Stories

Why AI errors multiply in a month-end close, and how an old vendor key opened a security firm's cloud

Warehouse workers can ask an AI questions answered from company documents, and AI can fill a sales deck from customer records while the price stays fixed.

By , Senior AI ConsultantEdition of

4stories
3minute read
In This Edition

Each step works on what the step before it produced, so the chances multiply. Five steps that are each 80% accurate come out right together 33% of the time, which is 0.8 multiplied by itself five times.

An email draft is a single step, and a person reads it before it goes out, so a weak draft costs a minute of editing. A month-end close is a chain. Picture an AI that pulls the bank lines, matches them to invoices, posts the accruals, reconciles the accounts between group companies and writes the commentary. A wrong match in the second step does not stop anything. It looks reasonable, the next step builds on it, and by the commentary it has become a variance that someone must explain to the board.

So the question to ask a vendor is how many steps its AI runs before a person looks. A controller who spends 15 minutes on the unmatched and doubtful items after each step catches an error while it is still one wrong line in a list. Five such reviews take just over an hour, and the AI still does the days of matching and posting.

A rise from 80% to 98% per step saves a few minutes of editing on a draft, but it lifts a five-step job from 33% to 90%. So multi-step finance work will move to AI in jumps, one model release at a time, and each jump takes a review out of the close.

Share this story

The security company Barracuda had given the key to an outside AI firm for one project. An attacker stole it from a settings file in the firm's GitHub project. Artex, an AI hacking tool, then used the key to get into Barracuda's cloud in under two days. Barracuda found the intrusion two days after it ended.

Artex runs on paid AI models, and its users pay with other people's keys. Its agents search the internet for exposed AI apps, copy the keys stored inside, and spend them hunting for more. A chatbot that a team built and left open online, with the company's AI key inside, is a target.

At this speed, closing access protects better than watching for intruders. List every login that outside firms hold, close the finished ones, and set a spending limit on each AI key.

Share this story

A picker can ask whether a damaged carton may still ship, and an AI assistant answers from the company's current documents. The voice picking systems that warehouses have used since the 1990s worked in one direction only: go to this aisle, take this many cases, confirm with a fixed word.

Those questions used to go to the supervisor. At 30 questions a shift and two minutes each, the supervisor gets an hour back. Each question that the assistant cannot answer shows a rule that exists only in a supervisor's head, and that is the rule to write down next.

Share this story

A team uploads its approved PowerPoint to Presentations.AI, marks the sections that change for each prospect, such as the case studies and the price, and links each section to a field in the customer record. The rest of the deck stays as the company wrote it.

The price is copied from the record, so the AI never writes a number of its own. The real decision moves into the template: which lines are locked, which come from a field, and which the AI may write. Whoever owns that template decides what every salesperson shows a customer.

Share this story

THE DAILY BRIEF

Get the next edition in your inbox.

A five-minute read, every weekday morning.

Free forever · Unsubscribe anytime

Share this

Other Editions

Newest first


THE DAILY BRIEF

Read the next one first.

A five-minute read, every weekday morning.

Free forever · Unsubscribe anytime