ETRM Software Problems: The 7 That Leave Your Most Convex Positions Sitting in a Spreadsheet

Why do structured deals keep ending up as placeholders?

Because vendor deal templates are built around instruments that repeat across many clients, which is the correct product decision and the reason your vanilla book works fine. Fixed for floating swaps, physical forwards at a hub and options with standard exercise are the same shape at every firm. Contracts that were negotiated rather than traded are not.

Tolling agreements with availability tests and outage provisions. Storage with injection and withdrawal ratchets, cycling limits and park and loan features. Transport capacity with fuel retention that varies by path. Load following supply with a shaping obligation and a bandwidth penalty. Heat rate calls with a strike expressed against two commodities. Each of those is genuinely a different valuation problem, and forcing it into a user defined field means the deal sits inside the system without participating in it.

The scope failure is deciding to fix this by replacing the system. Trade capture for vanilla instruments, confirmations and basic settlement are commodity capability, and rebuilding them buys nothing. What is worth owning is the layer where your firm is actually different: deal economics modelled as first class structures with their own valuation functions, feeding the same position, exposure and profit and loss machinery as everything else. The objective is not to replicate the vendor's forwards handling. It is to stop having a second book.

What goes wrong with the curves and history you migrate?

The migration exposes how much of your valuation lives outside any system. Three problems show up every time.

The first is that historical marks cannot be reproduced. Curves were built with a mixture of broker quotes, settlement history and judgement, in scripts on a laptop, with inputs that were not stored. So the mark on a given day exists as a number with no derivation behind it. You cannot recreate what was not captured, so the honest approach is to import historical marks as stated values with their provenance recorded as unreproducible, and start full reproducibility at cutover. Pretending otherwise creates a false audit trail, which is worse than an acknowledged gap.

The second is deal amendment history. Negotiated contracts get amended, and the amendments frequently live in a contract folder rather than in the trading system, so the deal record you migrate reflects current terms with no history of when they changed. Any attribution across a period that includes an amendment will then be wrong and nobody will be able to say why.

The third is unit and convention drift. Heat rates, basis conventions, time zones and day counts differ between the sources you are consolidating, and a silent mismatch produces valuations that look plausible and are wrong by a consistent factor. Validate a sample of migrated deals by revaluing them against a known date and comparing to the number the business already published. If the two do not tie, stop and find out why before loading anything else.

Why do the market data, market operator and pipeline integrations break after launch?

They break because each one is a separate contract, a separate format and a separate revision schedule, and teams scope them as one line item called data feeds.

Market operator settlement data revises. A value published today is not necessarily the value that will stand in three months, and a system that stores the first published figure without a revision mechanism will drift permanently away from the settlement statement. Every ingest needs to carry a publication version and support restatement, with the effect on prior attribution visible rather than silently absorbed.

Pipeline and scheduling interfaces break on nomination cycles and on their own calendars. Cycle deadlines, imbalance rules and confirmation formats differ by pipeline, and a design that assumes one nomination model will need reworking the second time you add a counterparty.

Market data subscriptions break on entitlement rather than on technology. Redistribution rights, per user entitlement and the difference between what you may store and what you may only display are contractual constraints that shape the architecture. Read the agreements before designing the cache, because discovering an entitlement restriction after go live means rebuilding how the data is served, and it is a conversation nobody enjoys having with a vendor's compliance team.

What happens when controls around curves and amendments are not covered?

This is the difference between a valuation system and a system anyone will trust. The engineering is straightforward. The discipline is what is usually missing.

• Four eyes approval on any change to a production curve. A single person who can change a curve and revalue the book is a control weakness regardless of their integrity, and it is the first thing an auditor or an acquirer asks about.
• Versioned methodology. How you blend broker quotes with settlement history, derive basis at an illiquid node, shape a monthly block into hourly and extend beyond the liquid horizon must be stored as a versioned method with its inputs, so any historical mark can be reproduced exactly.
• Separation of duties between capturing a deal and publishing a curve, enforced by the system rather than by an organisation chart.
• Immutable amendment history on every deal, so a term change is a dated record with an actor rather than an edited field.
• Independent model validation where the output reaches board level risk reporting or supports hedge accounting, budgeted separately rather than treated as part of delivery.

