ERP Stall and Gaming at Kahatagaha Group

Reframe first. Every symptom described is a business-transformation symptom, not a system defect: late POs, duplicate records, paper invoices, parallel spreadsheets and 34% turnover are signals of ownership, incentives and capacity, not of software configuration.

Diagnosis using a change lens

• Vision and case (Kotter steps 1-3): the programme has been run as an IT implementation with an IT steering committee. There is no compelling unit-level answer to 'what changes for me', so unit heads have rationally protected themselves by keeping 'the real numbers'. Parallel spreadsheets are the visible sign of a withdrawn commitment, not resistance to be overcome by communication alone.
• Strategy and systems (7S): the standardised processes do not fit the business reality of the packaging unit - 21% of POs after goods receipt means the process design ignored how urgent orders are actually placed. Specification failure, not user error.
• Shared values and incentives: the shared-services KPI 'invoices processed per FTE' rewards throughput and is silent on accuracy, completeness and cycle time. That is why month-end close lengthened while 'productivity' rose: volume was processed, exceptions were parked, and rework moved downstream. Classic KPI gaming through the measure chosen, not through dishonesty.
• Staff and skills: 34% turnover in a new operating model means the centre is permanently in training. Three of nine staff short is a capacity decision taken at the top and then blamed on the team.
• Data ownership: 4,300 duplicate supplier records exist because no one owns master data. Without a data owner, governance, and a change-request process, duplicates regenerate faster than they are cleaned.

Root cause statement for the board The programme was scoped as a technology delivery with a fixed budget and a timeline, governed by IT, with no accountable process owners in the business units, no data governance, no backfill during go-live, and an incentive system that rewarded the wrong behaviour. Those are the four levers; the ERP module is not one of them.

Design principles

1. Every volume or efficiency measure must be paired with a quality or risk counter-measure; a single unpaired measure will be optimised at the expense of whatever it ignores.
2. Lead measures (process discipline) sit alongside lag measures (cost, close days).
3. Fewer measures, each with one named owner and a written data definition - the 'parallel spreadsheet' problem is partly a definitions problem.
4. Measures are set for the shared-services centre as a whole, not per team, so teams cannot pass exceptions downstream.

Proposed scorecard

Perspective	Measure	Counter-measure / pairing
Financial	Fully loaded cost per invoice processed	Cost-to-serve per business unit; not the bonus driver on its own
Internal process	Touchless invoice rate	Exception and rework rate; % invoices requiring manual intervention
Internal process	Cycle time, receipt to payment; month-end close days	% invoices paid within agreed terms (protects suppliers and discounts)
Compliance	% purchase orders raised before goods receipt	Reported jointly with procurement, not shared services alone
Data quality	Duplicate supplier records per 1,000; master data change lead time	Requires a designated data owner in each business unit
Customer	Business-unit satisfaction and query resolution time	Replaces internal complaint escalation as the informal channel
Learning and growth	Trained-and-certified staff retention; vacancy rate	Turns the 34% turnover from an excuse into a managed metric

Ownership rules

• One owner per measure, published with the definition, source system and calculation method.
• No measure enters a bonus plan unless its counter-measure is also in the plan with at least equal weight.
• 'No PO, no pay' is a procurement-owned control, not a shared-services throughput target.
• Review the scorecard quarterly for gaming: check whether the measure and the quality measure are moving in the same direction. Divergence is the alarm.

Decision framing. The choice is not one decision for the whole programme but a gated sequence per unit. 'Accelerate everything' repeats the packaging failure three times; 'stop' writes off working capability and leaves units on unsupported legacy systems.

Readiness gates (assessed per unit, 8 weeks before go-live)

1. Process fit: the top 20 transaction types are mapped and tested against the unit's actual practice, with the unit's process owner signing off - this is the control that would have caught the 21% late-PO problem.
2. Data quality: supplier and customer master deduplicated below an agreed threshold (for example 0.5% duplicates), with a named data owner in the unit.
3. Capacity: backfill in place and vacancy below 10% in the receiving team, or the go-live date moves. Turnover at 34% is a barrier to entry, not a training problem.
4. Incentives: the unit's scorecard already carries the new measure pairs, so behaviour is aligned before cut-over, not after.
5. Exit criteria from the previous wave: month-end close at 7 days or better and touchless invoice rate above an agreed level. If wave two has not met its exit criteria, wave three does not start.

Recommendation

• Pause the packaging unit and remediate it directly; it has already failed and is the reference case for what goes wrong.
• Re-scope with a reduced first-release scope per unit - finance and procurement only, HR deferred - so each release is genuinely deliverable in the window.
• Continue, gated, with logistics and agri-inputs, which are sequential and more standardisable.
• Defer tea exports to a low-season window; its peak trading period is the wrong time to change systems.

Report to the board: re-baseline the LKR 5.8 billion, publish the gate criteria and the current status against each, and make the next release decision on evidence rather than on sunk-cost pressure.