Request for Proposal RFP

Why MW and MWh do not specify a battery plant

Nameplate is the top of the energy chain, not the bottom. A DC nameplate shrinks through the state-of-charge window, then through PCS, transformer and collection losses and the auxiliary draw, so deliverable AC energy at the point of interconnection commonly lands around 85-92% of DC nameplate at beginning of life.

Contract energy sits lower again, and it is only a contractual quantity once three coordinates are fixed: a meter, a condition set, and a year. An RFP that names 400 MWh and stops has specified none of them, which leaves each bidder free to answer a different question and price it honestly.

The year is the coordinate with the largest price attached, because it decides who buys the degradation. Holding a 400 MWh obligation to year 20 on pure day-one overbuild needs roughly 400 ÷ 0.70 ≈ 570 MWh of usable-equivalent capacity at commercial operation on a ~70% year-20 retention curve, while a staged plan starts nearer 450 MWh and adds tranches later.

Those are two different plants and two very different prices for the same headline. If the RFP does not say which one it wants, the bid that assumed the cheaper structure wins the evaluation and the owner discovers the missing capacity somewhere around year eight.

The technical content that makes bids comparable

Start with the duty profile, because every guaranteed number downstream is conditional on it. State the cycles or annual throughput, the depth of discharge, the C-rate, the resting state of charge between dispatches, and the cell-temperature window the site will actually produce — the duty you intend to run, not the continuous lab protocol a vendor's cycle-life data was measured at. If the plant will run a blended stack of services, say so, because a jagged blended duty is not the sum of two clean profiles and the warranty envelope will be drafted from whatever the RFP declared.

Then the guaranteed numbers, each carrying a boundary and a year. Guaranteed energy stated at the bus the offtake pays at, in the worst contracted year, not nameplate at the racks. Round-trip efficiency on a named basis, because the same hardware reads roughly 92-96% DC, 88-93% at the PCS AC terminals and 85-90% net AC at the POI, and a bid two points better than its rival has often just drawn the boundary two buses earlier.

Availability stated as a definition rather than a percentage: time-based and capacity-weighted metrics are different measurements of one plant, and excluded hours can be struck from the denominator or deemed available in the numerator, which gives different answers on identical downtime. Utility-scale guarantees commonly land at 95-98% measured annually, with mature LFP sites targeting 97% plus — worth converting to hours before you accept one, since 98% allows about 175 h/yr of unexcused downtime, 97% about 263 h and 95% about 438 h.

Then the scope that holds those numbers up over time. Say whether augmentation is inside the price or the owner's cost, against which warranted curve tranches are timed, and what physical provisions the plant must be built with — reserved foundations, feeder positions, conduit, DC bus and conversion headroom — because reserved bare land is a provision on paper only.

Say how many years of service are included and which contract each performance promise lives in, since the delivery-stage numbers usually sit in the supply agreement and the long-tail ones in the LTSA. Ask for the spares list with the physical answer attached: what is stocked on site, what sits in a regional depot, and what is only an untested lead time.

Standards, the AHJ, and where the scope stops

Name the code stack and the evidence, not just the standard numbers. UL 9540 is the safety certification of the assembled ESS product; UL 9540A is the fire-propagation test method that produces data at cell, module, unit and installation level. Require the full report at the level that supports the spacing actually proposed, for the cell part number and revision that will ship — a summary letter proves nothing, and a report on one cell does not carry over to its successor in the same container.

State the NFPA 855 edition the authority having jurisdiction enforces, name who authors and stamps the hazard mitigation analysis, and ask which explosion-control strategy the drawings carry. If the project will be insured against carrier engineering criteria that go beyond code, put that in the RFP too rather than discovering it at the lender's insurance review.

Draw the scope boundary at a physical termination and issue the responsibility matrix with the RFP. The high-voltage interface is where this bites hardest: whether the switchyard is utility-built, developer-built and transferred, or developer-built and retained changes the cost line, the schedule risk and the operating obligation without changing one item of steel.

Bidders pricing the same rows return comparable numbers; bidders left to assume return a spread that is mostly assumption. The rows that go unassigned are predictable enough to pre-empt — transfer-trip and the telecom behind it, the control-centre point list, metering instrument transformers at the named accuracy class, utility witness-test dates, and station service energised before backfeed.

Employer's requirements: the same content under FIDIC

On FIDIC-based contracts the technical package has a name and a job in the contract. The Yellow Book, for plant and design-build, and the Silver Book, for EPC and turnkey, both make the contractor responsible for the design and both define the employer's requirements as the contract document stating the purpose, scope and design or other technical criteria for the works.

Under those forms the design obligation is commonly tied to fitness for the purposes defined in the contract, which makes the employer's requirements the yardstick the finished plant is measured against rather than an aspiration attached to a tender.

The practical consequence is that ambiguity is not neutral. A contractor designing to your words is entitled to satisfy them the cheapest compliant way, so a requirement reading "the system shall be designed for high availability" buys nothing, while the same requirement written as a percentage, a definition, a measurement boundary and a test becomes a guarantee with a remedy. Write the RFP knowing which annex it becomes.

What the term obliges is set by the specific contract, not by statute. FIDIC forms are routinely amended, US-style EPC and owner forms call the same package a technical specification, a scope book or an exhibit, and each allocates design responsibility differently. Check which document your technical content becomes, and what design obligation attaches to it, before assuming the standard form's allocation survived the negotiation.

What an underspecified RFP costs

The first cost is analytical. Coordinates the RFP never asked for cannot be recovered at evaluation, so the work moves downstream into bid normalization, where restating incomparable bids onto one basis is slower, more contestable and less accurate than specifying them in the first place. Ask for the guaranteed numbers in a fixed table with the boundary, the condition set and the year pre-printed, and require a written deviation list — a bid marked "compliant" with no exceptions stated is usually a bid nobody read against the spec.

The second cost is schedule. A technical requirement changed after award reopens whatever it touched: swapping the cell or the enclosure forces the hazard mitigation analysis to be re-analysed and resubmitted, and AHJ review cycles of weeks to a few months per round are normal in jurisdictions with little storage experience. Which is why the pre-application meeting with the fire authority and a settled enough interconnection position belong before the RFP goes out, not after the bids come back.