Leading Analytics and AI Change

Slide 1 of 4

A standard protects readers

A work standard is a check another reviewer can repeat before a report is shared.

Jelmer has seen repeat errors in draft reports.

Slide 2 of 4

What can go wrong

One table has 1,805 absences, but they belong to only 421 employees.

An absence is an event, not a person.

Slide 3 of 4

Keep people protected

A work sample can contain personal data, so use reference codes and only necessary fields.

Escalate a suspected exposure rather than circulating the sample.

Slide 4 of 4

Quick check

A useful standard tells a reviewer what to check.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Group the defects

Repeated mistakes are easier to fix when grouped by their failure type.

This log is a labelled practice scenario, not a recorded employee performance history.

Slide 3 of 4

Missing is not late

An unconfirmed delivery has no verified arrival outcome and must not be called late.

Separate incomplete evidence from recorded results.

Slide 4 of 4

Quick check

Sort mistakes by the check needed to catch them.

Slide 1 of 5

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 5

Commit to a choice

Which kind of standard catches more recurring defects?

Choose before seeing the model.

Slide 3 of 5

Build a rule card

Each rule needs a reason, a passing example, a failing example and a short check.

Reviewer check: compare Count rows with Count different employee_ref.

Slide 4 of 5

Six short checks

The practice checklist gives a reviewer clear pass or fail actions.

These proposed checks are not an adopted company policy.

Slide 5 of 5

Quick check

A rule is useful when it would catch a real mistake.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Replay the mistake

A strong rule must catch a defect when tried on yesterday’s bad work.

Each sample is an authored exercise, not an actual team rework log.

Slide 3 of 4

Know the limit

Some repeat mistakes need a tool fix, not another reminder.

Log the risk and ask the owner to correct the default.

Slide 4 of 4

Quick check

Replay tells you whether a rule actually works.

Slide 1 of 4

Not every report is equal

Review effort should follow consequence, not just the order in which reports arrive.

In this practice week, 14 reports arrive and five can receive full review.

Slide 2 of 4

Look at the exposure

A report covering six rocket-builder customers deserves care when its financial claims are material.

Consequences depend on audience, decision and data sensitivity.

Slide 3 of 4

Make review levels clear

A full review tests numbers, sources, rules and claims; a light check tests obvious risks.

The practice plan does not assume every report is safe.

Slide 4 of 4

Quick check

Choose the review depth by risk.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Commit before deciding

Does equal checking always improve quality?

Choose before the sampling rule is shown.

Slide 3 of 4

Add a random check

After the riskiest work, check one random lower-risk report in full.

The random pick helps find a problem that the risk ranking missed.

Slide 4 of 4

Quick check

Document both priority and selection method.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Severity follows consequence

A material defect changes a decision or exposes people, unlike a small formatting mistake.

A material error means rework or escalation, not cosmetic polish.

Slide 3 of 4

Choose a release decision

Release only when checks support the important claims.

A safe partial release may be possible when the affected claim is removed and that decision is documented.

Slide 4 of 4

Quick check

The defect matters when it changes what readers decide.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Compare two reviewers

Agreement means reviewers make the same decision on the same sample.

This is a simulated comparison, not recorded performance data.

Slide 3 of 4

Revisit the difference

When reviewers disagree, recalibrate: find the rule each person read differently.

Agree on a worked example and a future recheck.

Slide 4 of 4

Quick check

Good calibration compares decisions, not personalities.

Slide 1 of 4

Feedback should repair work

Useful feedback helps a colleague change the right part of an analysis.

This is a labelled practice review, not an actual staff evaluation.

Slide 2 of 4

Commit to a prediction

Which note is more likely to lead to a corrected resubmission?

Choose before the example outcome.

Slide 3 of 4

Reveal the difference

The original draft calls 1,805 absence events 1,805 people, but there are 421 people.

The practice correction changes the count, not the chart decoration.

Slide 4 of 4

Quick check

Good feedback names an actionable repair.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Use three clear parts

A feedback card names the defect, its consequence and one next step.

Keep the focus on the work rather than the colleague.

Slide 3 of 4

Evidence makes it fair

Show the source evidence that another reviewer can reproduce.

Give them the check, not a replacement report.

