Forecast Accuracy Audit · October 2026

Penrose Garden Products Ltd

Wholesale garden tools, planters, watering and outdoor lighting for independent garden centres and hardware shops across the South West. 46 staff. 60 products over 12 months of forecasts, sales and stock.

Worked example. Penrose Garden Products is a fictional business; every name and figure is invented. All figures come from the worked-example spreadsheet: ALL CHECKS PASSED (with 2 notes): products where the forecast did worse than the baseline, and months with no sales while stock was held.
The headline

How good is the forecast, and what do the errors cost?

Forecast error
30%
against 44% for simply using last month's sales
Bias
+2%
overall; hides big misses both ways
Cost of error
£174,312
a year: gross profit lost to stock-outs and the cost of holding excess
Excess stock
£49,264
cash tied up above the stock you aim to hold, once
Where

Value and demand groups

X: steady demandY: variableZ: erratic
A
AX2 products · 4% of valueerror 16% (baseline 18%)excess £3,061 · lost £0 a year
AY18 products · 73% of valueerror 28% (baseline 46%)excess £7,444 · lost £147,557 a year
AZ1 product · 3% of valueerror 68% (baseline 71%)excess £23,298 · lost £0 a year
B
BX5 products · 3% of valueerror 20% (baseline 21%)excess £66 · lost £0 a year
BY14 products · 11% of valueerror 31% (baseline 41%)excess £5,727 · lost £11,445 a year
BZ3 products · 2% of valueerror 88% (baseline 106%)excess £3,195 · lost £5,457 a year
C
CX11 products · 3% of valueerror 18% (baseline 19%)excess £2,691 · lost £0 a year
CY6 products · 2% of valueerror 23% (baseline 31%)excess £3,782 · lost £0 a year
CZnone

A items make up 80% of value; their forecast error is 30%. Red marks a group where the errors cost more than £20,000. Effort pays most on the A and B items; C items need a simple rule, not more forecasting time.

What it costs

The products costing most

ProductErrorBiasMonths out of stockExcess stockGross profit lost a yearCost of error a year
Outdoor lighting 01
PG131 · AY
30%-18%7£0£84,075£84,075
Hand tools 02
PG102 · AY
54%+41%6£0£14,570£14,570
Outdoor lighting 08
PG138 · AY
24%-17%8£0£13,565£13,565
Watering 02
PG122 · AY
36%+0%1£0£12,658£12,658
Outdoor lighting 10
PG140 · AY
24%-14%3£0£9,551£9,551
Outdoor lighting 03
PG133 · AY
129%+43%6£0£7,598£7,598
Outdoor lighting 06
PG136 · AY
22%-22%6£0£5,540£5,540
Hand tools 01
PG101 · AZ
68%+47%0£23,298£0£4,660

Months that ended with no stock are left out of the error, because sales then show only what was available; they are costed as lost gross profit instead. 18 products did worse than simply using last month's sales; the largest by value: Planters 04, Gloves and kneelers 09, Outdoor lighting 03, Bird care 04, Planters 09, Watering 05.

How it's done today

The planning routine

Who prepares the forecast?The purchasing manager, alone, from the sales history in the accounts system
How often is it updated?Once a year in August for the whole year; changed only when a shortage is noticed
What method is used?Last year's sales for the same month, plus 10% for growth, for every product
What information goes in?Sales history only; no input from the sales team on promotions, new accounts or lost accounts
Who decides the final numbers, and how is accuracy checked?The purchasing manager decides; accuracy is never measured
What to change

The improvement plan

#ChangeLikely effectEffortOwnerBy when
1Forecast monthly, not once a year: a 30-minute review in the first week of each month
Routine · All
HighLowPurchasing managerNovember 2026
2For outdoor lighting, plan the season from pre-orders and last season's sell-through, not last year's sales alone
Method · Outdoor lighting
HighMediumPurchasing managerJanuary 2027
3Sales team adds known promotions, new accounts and lost accounts before each monthly review
Who decides · All
MediumLowSales managerNovember 2026
4For C items, stop adding 10% growth; order to a simple two-month average and a lower stock cover
Method · C items
MediumLowPurchasing managerDecember 2026
5Record lost orders when stock runs out, so true demand is visible
Data · All
MediumMediumOffice managerDecember 2026

Ranked by likely effect on accuracy, then effort. Rerun the audit in six months to see whether the error and the cost fell.

