Where the time saved goes

For leaders responsible for a software product that customers, advisers or staff rely on. An essay worth your time, plus the sharpest things I've read, filtered hard. No hype.

Iress and Signal Advisory surveyed 839 financial advice professionals in Australia and New Zealand this year. Of those using AI, 82% say it saves them real time on file notes. Only 20% say it has lowered the cost of serving a client.

58% of all respondents take four weeks or more to get advice to a client, and the four weeks hold at every practice size. In practice, that usually means a client file waiting for the next person to pick it up, or for someone to check it again.

If you run a product that advisers or staff rely on, this is the gap to look at: minutes saved on one step, against weeks of waiting for the client. This issue follows the time saved, and looks at what advisers paid for when a tool built into their software stopped halfway.

The best things I read this fortnight on building product. I read them so you don't have to. Click the ones that earn it.

Building product in the age of AI · 11 min
Stop Asking Where to Put AI Agents. Map the Real Process First
By Jekaterina Jegoscenko, engineering manager, previously an IT process manager at a multinational telecom company.
Jegoscenko went looking for the official incident process and found three documents: one from 2010, one from 2014 and one on an internal wiki. None matched the work. So she asked each team to draw how a ticket actually moved. The delay sat in the hand-off to a specialist team in another country, and one status, "Pending", meant four different things. Her rule: "your documented process is rarely how the work actually gets done." Her closing question is the one this issue's check asks: "Show me where the ticket actually stops."

Craft & judgment · 6 min
Everyone Is Learning to Grade AI. Who Owns the Answer Key?
By Michael Albers, former product lead on search at Yahoo! and later a chief product officer.
Albers's AI agents worked overnight and returned ten code changes, each with a preview of the page. The report said every preview built without an error, and two review passes had approved all ten. "Nobody had opened a preview in a browser, until I did. Every one of them was a blank page." The standard said the code had to build, and never said the page had to show anything. His answer: whoever runs product owns the written standard for what good looks like.

Building the business · 8 min
Faster AI delivery needs time from the business
By Richard Hogan, Global Chief Architect in IBM's Microsoft Practice.
Hogan starts from a working paper covering 718 firms. After those firms adopted coding agents, they wrote about 30% more code. The work completed and deployed did not change significantly. Code changes took around 49% longer to get through review. The paper has not been peer reviewed, and Hogan says to hold some of it loosely. The faster step lands on people who already have a full-time job, such as the owner who signs off a requirement. His line for anyone planning the work: "Review and exception handling are real work. Treating them as spare capacity is an excellent way to build the next queue."

The counter-argument · 3 min
Kitces: AI Gives Advisers Depth and Quality, Rather Than Efficiency
PLANADVISER's report of a Q&A with Michael Kitces, financial planner and industry commentator.
The essay looks for the time saved in a lower cost or a faster result for the client. Kitces expects it to go into deeper advice: "What I see is a lot of tools that might save us 30 minutes or an hour in plan prep and analysis, and then create one, two, three more hours of additional conversations and implementation actions with clients." Advisers once served hundreds of clients each, and many now serve closer to 100 households. If he is right, asking whether AI lowered the cost of serving a client is the wrong question.

Pick one task your users already use AI for, inside your product or outside it. Follow one real client file from that task to the moment the client receives the result. Write down every hand-off between people or systems, and every time someone checks or re-types a record, and time each one. The longest wait is where your product should act first.