Knowledge Base

How to Keep Accountability Clear Across Shifts, With AI in the Loop

DBR77 Team7 min read

On this page

Accountability stays clear across shifts when every open issue has one named owner, ownership moves only when the next person accepts it, and escalation runs by role and clock time instead of by who happens to be on shift. AI can suggest owners and priorities, but a person approves each one and stays accountable for it. If a suggestion has not been reviewed by the end of the shift, it is handed over like any other open item.

The shift change is where most plants lose track. Work started at 21:30 is half done at 22:00, the person who knows the details goes home, and the next crew finds a note that says "check press 4".

Why issues get lost between shifts

The UK Health and Safety Executive states that many accidents have occurred because of failure of communication at shift handover, most of them involving planned maintenance work. It also lists failure of information transmission at shift handover among the factors behind the Piper Alpha disaster. Its advice: handover should be face-to-face, two-way with both people taking joint responsibility, verbal and written, and given as much time as it needs.

In most plants the problem is less dramatic but daily. Four patterns repeat:

  • Team owners. A task "belongs to maintenance", which means it belongs to nobody at 02:00.
  • Silent handover. The outgoing lead writes a note, the incoming lead reads it later, and nobody confirms anything.
  • Relative deadlines. "Next shift" moves every eight hours.
  • Wiped boards. The whiteboard is cleaned at the start of the shift, and the history goes with it.

Using Machine Data in Shift Handover shows how machine data can feed the handover without more reporting.

At 05:40 the AI suggests moving the press 4 mold change to the morning shift.

Comparison of an unreviewed AI suggestion at shift handover, without a handover rule and with a clear rule

Four rules for ownership across shifts

  1. One person, not a team. Every open task has one named owner. The team can help, but one person answers for it.
  2. Ownership moves only on acceptance. The outgoing owner stays the owner until the incoming person accepts the task. A task nobody accepts at handover goes to the incoming shift lead by default.
  3. Deadlines in clock time. "By 06:00 Tuesday", never "next shift".
  4. The record travels with the task. What was done, what was found, photos, readings and the next step stay attached, so the new owner does not start from zero.

These rules sound basic. Most plants apply them to safety permits and almost nothing else.

Escalation that does not depend on the roster

Escalation should name roles and times, not people. On a day shift, a late maintenance task may go to the maintenance supervisor after 30 minutes. On a night shift with no supervisor on site, the same rule must point somewhere real: the on-call engineer, or the shift lead with a call-out list. Write one ladder per issue type and check it against every shift pattern, including weekends.

A good ladder has three steps at most: owner, shift lead, then the manager or on-call role. More steps slow the response and blur who decides.

Exceptions that break the rules

Most handover failures come from the cases nobody planned for. Agree these in advance.

SituationWho owns itRule
Task open at shift endOutgoing owner until acceptedHanded over in person, accepted in the system
Next shift lacks the skillShift lead of the next shiftReassign or schedule for a shift with the skill, with a clock-time deadline
Owner absent or sickShift leadReassigned before the shift starts, not when someone notices
Safety-related issueShift lead, never a teamNo carry-over without a containment step recorded
Same issue on three shiftsArea managerBecomes a root-cause task, not a fourth quick fix
AI suggestion not reviewedIncoming shift leadHanded over as an open item, or closed as rejected with a reason

Where AI fits, and where it does not

AI is useful at the shift boundary. It can read shift logs and open tasks, write a short summary, rank what needs attention first and suggest an owner based on skills and the roster. The judgment stays with people: whoever approves a suggestion owns the result, as if they had made the decision alone.

The risk is that people stop checking. The EU AI Act names this for high-risk systems: the people overseeing them should stay aware of the tendency of automatically relying or over-relying on the output (automation bias), and be able to disregard, override or reverse it. Most shop-floor tools are not high-risk systems under the Act, but the principle is sound on any shift. NIST's AI Risk Management Framework states that trustworthy AI depends upon accountability, and that accountability presupposes transparency.

Illustrative example: at 05:40 the AI suggests moving a mold change on press 4 to the morning shift because the night setter is busy with a die repair. The night shift lead does not review it before handover. Under a clear rule, it appears on the handover list as "suggestion pending", the morning lead accepts it at 06:05 and becomes its owner. Without the rule, the suggestion sits in a queue, both leads assume the other saw it, and the mold change starts 90 minutes late. When AI should only advise and when people must decide is covered in When AI Should Recommend and When People Should Decide.

How this works in DBR77 IRIS

Shift handover in DBR77 IRIS is recorded in IRIS MES. The outgoing shift records open issues and the incoming shift confirms acknowledgment, with a full audit trail. Each task in IRIS Tasks has an owner and moves through open, in progress, review, done and verified, so it is always clear who holds it. Skill data from the People module decides whether an operator is qualified for a station, and unresolved issues on the IRIS Gemba boards escalate through the tiers by SLA rules.

The IRIS assistant can answer questions about the shift from the same data, such as what stopped and which orders are late. Data & AI recommendations wait for a person to apply, approve or reject them, so it stays clear who decided. From Signal to Task to Closure: The Execution Loop follows the full path from signal to closed task.

FAQ

Can a task be owned by a team?

A team can do the work, but one person should own it. Shared ownership is the most common reason tasks are lost at night.

What if the incoming shift will not accept a task?

It goes to the incoming shift lead by default, who either accepts it or reassigns it with a clock-time deadline. It never stays without an owner.

Who is accountable when someone follows an AI suggestion?

The person who approved it. The AI's reasoning and the approval are both logged, so the decision can be reviewed later.

How long should handover take?

As long as the open items need. HSE advises giving handover as much time as necessary and identifying higher-risk handovers in advance.

Conclusion

Take your current handover sheet and give every open item one named owner and a clock-time deadline, with ownership that moves only when the incoming person accepts it. Write an escalation ladder of three steps at most for each issue type and check it against every shift pattern, including weekends. Hand over unreviewed AI suggestions like any other open item, so each one has a person who approves it and owns the result.

Sources