Approvals

Two hosts work through the single distinction that makes approval decisions tractable — irreversibility, not importance — and the four practical fixes that speed up a stalled approval without removing the human from it.

A: The quote is finished at 11am. It goes out at 4.

B: And nothing happened in between, that's the strange part. The work was done. It just sat in a queue waiting on one person who was on site, then in a meeting, then had ninety unread emails.

A: Meanwhile the customer reasonably concludes you weren't that interested.

B: This is the workflow where AI looks most attractive and is honestly most dangerous — because the obvious fix is, let the machine send it.

A: And the arithmetic looks unanswerable on paper. Remove the person, remove the bottleneck, response time collapses from hours to seconds.

B: What the arithmetic doesn't show is the distribution of outcomes. Most automated replies would be fine. A small fraction are wrong in expensive ways — a price you can't honour, a deadline you'll miss.

A: And you don't find that out from a log. You find out from the customer, usually late, and by then it's been repeating the mistake for a couple of weeks.

B: So — keep the gate. Attack the delay a different way. Which starts with a distinction that, once you see it, makes most of these decisions easy.

A: Approval is about irreversibility. Not importance.

B: An internal report can be extremely important and still not need a gate — if it's wrong, you fix it, and nobody outside the business ever knows.

A: A two-line reply to a customer is far less important in the scheme of things, and it absolutely needs one, because you cannot unsend it.

B: Most teams get this backwards, actually. They gate the important stuff and automate the irreversible stuff, because importance is easy to feel and reversibility takes actual reasoning.

A: So what fixes the stall, once you've sorted things that way?

B: Four things, roughly in order of impact. First — make the decision small. Most of the delay isn't deliberation, it's context-loading. Put the draft, the original request, and the history on one screen, and the decision takes seconds.

A: Second — batch, don't interrupt. Twelve separate notifications is twelve context switches, and they get ignored. One queue, cleared twice a day, actually gets cleared.

B: Third — widen who can approve. Very often one person is the bottleneck only because nobody ever defined a second approver. A value threshold — under some amount, a team lead can sign off — removes most of the delay without removing control.

A: And fourth — auto-approve the boring tail, explicitly. Appointment confirmations, "we got your message." Carve those out deliberately, write the list down, review the list.

B: That last one's worth being careful about — a named carve-out list is a completely different risk than removing approval by default because it was in the way.

A: Where does the human stay, through all of this?

B: On anything that leaves the building or commits the business. Customer replies, quotes, prices, dates, refunds. If it's wrong, the customer finds out before you do — that's the actual test.

A: And the log matters more than it sounds like it should. What was drafted, who approved it, what they changed, when it went out.

B: When a customer disputes what they were told, "here's the message, who approved it, and when" ends the conversation in a minute instead of an afternoon.

A: One honest limit — approval queues need a service level or they rot. If nothing enforces "cleared twice a day," the queue just becomes the new place work goes to die.

B: One thing to check this week, then.

A: Take one recent quote or reply. Time how long it sat, finished, before it actually went out. Not how long it took to produce — how long it waited once it was ready.

B: If that number's in hours and the cause was availability rather than genuine deliberation, that's this workflow.

A: Written version's on the site — atinnovators.in, under insights.