Before you automate, fix the work

Operations leaders can learn a lot about AI readiness from the work that repeatedly requires someone to save the day.

A manager clears the approval queue over the weekend. An experienced employee knows which spreadsheet is actually current. A customer gets onboarded because a director personally chases three teams. Everyone appreciates the effort, and the same rescue happens again the following month.

An occasional crisis calls for exceptional effort. Recurring rescue work deserves a closer look at how the work is designed. There may be a missing capability, unclear ownership, a dependency in the wrong order, or more work arriving than the next team can finish.

Adding AI at the busiest-looking step can make those problems harder to see. More drafts and requests move forward, while the people responsible for reviewing them inherit a larger queue.

Find the queue behind the hero

Start with one recent request that needed an intervention. Follow it from arrival to completion with the people who actually handled it.

Record when someone worked on it, when it waited, and why it moved again. Ask what the person doing the rescue knew or could do that the normal process didn’t provide. Was it decision authority, access to missing information, or simply extra hours?

Those are different problems. They call for different changes.

A useful workflow map includes the places work waits: an inbox, an approval list, a spreadsheet tab, or a message that has no obvious owner. Count arrivals and completions over the same period. Look at the oldest open items as well as the average. A queue that appears manageable can still contain requests that have been forgotten.

Follow the work after the automated step

Consider a hypothetical service team preparing proposals. An AI tool helps staff produce drafts much faster. Managers still need to review the scope, pricing, and delivery commitments before a proposal can go to the customer.

If managers are already at capacity, faster drafting creates more proposals awaiting approval. Customers may experience no improvement in turnaround. Staff may spend more time asking for updates, revising drafts whose assumptions have changed, or explaining why a nearly finished proposal is still stuck.

That doesn’t mean drafting assistance has no value. It means the team needs to examine the whole path before deciding how broadly to use it. Some proposals may be routine enough for a clearly defined approval rule. Others may require expertise or authority that the organization hasn’t assigned. Missing information might be creating most of the review effort.

Here is a simplified illustration, not a client result. Assume enough incoming work, no starting backlog, and the same working day for drafting and review:

Workflow	Drafts ready for review per day	Reviews completed per day	Approval queue grows by
Before drafting assistance	10	10	0 per day
Faster drafting, same review capacity	20	10	10 per day

After five days in the second scenario, 50 proposals await review. The team has produced more drafts, while customer-ready output remains ten per day. More review capacity, clearer approval rules, or better inputs could change that result. The pilot needs to test which change helps.

DORA’s value-stream mapping guidance makes the same distinction between improving a local activity and advancing the overall outcome. Its focus is software delivery; tracing work, waiting, and handoffs is also a useful diagnostic for this proposal example.

Each possibility suggests a different next move. A faster drafting tool can’t settle the ownership or capacity question on its own.

Give the recurring rescue an owner and a rule

Before buying another tool, write down the decisions that currently depend on intervention.

Who owns the request from beginning to end? Who can approve an exception? What information must be present before review starts? What happens when the usual reviewer is away? Which requests should the team decline or defer rather than start?

The answers should be usable by the next person doing the work. A process that depends on knowing whom to message is difficult to operate consistently, with or without AI.

Sometimes the first change is a clearer intake form or a visible queue in software you already have. Sometimes it is changing approval authority, limiting work in progress, or reserving review capacity. Sometimes the missing capability does require custom software. The point of the investigation is to distinguish those cases before committing to a solution.

Our sequencing questions address the broader decision about which move should come first. For this workflow, stay close to the specific reason the last request needed rescuing.

Test whether the whole workflow improves

Choose one request type and one accountable owner for a pilot. Establish a baseline before making the change, including time from arrival to completion, queue age, rework, and the interventions needed to finish.

Agree on the conditions for continuing or stopping. Faster drafting alone shouldn’t decide whether the pilot expands. If completed work increases but corrections or downstream waiting also rise, the team needs to understand that tradeoff. If staff maintain an unofficial spreadsheet to keep the pilot running, treat it as evidence of an unresolved requirement.

Use the review to answer a short set of questions:

• Does the request finish sooner, including the time spent waiting?
• Can the next person act without chasing information or permission?
• Has work disappeared, or moved onto another team?
• Can routine and exceptional cases be handled within the agreed rules?
• Can the workflow continue without the person who usually rescues it?

The anonymized assessment example shows how an assessment can separate immediate foundation work from decisions about buying and building software. The organizational-gaps article covers the wider strategy, fluency, and ownership issues around AI adoption.

Appreciate the person who helped in a crisis. Then use the recurring pattern to improve the system they work in. The useful result is a workflow the team can sustain, with AI applied where it helps that workflow succeed.