A real half-hourly data analysis: how the right size moved IRR by 4 points

Last month we worked through a 1.4 GWh annual consumption food production site where the original installer quote was for 700 kWp solar on a rule-of-thumb 60% sizing. The site looked like a standard daytime-heavy industrial profile until we pulled the half-hourly data — and it told a meaningfully different story.

The annual consumption number

1,380 MWh/year was the annual headline. At a 60% sizing rule of thumb, the installer's 700 kWp proposal would deliver roughly 665 MWh/year of generation — comfortable below annual demand and (on the surface) appropriately sized for a daytime-heavy operation.

The investor case at 700 kWp: £560k turnkey, year-one electricity saving £128k, payback 4.4 years simple, IRR 12.8%. Not bad. Not exceptional.

The half-hourly file

The supplier provided 12 months of half-hourly data on standard request — a CSV with 17,520 rows. Average demand 158 kW. Maximum 412 kW (production-line peak summer). Minimum 38 kW (3am Sunday morning, refrigeration only). Standard deviation 64 kW.

The notable feature: overnight load consistently 70–110 kW from cold-storage refrigeration. The site never went below 38 kW. That's not unusual for food production — but it materially changes the optimal solar size because solar generation is concentrated in 6 hours around midday, while overnight consumption can't use solar at all.

Re-modelling with half-hourly data

We modelled self-consumption percentage at solar sizes from 400 kWp to 1,300 kWp in 50 kWp increments. The results:

The optimum landed at 1,100 kWp — 36% larger than the installer's proposal — with a project IRR of 16.9% versus the installer-proposed 12.8%.

Why the optimum was bigger than expected

The continuous overnight refrigeration load (70–110 kW) created enough demand that even at 1,100 kWp solar, summer-midday generation didn't overwhelm the site's ability to consume. Self-consumption stayed at a respectable 71% at the larger size. The marginal kWp from 700 to 1,100 captured most of its value through avoided cost (~17p/kWh) rather than export (~7p/kWh), justifying the additional capex.

The installer's rule-of-thumb sizing was anchored on annual consumption (1,380 MWh) rather than the demand profile shape. The rule of thumb works on offices and retail. It systematically under-sizes industrial sites with continuous loads.

What changed in the recommendation

We recommended 1,100 kWp at £880k capex versus the original 700 kWp at £560k. The customer accepted the larger sizing on the back of:

• £320k additional capex generating an additional ~£40k year-one cash benefit (avoided cost + export revenue minus marginal O&M).
• Marginal IRR on the additional 400 kWp: 12.5% — clearly above the 8% hurdle rate the customer set.
• FYA scaling — additional £40k corp-tax saving year-one on the larger capex.
• 25-year cumulative cash improvement: ~£1.2m vs the smaller system.

The general lesson

Half-hourly data costs nothing to obtain (the supplier provides it on request), takes a couple of days to model properly, and on industrial sites with continuous loads typically changes the optimal sizing by ±20–40% from the rule-of-thumb. That delta translates into 3–6 percentage points of project IRR, which is large enough to matter materially.

Two patterns to look for: (a) sites where overnight load is above 30% of summer-midday peak — these are typically under-sized by rule-of-thumb; (b) sites with strong summer/winter seasonality — these are typically over-sized in summer-light models. Half-hourly data is the only reliable way to identify which case applies.

For any commercial solar project above 100 kWp, we now insist on half-hourly data as part of discovery. The cost of getting sizing wrong is too large to skip the analysis.