SignalsOperating intelligence
Open navigation

Operating question

The useful question is not whether AI made a task faster, but whether a team can explain the cost, test the handoff and recognize the judgment that shaped the result.

AI Operating Models

3 Things AI: The “Where Did the Time and Money Go?” Edition

3 Things AI 5 min3 sources

For

Leaders and workflow owners

You will leave with

3 operating decisions

Reading mode

5 min · 3 verified sources

Reading guide3 decisions · 4 sections+

Decision points

  1. 01Pair one AI workflow's provider cost and review time with a business measure before building a detailed allocation model.
  2. 02Test an onboarding transformation against messy completed cases and keep exception acceptance with the accountable expert.
  3. 03Name the recognizable brand elements a human reviewer must preserve before asking AI to create more campaign options.

Three practical ways to connect AI cost to an outcome, test a difficult handoff before scaling it and keep creative judgment visible in faster work.

AI can save an hour and still leave a business wondering whether the work improved. The missing piece is often a simple comparison: what did we spend, what became easier, and which part still needed a person to make the call? Three developments this week offer practical ways to run that comparison.

1. Make the AI bill answer one business question

A small software firm pays for a team assistant, model API calls and cloud infrastructure. Finance sees three bills. The product lead sees faster releases. Neither can say whether the extra model use reduced support work, improved delivery time or simply produced more activity.

On November 9, IBM announced new Apptio connections and planning features for AI spending. IBM says the additions can bring model consumption and token data into broader technology-cost reporting, support forecasting and connect AI investment with business outcomes. Some capabilities are generally available, while others are planned for later quarters. This is a product announcement from IBM, not proof that every organization can calculate a clean return.

The benefit is a shared view for the people approving spend and the people using the tools. The tradeoff is that better cost detail can create false confidence about value. A token total is measurable; the avoided delay, better decision or customer outcome may be harder to isolate.

Pick one recurring AI-assisted workflow and give it one business measure for four weeks: minutes to resolve a support request, proposals reviewed per week or exceptions caught before delivery. Put provider charges and staff review time beside that measure. Do not allocate every platform cost on day one.

2. Test the messy handoff before automating the whole journey

A bookkeeping practice wants to speed up new-client onboarding. The neat demo reads a standard spreadsheet. Real clients send renamed columns, missing dates, scanned statements and explanations buried in email. Automating the welcome message is easy; rebuilding trustworthy starting data is the difficult handoff.

Northern Trust and Microsoft reported on November 9 that an AI-assisted onboarding workflow transformed historical transaction data into a standard format. The companies say measured preparation time fell from about two and a half weeks to three days. They also report a 96.5% first-pass F1 score, a quality measure that combines precision and recall, across seven representative anonymized test cases. Analysts reviewed outputs, handled exceptions and approved the work before go-live. Those figures come from one large financial institution and should not be treated as a forecast for a smaller firm.

The benefit is a clearer pattern for choosing where AI helps: transform a bounded input, compare it with a reference and route exceptions to an expert. The tradeoff is the work required to create representative test cases and trusted answers. If the examples are too clean, a strong score may say little about the files customers actually send.

Take ten completed onboarding cases, remove personal information and list the five variations that caused the most rework. Test the proposed workflow on those cases before connecting it to a live intake. Record correct fields, incorrect fields, missing fields and the time a reviewer spends resolving each exception. A reasonable option is to automate only the transformation step while keeping acceptance with the person accountable for the client record. Ten cases will expose obvious failure modes, not every future one.

3. Keep the part of creative work customers recognize

A two-person retailer can now make product images, short videos and campaign variations faster. After a few rounds, everything is technically polished but sounds like every other promotion in the category. The time saving is real; so is the feeling that the brand disappeared during the last mile.

Adobe published a November 9 survey of more than 2,000 U.S. creators about creative AI. Among creators using AI, 68% said they at least sometimes settle for a result that does not match what they had in mind, while 77% of all respondents said the final creative decision should remain with the human creator. Adobe and The Harris Poll conducted the survey in October. It reflects self-reported U.S. creator experience, not Canadian customer response or an independent product trial.

The benefit is permission to use AI for more drafts and technical chores without pretending that speed is the whole job. The tradeoff is that review can become a vague request to “make it feel right,” which adds cycles without protecting anything specific.

Before the next campaign, name three elements a customer should still recognize: perhaps the plainspoken opening, the product shown in real use and one detail only your team would know. Let AI vary the rest. Ask one person who knows the brand to reject any draft that misses one of those elements, even if it looks polished. This will not prove which creative will sell best, but it makes the human decision visible and repeatable.

The bigger pattern

These developments point to a balanced test for useful AI work. Cost matters, but only beside an outcome. Speed matters, but only against difficult examples. More creative options matter, but not if nobody can say what should remain recognizably yours. None of those tests requires a large transformation program; each starts with one workflow and a comparison the team can understand.

Which recurring task in your business would you measure first for cost, exceptions and human review?

Verified sources

Continue your decision path

Move from understanding to action.

02 · Go deeper

3 Things AI: The “What Are We Actually Buying?” Edition

Three practical ways to test a supplier exit, teach an alert assistant with experienced staff and evaluate current web answers before a wider rollout.

Read next
03 · Assess

Apply this signal to your architecture.

Identify the workflow, context, and controls to structure first.

Open Architecture Assessment