ISSUE 03 / SEP 28, 2026
This week in enterprise AI: what happened, why it matters, what I'd do.
Boring AI wins.
A bank, a drug maker, a dialysis provider and a tech company each tied AI to a number the board already tracks.
|
■ FROM THE PEOPLE DOING IT “We took what used to take two weeks, and we reduced it down to three minutes.” Matthew Prince, CEO, Cloudflare |
|
PATTERN WATCH Three of the four stories tie AI to a number someone senior already watches: hours per mortgage, missed treatments, people-days per earnings release. The fourth, Novartis, is about refusing to count anything else. |
|
|
This week's evidence came mostly from regulated industries. They had to build governance first, and it now looks like a head start. Each story ties AI to a number someone senior already watches: hours per mortgage, missed treatments, days to report earnings.
— Manny
|
| THIS WEEK'S THREE STORIES |
|
|
■ SPONSORED SOURCE · BANKING
A 15-hour mortgage check now takes about 3 minutes.
What happened TD applied AI agents to the first review of mortgage applications. In qualifying cases, the assessment fell from 15 hours to about three minutes. TD has also given Microsoft Copilot to more than 40,000 employees in Canada and the US. Manulife is expanding Copilot to over 30,000 staff, and BMO says it frees about one working day a month for its Nesbitt Burns advisor teams. Microsoft paid the Financial Post to publish this article and the figures weren't independently checked. Nobody paid me to include it.
Why it matters The gain came at one step where applications sat waiting for a person. Most core processes have a step like that, usually a handoff nobody owns end to end.
WHAT I’D DO Map one core process end to end and find the step where cases wait longest for a person. Test AI there first, and keep a route to a human for the cases that don't fit. |
TO MAKE THIS WORK First: Find out why cases wait, because missing data, unclear rules and a person with no time for it each need a different fix. As you do it: Design the fast path and its controls together. Decide which cases qualify and what gets checked, and set the point where a person takes over. Then: Run a share of real applications through the new path and compare the results with the old one before you widen it. |
Read the source →Financial Post, sponsored content for Microsoft Canada, Sep 25, 2026
BY ROLE Transformation lead: Name one owner for the whole journey before automating a step inside it. Technology and data: Get the data into one governed place first, as CIBC did before scaling AI. Finance and risk: Decide which cases qualify for the fast path. TD limits it to qualifying applications. | BEHIND THE NUMBERS CIBC says most of its consumable data now lives in Microsoft's cloud, as the base for AI. A Microsoft-commissioned survey in the same piece found 81% of Canadians want to reach a real person even if their bank uses AI. | HOW SOLID IS THIS Sponsored content paid for by Microsoft. Figures are company-supplied and weren't independently checked. |
|
|
|
■ CASE STUDY · PHARMA
Novartis: “cool does not scale.”
What happened Novartis's data platform holds data from over 3,000 clinical trials and Novartis now uses AI for literature review, molecule design and early safety checks. Christian Diehl, its chief data and digital officer for research, says impressive demos usually haven't been rolled out six months later; AI has to become a boring part of daily work. One lesson came the hard way: the team prepared a dataset nobody asked for. It now starts from scientists' questions and works back to the data.
Why it matters A demo proves an idea can work. Daily use proves it survives reality. Novartis counts only the second.