Slide 4 of 4

Quick check

The next step should be checkable by its owner.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Choose what matters

When a draft has several issues, start with the ones that change decisions.

The order here means material defects are corrected before formatting.

Slide 3 of 4

Weight by line counts

A simple average of category rates may differ from the rate across all eligible lines.

For all lines together, divide 8,195 on-time lines by 17,758 confirmed on-time or late lines.

Slide 4 of 4

Quick check

Correct the material issue before cosmetic details.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

An assistant is not proof

A draft made with an AI assistant still needs its claims checked against source records.

Ask how the claim was produced rather than arguing about the tool.

Slide 3 of 4

Hold the evidence line

When someone disagrees, restate the defect with the data and agree a repeatable test.

This is evidence of separate outcome categories, not proof of why any delivery was delayed.

Slide 4 of 4

Quick check

A useful reply ends with a shared verification step.

Slide 1 of 4

Ratings are not proof

A self-rating tells you what someone believes about their own skill.

A work sample shows what they actually did. You will predict before you look.

Slide 2 of 4

Commit before the reveal

Noor wants to know who needs development this year. Commit to an answer first.

Two practice analysts rated their skill at reconciling sources.

Slide 3 of 4

Review the actual work

A capability gap is a skill someone cannot yet show reliably in their work.

The analysts are practice. Every finding is a real figure.

Slide 4 of 4

What to look at first

Before you plan any course, ask what failed in the work and why.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Four different causes

A shortfall has one of four causes: capability, process, tool or workload.

Only a capability gap is a development need. Other causes go to their owner.

Slide 3 of 4

Two competing accounts

A source assertion is one unit’s account of a cause. Nobody has settled it yet.

So never say titanium was proven late.

Slide 4 of 4

Short report, which cause?

Classify the failure point before assigning development.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Prioritise the consequence

Consequence means what a shortfall could change for the reader.

One wrong conclusion can matter more than a frequent small error.

Slide 3 of 4

Route the real cause

An owner is the person or function able to repair the actual cause.

Slide 4 of 4

Which need first?

Order needs by their decision risk, then assign owners.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Keep individuals private

A finding about one person goes to their line manager only.

F-01 has two analysts, one at 0.6 FTE. So “the part-time analyst” names a person.

Slide 3 of 4

Needs can change

In a changed case the work changes, so check the need again.

Counting duplicates by hand closes. Checking the tool’s output opens.

Slide 4 of 4

Safe to share?

A good team summary keeps evidence useful without identifying an individual.

Slide 1 of 4

Choose before seeing

Do courses or real deliverables create better evidence of development?

Make your prediction before seeing the practice results.

Slide 2 of 4

The practice result

A development assignment is work already due, used to build and check a skill.

Slide 3 of 4

Build three assignments

A useful work assignment has a real output, named support, a reviewer and a later check.

Each output is due anyway.

Slide 4 of 4

Which is real work?

Choose the work product that can be reviewed later.

Slide 1 of 5

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 5

Predict the coaching result

Coaching keeps the thinking and the work with the colleague.

Slide 3 of 5

Ask before supplying

The customer counts disagree, but that does not show lost customers.

Practice colleague: “The assistant found 22 lost customers.”

Slide 4 of 5

Three questions, in order

A coaching plan starts from what the colleague can do, then asks three questions in order.

Slide 5 of 5

What can they do already?

Coaching starts from what the colleague can already do. Their attempt shows it.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Hold the difficult moment

A worked step shows one move, then returns the problem to the colleague.

Resist taking over even when work is urgent.

Slide 3 of 4

Equal totals can differ

A matching total does not prove that the same customers are in each group.

Slide 4 of 4

What shows it worked?

Test whether the next output is independent before crediting a development activity.

Slide 1 of 4

Warm-up

Two quick questions from earlier.

Answer both to continue.

Slide 2 of 4

Return an incomplete plan

A development plan must specify real work, support, a reviewer and later evidence.

Amend or return each line before approving.

Slide 3 of 4

Decide under a deadline

A capability risk is the harm if a needed skill is not strengthened in time.

F-01 has two analysts in post, so spare review time is limited.

Slide 4 of 4

Answer Noor

State what stays, what moves and the risk accepted when Noor asks to pause development.