How Much Does ETRM Software Cost to Build in 2026?

What ETRM work actually costs

Vendor pricing in this category is opaque by design and usually arrives bundled with implementation services. Priced as a build, the work splits cleanly along one line: whether you are extending a packaged system of record or replacing it.

• Structured deal and risk layer: $150,000 to $350,000, 16 to 24 weeks. Your own curve construction, the deal types the vendor template cannot hold, credit exposure by counterparty under your netting and collateral agreements, and profit and loss attribution the risk committee can actually read. Sits alongside the packaged system rather than replacing it.
• Full custom platform: $500,000 to $1,200,000, 12 to 24 months. Deal capture, valuation, scheduling, actualisation, settlement and regulatory reporting in one system. The right call for a narrow set of firms and the wrong call for most.
• Each additional market or commodity: $40,000 to $120,000. A second ISO brings new market results, new scheduling mechanics and new settlement data. Adding physical gas to a power book brings transport, storage and imbalance, which is a different set of models entirely.

Trade volume is almost irrelevant to the price. A firm executing thousands of vanilla financial swaps a month is cheaper to build for than one executing forty tolling and heat rate option deals a year, because the second firm's optionality lives in structures no template holds.

What pushes the number up

• Deal structure variety. The dominant driver. Tolling agreements, heat rate options, storage, physical transport, load following and environmental attribute deals each carry their own valuation, their own risk decomposition and their own settlement behaviour.
• Curve construction. Proprietary forward curve building, basis and shaping methodology is usually the reason the firm is building at all, and it is quantitative work rather than application development.
• Number of markets and pipelines. Each ISO and each pipeline has its own data, its own scheduling process and its own settlement timeline.
• Credit and collateral complexity. Netting under master agreements, collateral thresholds, independent amounts and rating triggers all have to be modelled per counterparty, and the exposure number has to be defensible to your credit committee.
• Daily mark cycle expectations. A position that has to be marked, risked and reported before the desk opens is an operational commitment with real engineering behind it.
• Regulatory and audit reporting. Position reporting obligations, model validation evidence and the audit trail behind any published mark all add scope that trading teams tend to omit from the first specification.

What brings it down

• Keeping the packaged system as the book of record. The single largest saving in this category. Let the vendor handle vanilla capture, confirmations and settlement, and build only the valuation and risk your book actually needs.
• One commodity, one market, in phase one. Prove the model on the market carrying most of your risk, then extend.
• Deferring scheduling and actualisation. These are operationally heavy and frequently already handled adequately elsewhere in the business.
• Using market data you already subscribe to. Curve inputs are usually already licensed somewhere in the firm. Reusing that entitlement rather than adding a new one avoids a recurring cost larger than most people expect.

A worked example: generator and marketer, three ISOs

Generation owner and marketer, assets in three ISOs, a physical gas book supporting the fleet, tolling and heat rate option deals that the incumbent system holds as free text with a spreadsheet valuation beside them. Structured deal and risk layer, by line.

• Discovery with trading, risk and back office, cataloguing every deal structure in the book: $18,000
• Deal capture and valuation models for tolling, heat rate options and storage: $58,000
• Curve construction and daily mark to market engine: $64,000
• Credit exposure by counterparty under netting and collateral terms: $44,000
• Profit and loss attribution readable by the risk committee: $36,000
• Market data and ISO result feeds across three markets: $32,000
• Integration back to the packaged system of record: $27,000
• Acceptance with a parallel mark cycle over one month: $19,000

That totals $298,000, near the top of the structured layer band because of three markets and three non standard deal types. A single market power only firm with one structured product typically lands near $175,000. This firm has stayed on the packaged system for capture and settlement and has no plan to replace it, which is the outcome we recommend for most firms in this position.

Where the money goes across phases

• Discovery and deal cataloguing, roughly 6 percent. Writing down every structure in the book, including the ones traded once and never documented. This phase reliably finds deals nobody on the technology side knew existed.
• Valuation models, roughly 20 percent. Quantitative work, and the part that needs a trader in the room every week.
• Curve engine, roughly 21 percent. The firm's own methodology, which is usually the reason for building.
• Credit and attribution, roughly 27 percent. The outputs the committee and the auditor consume.
• Feeds, integration and acceptance, roughly 26 percent. Including a full parallel mark cycle, which is the only acceptable proof.

The annual run cost

• Support and change, 18 to 25 percent of build cost per year. Higher than most categories because deal structures evolve continuously and every new structure is a model change.
• Market data subscriptions. Priced by your data vendors rather than by any developer, and usually the largest recurring line in the whole stack. Confirm entitlement for any new use before designing around a feed.
• Tariff and market rule changes, $20,000 to $75,000 a year. ISOs amend tariffs and change charge structures, and anything feeding valuation or settlement has to follow.
• Model validation, annually. Independent review of valuation models is standard practice at firms of any size and it consumes both external cost and internal quantitative time.
• Hosting and compute, $12,000 to $45,000 a year. Overnight mark runs and scenario analysis are bursty, compute heavy workloads with a hard morning deadline.
• Audit and committee support. Explaining a published mark to an auditor requires reproducible runs and retained inputs, which is a retention and process cost as much as a technical one.
• Analyst and trader training, $6,000 to $18,000 a year. A valuation tool nobody trusts gets shadowed in a spreadsheet, and then you are paying for both.

Timeline and what governs it

A structured deal and risk layer runs 16 to 24 weeks. The pace is set by trader availability rather than by engineering capacity, because every valuation model needs a desk conversation to validate. Book those sessions in advance and treat them as the project's critical path, since a model built without the trader who structures those deals will be rebuilt.

Run at least one full month of parallel marking against the existing process before anyone relies on the new numbers. Differences will appear, and most of them will turn out to be the old spreadsheet being wrong. That is the point of the exercise, but it takes a month and a calm room.

When you should not build

If your book is mostly vanilla forwards and financial swaps, buy a packaged system and stop there. The valuation is standard, the risk decomposition is standard, and there is nothing about your positions that justifies bespoke quantitative work. Firms in that position who build usually do so because they dislike a vendor rather than because their deals demand it, and they end up owning a maintenance obligation for a capability they could have licensed.

Build when the deals carrying most of your optionality are the ones your current system holds as free text with a spreadsheet valuation beside them. That spreadsheet is your real risk system, it has one author, and it is the argument for a build far more than any feature comparison will be.

What to insist on in the quote

Price each deal structure as its own line. Tolling, heat rate options and storage carry very different modelling effort, and a single line item labelled structured products hides which model is actually being estimated and which one will overrun. Ask for the parallel mark cycle to be quoted separately as well, since it is the phase that decides whether the desk trusts the output enough to stop maintaining its spreadsheet.

Ask also who owns the model documentation. Independent model validation will request methodology and assumptions in writing, and a build that produces correct numbers without that documentation simply creates a second piece of work at review time, usually at a worse moment.

How to size your own budget

• Count deal structures, not deals. List every distinct structure traded in the last two years. That list, more than volume or headcount, is what prices this project.
• Identify which positions are currently valued in a spreadsheet. Those are your phase one scope and your business case in the same breath.
• Confirm market data entitlement before design. Data licensing can cost more annually than the software you are about to build, and it is set by your vendors, not by us.
• Reserve 15 percent for model iteration. Valuation models are refined against real marks, not specified once. Budgeting for that loop is what keeps the project honest.

When you are ready to turn this into a specification, Digital Heroes writes a product requirements document before any code exists, so the scope is fixed and priced rather than discovered later at a day rate. You can take that specification to any other firm on your shortlist.