WHAT I’D DO List your AI pilots older than six months. For each, ask: would the business notice if we switched it off? Keep those and retire the rest. |
TO MAKE THIS WORK First: For each pilot, know who uses it and what they stop doing. If nobody can say, it isn't ready to judge. As you do it: Use one test for every pilot: daily use, a business owner and a measured result. Then: For the pilots you keep, fund training and support before the rollout starts. |
Read the source →McKinsey, Sep 21, 2026
I wrote about this: Your pilot didn't prove what you think it did →
BY ROLE Executive sponsor: Put your name on the decision gate for each pilot: proceed, narrow or kill. Technology and data: Clean the data someone has asked for, not the data you think will be useful. Finance and risk: Govern agents like any regulated process: access rights, audit trails, named points where a person decides. | BEHIND THE NUMBERS Its data42 platform combines 3,000+ trials with real-world data such as health records, and a control tower called Sense monitors trials in real time. Novartis has committed $1B to a San Diego “lab-in-the-loop” site, where it hopes AI ideas will feed automated labs and results feed back. | HOW SOLID IS THIS An executive interview published by McKinsey. |
|
|
|
■ CASE STUDY · HEALTHCARE
AI that flags a missed dialysis appointment before it happens.
What happened DaVita, the kidney-care provider, uses AI to flag patients likely to miss dialysis so caregivers can step in early. Its IT team is also scaling, with clinicians, AI that listens in clinic visits, and has rolled out an AI search tool across medical, insurance and admin records. Lender Finance of America paired its AI strategy with one shared platform for AI work, so it wouldn't end up with a sprawl of separate tools.
Why it matters AI cases tied to a number the board already tracks get funded and stay funded. For DaVita, that number is missed treatments.
WHAT I’D DO Pick one metric your board already reviews and work backwards to the AI use case that would move it. |
TO MAKE THIS WORK First: Check that you can get the data behind the metric, and that someone has authority over the process that drives it. As you do it: Write the case in this order: the metric, then the workflow, then the tool. Then: Get the person who owns the metric to sponsor and fund the work, so the case keeps its budget when priorities shift. |
Read the source →CIO.com, Sep 21, 2026
BY ROLE Executive sponsor: Start every AI business case with the outcome, then work back to the technology. Technology and data: Build one shared AI platform before use cases multiply, as Finance of America did. HR and L&D: Pair data-quality work with career paths for IT staff, as Schneider Electric does. | BEHIND THE NUMBERS Schneider Electric pairs data-quality work with new career paths for its IT specialists. Gartner finds 63% of CIOs name process improvement as a goal for AI, behind business outcomes at 74%. | HOW SOLID IS THIS Reported by CIO.com. DaVita gave no figures. |
|
| THE PROOF IS IN THE PUDDING |
|
32 of 112 32 of 112 companies showed a number for what AI did for their organization: 29%. The rest said they use AI but showed no result: no cost cut, no hours saved, no error rate. One that did: Nestlé cut product ideation time from six months to six weeks by applying AI to end-to-end product innovation. (Harvard Business Review) |
|
Two weeks and 20 people, down to three minutes. Before each earnings release, about 20 people at Cloudflare spent two weeks preparing investor documents. AI tools now do it in about three minutes, and CEO Matthew Prince says investors report fewer errors.
Decoder with Nilay Patel, "Can Cloudflare save the web from AI?", Sep 26, 2026 Act: Anthropic dropped its mandatory 30-day data retention for business customers. After customer pushback, its new Enterprise Frontier Safeguards will let businesses control how their data is stored and reviewed, at no charge, rolling out in phases from later this fall. Anthropic also cut one part of its pricing (reusing cached input) by 75%. Revisit any decision you made on the old terms.
Stratechery, Sep 21, 2026 Nationwide is putting about $300M into AI, after $5B on its core systems. The insurer is spending $1.5B on technology through 2028, about 20% of it on AI. CTO Michael Carrel reports a nine percent rise in how staff rate AI in its engagement survey.
Technovation with Peter High, Sep 24, 2026 |
|
What are you being asked to decide about AI, or where is a change stuck? Send me your question. I’ll answer the best ones in future issues, and I never use your name. |
|
■ TRANSNOVA.AI
Where is your team stuck?
AI isn’t the transformation. Redesigning the work is. Transnova.ai, my firm, finds the workflow that keeps slowing your team down, redesigns it, and measures the result.
|
|
Invite me to speak · Connect on LinkedIn
|
|