How confident are we?

What this rests on

Built from 60 products over 12 months; sales at average prices are 99% of sales in the accounts. Error and bias compare months 2 onwards, value-weighted, against a simple baseline of last month's sales, leaving out months that ended with no stock, because sales then show only what was available. Lost sales take the larger of that month's and the previous month's sales as demand. Excess stock uses your aim of 1.5 months of demand; holding cost 20% a year; and lost sales assume 30% of a month's demand is lost when you run out. These three are judgements.

The months after

What the agent flagged, and what was decided

At handover, the purchasing manager started the monthly forecast review (change 1 on the plan). The agent was then set up in Penrose's own AI assistant account, with read-only access to the stock system and the purchasing spreadsheet. On the third working day of each month it adds last month's forecast, sales and month-end stock to a new dated copy of the audit, drops the oldest month, runs the checks and sends the purchasing manager a short note, copied to the managing director. It uses the rules on Inputs: a 30% line for the error on A items, a 3-point rise in the overall error, and the five products costing most. Nothing left Penrose, and no decision changed, without one of its own people approving it. These events continue the fictional example.

Monthly runs
4
November to February 2027
Flags raised
3
the other notes were the monthly score only
Error on A items
27.9%
in February, back within the 30% line after rising to 32.4% in December
Data problems caught
1
no figures reported until Checks passed
DateWhat the agent foundFlag?What people decided
4 NovOctober added. Overall error 29.7%, A items 29.5%, both much as at the audit. The note listed the five products costing most, led by Outdoor lighting 01.Note onlyNothing needed. The purchasing manager used the note at the first monthly review.
3 DecTwo A items in outdoor lighting, Outdoor lighting 01 and 08, ran out at the end of November, at the start of their busiest months.FlagThe purchasing manager brought forward January's order for both by four weeks, and the sales manager asked the five largest garden centre customers for their December lighting orders.
3 DecThe error on A items rose to 32.4%, above the 30% line. The biggest driver was Outdoor lighting 03, forecast 40% below November's sales: the line the audit had already found worse than simply using last month's sales.FlagThe managing director brought forward change 2 on the plan, planning outdoor lighting from pre-orders, to the spring order instead of January.
6 JanThree new planter lines sold in December were not on Products, so Checks showed CHECK and the note gave no figures that morning.FlagThe office manager added the three lines from the price list. Checks passed again the same morning.
6 JanDecember added. Overall error 28.9%. A items 31.1%: still above the line, so listed in the note but not flagged a second time.Note onlyNothing new needed. Pre-orders for the spring lighting range were being collected.
3 FebA items back within the line at 27.9%. Products worse than last month's sales down from 18 to 15.Note onlyThe managing director kept the monthly review and will look at the 30% line again at the six-month rerun.

At the check-in on 27 November, about six weeks after handover and after the first monthly run, Geoff and the managing director read the log together. They kept the 30% line and the 3-point rise.

What the agent added. The method is unchanged: same spreadsheet, same checks, same decisions. The agent made sure the score was kept every month, and the two stock-outs and the rise in the A-item error reached the purchasing manager within three working days of month end, not at the six-month rerun. The figures above are fictional, and no £ figure is claimed for the agent.

Fictional and labelled: these events continue the worked example. The same six events are the sample rows in the Agent log of the worked-example spreadsheet. The agent is part of the guided and done-for-you routes; the self-implement kit does not include it.