AI Knowhow Episode 94 Overview
- POC ≠ Product: the bar for testing out something new with AI is low; the bar for adoption is workflow-level usefulness.
- Treat AI as a product: think real users, real stakes, and real incentives to change behaviors.
- The product IS the workflow: integration into the day-to-day drumbeat of business beats revolutionary features.
- People > tech: AI isn’t the “easy button.” Change management and incentives make or break launch.
- Be “BORING” on purpose: Aim for predictable value and reliability before you chase flash, says Jim Garrity of SingleStone.
Getting an AI proof of concept off the ground? That’s relatively easy. Turning that initial spark into a product people actually use, love, and adopt is where the real work happens. In this episode, the team gets practical about what it takes to move beyond demos or lightweight POCs: choosing the right problems, treating AI like a product (not a project), designing for workflow integration, and managing the human side of change. Jim Garrity, the Chief Client Officer of SingleStone Consulting, also joins us to explain why being a little “boring” about AI may not be a bad thing.
Roundtable Highlights
1) Stop thinking in terms of demos. Start shipping products.
POCs are cheap to start, and easy to ignore. What separates teams that graduate to product? They pick a specific job-to-be-done, identify who actually benefits, and design for daily workflow. That framing raises the stakes and creates natural pull from the business.
If you’ve already seen your fair share of POCs that generate a lot of excitement up front, only to quickly fizzle out, then try this:
- Start with a high-friction task owned by a named role.
- Define “productive use” (e.g., replaced a manual step, reduced decision latency).
- Instrument it: log usage, completion rates, error/fallbacks.
2) The adoption triangle: problem × product mind‑set × incentives
Teams stall when any one of these core pieces is missing. Picking the right problem for your context matters as much as adopting a Product mindset (deliver value quickly, learn fast) and aligning incentives (leaders get value from adoption, not just launch).
Signals you’re on track:
- You have a one‑sentence problem statement tied to a business metric.
- You can name the user and the moment of use.
- Your rollout plan changes someone’s process on Day 1 (training, comms, SOP updates).
3) The product is the workflow
Model quality and a slick UI won’t save you if the AI product you’re imagining doesn’t fit how people actually work. The team underscores that workflow integration is the hardest part—and the real product.
Design checkpoints:
- Where in the flow does AI enter? Who hands off to whom?
- What’s the fallback when the model is wrong or uncertain?
- What telemetry proves it’s helping (time saved, accuracy, satisfaction)?
Expert Interview: Jim Garrity of SingleStone
Jim Garrity, Chief Client Officer at SingleStone Consulting, joins us to discuss their deliberately BORING framework to implementing AI and how they put it into practice with a number of specialty insurers. These are companies who have to underwrite risk on products that have no real direct comparisons, so they need to be able to consume vast quantities of information to accurately predict risk. Because of its ability to absorb far more information far faster than humans, this is the kind of work that AI is uniquely suited for.
Jim recommends starting with dependable, readily automatable work that compounds trust and ROI. In a hype‑heavy market, “boring is good” means:
- Favor reliability over flash. Pick use cases with stable data, unambiguous outcomes, and clear owners.
- Design for human + machine. AI augments judgment and prep, not just clicks.
- Keep value measurable and repeatable. Leaders fund what shows up in KPIs and customer experience.
Why it resonates for services leaders: The path from idea → income is shorter when you solve frequent pains in client delivery, risk, and more.
In the News: AI flags hidden heart disease with 77% accuracy
For our In the News segment, Pete Buer and Courtney Baker unpack a recent story on an AI tool called EchoNext that was developed by researchers at Columbia University and NewYork‑Presbyterian. It uses ECGs to triage who should get an echocardiogram, reporting that it can detect a “hidden” heart disease 77% of the time, far outperforming human cardiologists. The executive takeaway beyond healthcare? Early‑detection patterns like this can be fare more broadly applicable (financial risk, supply‑chain bottlenecks, employee burnout) when you have signals and workflows ready to act.
Watch the Episode
Watch the full episode below, and be sure to subscribe to our YouTube channel.
Listen to the Episode
You can tune in to the full episode via the Spotify embed below, and you can find AI Knowhow on Apple Podcasts and anywhere else you get your podcasts.
Show Notes
- Connect with SingleStone CCO Jim Garrity on LinkedIn
- Learn about SingleStone Consulting
- Connect with David DeWolf on LinkedIn
- Connect with Mohan Rao on LinkedIn
- Connect with Courtney Baker on LinkedIn
- Connect with Pete Buer on LinkedIn
- Get a guided Knownwell demo
- Follow Knownwell on LinkedIn
