day submission-to-decision cycle (improvement: 57%)

Need a vendor-neutral second opinion on an AI proposal?

Paisol's Vendor Proposal Review is a flat $4,500 engagement — we read the proposal, talk to your team, and send a written second opinion within 5 business days. We've saved clients millions by spotting overscoped or under-architected proposals before they get signed.

Or skip straight to a 2-day strategy workshop: book a free 30-minute call to see if it fits your situation. Or learn more about our AI consulting services.