Concentration risk deserves a specific mention. When curve construction lives in one senior analyst's scripts, the exposure is not only control. It is that the methodology walks out of the building with them, and it is often the reason the firm makes money.

Should you build custom or configure what you already own?

Buy, and we say this before quoting anything, if your book is mostly vanilla forwards and financial swaps. Molecule has genuinely changed the mid market: it deploys quickly, it is cloud native, and for a straightforward power and gas book it is a strong purchase that many firms should make instead of anything custom. Its opinionated model is both its strength and its limit.

ION Openlink Endur remains the enterprise default and earns that on breadth. It will handle almost anything, with the honest caveats being cost and pace: implementations routinely run over a year, extension happens through the vendor's own scripting and specialist consultants, and small changes take longer than the business expects. Allegro has a similar profile with real depth in gas and power. Amphora and Enuit are credible mid market options with smaller ecosystems, which matters mainly when you need people who already know the system. Aspect has its deepest heritage in oil and refined products, so it is a less natural fit for a power and gas portfolio.

The usual right answer is not either or. Keep the packaged system for capture, confirmations and settlement, and build the layer where you are different: curve construction with reproducibility, valuation for your negotiated deals, credit exposure under your specific agreements, and attribution shaped to how your risk committee thinks. Full custom becomes defensible only when your entire portfolio is structured, when your asset class is one no vendor models, or when a packaged implementation has already failed twice for genuinely non standard requirements rather than badly specified ones.

How do hidden costs get into the quote?

In Digital Heroes delivery experience a structured deal valuation and risk layer alongside an existing system, covering curve construction, credit exposure and attribution, runs $150,000 to $350,000 and reaches production in 16 to 24 weeks. A full custom trading and risk platform runs $500,000 to $1,200,000 across 12 to 24 months. The drivers particular to this category are specific.

• Structured deal type count. Each is its own valuation model plus a test suite, and the tests are the larger half of the work.
• Market coverage. Every market operator and every pipeline brings its own data feeds, settlement formats and scheduling interfaces.
• Whether scheduling and actualisation are in scope or only valuation, which is close to a doubling decision.
• Regulatory reporting obligations, which add a workflow with its own deadlines and its own failure consequences.
• Model validation, which should be an independent exercise you budget for rather than an afterthought, and which produces documentation that saves real time during audits and diligence.

What separates a trading and risk build that works from one that fails here?

The first difference is whether attribution was a design requirement or a reporting request. Decomposing the daily move into curve movement on existing positions, new deals, actualisation as scheduled volumes become metered, settlement adjustments and an unexplained residual determines how positions and valuations must be stored. Built in, it is cheap. Retrofitted, it is close to a rewrite. And the residual is the number that decides whether your risk committee trusts the book or spends every meeting arguing about data instead of position.

The second is whether credit exposure recalculates on every curve publication rather than monthly. The failure mode in credit is timing, not arithmetic. A counterparty moves through a threshold during a price spike and nobody calls for collateral for three weeks, or a trader books a deal that pushes past a limit that existed only in a document. Enforcing limits at the point of capture closes the second half of that gap.

The third is how you test the developer before committing. Ask them to explain how they would value one of your structured deals before discussing anything else. You are testing whether they can hold the economics in their head, because someone who has never modelled optionality will build a system that stores your deal and cannot price it. Then ask how a historical mark gets reproduced, and listen for versioned methodology, stored inputs and an immutable record of published curves.

A useful next step costs two hours and no money: list every deal in the portfolio whose valuation currently lives outside the system, and total the notional and the optionality in that list. If the total surprises anyone in the room, you already have your answer. Settle repository and infrastructure ownership before kickoff either way. At Digital Heroes the client owns the code from the first commit.