Data migration is where trust lives or dies
The fastest way to kill confidence in a trading platform is not a bad UI.
It’s a trader asking, “Why doesn’t my position match?” and someone replying, “Just use the spreadsheet for now.”
That single moment changes the tone of the whole program. People stop believing the system. They stop believing the plan. Then the “temporary workaround” becomes the operating model.
If you want trust after go-live, treat migration like operational risk. Not a technical task you “get through” at the end.
Stop treating migration like a weekend event
Most programs talk about migration like it’s a cutover checkbox: “Load the data, run a few spot checks, move on.”
Trading firms don’t work like that. A platform is a chain: trade capture → risk → pricing → credit → settlement → finance. If one link drifts, the whole chain gets noisy. And noise becomes manual work. Manual work becomes delay. Delay becomes board pain.
Migration is the point where your new system either earns the right to be trusted, or gets treated like a demo environment with nicer screens.
Run migration like a risk workstream
Here’s the shift: treat “data” like cargo. Treat “reconciliation” like the port gate. Nothing leaves the port without a manifest match.
1) Define the “truth set” like you mean it
Don’t migrate “everything.” Migrate what matters to run the book.
Pick the minimum set that proves control, not completeness:
- Open trades and lifecycle state (amendments, novations, cancellations)
- Positions and exposures (by commodity, location, delivery period)
- Curves and pricing inputs used for MTM
- Counterparty limits, credit usage, collateral state
- Settlement status and payment runs
- For ERP overlap: inventory balances, GR/IR, open AP items, GL postings tied to trades
If you can’t say what “must match” on Day 1, you can’t manage the risk.
2) Build a migration manifest with owners
A manifest isn’t a spreadsheet of tables. It’s a decision record.
For each “truth set” item, you need:
- Source of record (old E/ETRM, spreadsheet, data warehouse, ERP)
- Target object (new platform object and key fields)
- Mapping rules (especially where the business logic changes)
- Owner (named person, not a team)
- Reconciliation method (count, value, tolerance, sampling rules)
- Acceptance criteria (what “good” looks like)
This is where programs quietly fail: “Everyone thought someone else owned it.”
3) Reconcile early, often, and visibly
If you only reconcile at the end, you don’t have a migration plan. You have a surprise generator.
Run recon in waves:
- Wave 0: a thin slice (one desk, one commodity, one region)
- Wave 1: last month + current month trades
- Wave 2: broaden to full scope
- Wave 3: dress rehearsal with freeze rules and cutover timings
Each wave must produce a recon pack that an exec can read in 3 minutes.
4) Use tolerances and error budgets
Perfect alignment is not the goal. Controlled alignment is.
Define tolerances that reflect trading reality:
- Counts must match exactly for certain objects (e.g., active trade count by status)
- Values can be within a tolerance (e.g., MTM within a defined $ threshold, by portfolio)
- Timing drift can be acceptable if documented (e.g., settlement status updates lagging by X hours)
Then track an error budget: how many breaks are acceptable at each stage, and how quickly they must burn down.
5) Treat breaks like incidents, not “data issues”
A break is not a discussion. It’s a ticket with a clock.
For every break, record:
- Impact: can it change risk, P&L, credit, settlement, or reporting
- Root cause: mapping, missing data, bad upstream, logic mismatch, timing
- Fix type: configuration, code, data correction, process change
- Owner and due date
- Decision needed: fix now, accept risk, defer with controls
This is how you stop migration from becoming a late-night blame loop between vendor and internal teams.
A real-world example: the Migration Trust Dashboard
This is what a weekly exec-ready view looks like. Short. Evidence-based. No theatre.
Migration Trust Dashboard (example)
| Area | Metric | Current | Trend | Acceptance |
|---|---|---|---|---|
| Trade population | Active trades migrated | 12,410 / 12,410 | ↔ | Must match |
| Lifecycle state | Trades with correct status | 12,372 / 12,410 | ↑ | 99.5%+ |
| Positions | Position breaks (by book) | 9 breaks | ↓ | <10 breaks |
| MTM | MTM variance vs legacy | +$38k on $412m | ↓ | within ±$50k |
| Credit | Credit usage variance | 0.4% | ↔ | within ±1% |
| Settlement | Unmatched invoices | 6 | ↓ | <10 |
| Break backlog | Breaks > 7 days old | 4 | ↓ | 0 by go-live-2w |
Top 3 break themes (this week):
- Delivery period mapping for shaped power products (mapping rule still inconsistent across legacy sources)
- FX rate source differences between ETRM and ERP (timing + rate type)
- Counterparty ID duplicates (master data clean-up incomplete)
That’s the kind of page boards trust. Because it reads like a control room, not a status call.
What leaders should ask in steering, every week
If you’re an exec sponsor, these questions cut through noise fast:
- What is the truth set for go-live, and what is explicitly out of scope?
- Show me the reconciliation results, not the plan to reconcile.
- What are the top 5 breaks by risk and effort, and who owns each one?
- How many breaks are older than 7 days, and why?
- What tolerance are we using for MTM and position variance, and who approved it?
- What is the fallback if one truth set fails at cutover?
- Are we running a dress rehearsal with freeze rules, approvals, and rollback triggers?
If your program can’t answer those cleanly, it’s not ready to claim confidence.
If you want a fast start this week
If you’re heading toward a trading/risk/ERP go-live and you want a clear view of migration risk, start with the Project Health Check. It flags the blockers, the missing owners, and the next steps that stop “spreadsheet fallbacks” before they happen.

