What the Assistant Wanted

In 2016, Landesbank Bremen faced a constellation that only regulated houses know in this sharpness. The MiFID II deadline (3 January 2018) was running, and the existing systems for voice, video, meeting-room booking and trading voice recording were so far from what MiFID II would require from 2018 that incremental extension was not an option. A platform was needed that recorded MiFID II compliantly, satisfied Basel II, met a bank's IT-security requirements — and at the same time held the trading floor and the entire house at 99.99% high availability across two sites.

The bank put the project out to tender. My team and I responded. We won. I led the project from the pre-sales phase through selection to acceptance as project manager on the integrator side — twelve months, January through December 2016.

Bank tenders of this magnitude are typically decided on a list of 200 to 400 requirements. Function checked, function checked, function checked. Whoever can check them all and isn't priced out wins. That is the mechanic every vendor in the market masters, and which therefore no longer carries differentiation.

Once the bid was won, the project's logic flipped. What looked like a platform migration — voice, video, meeting-room booking, ten trading desks with full redundancy, dedicated rooms in an external data-centre across two sites with hard switch-over requirements — was in truth three migrations in one: a compliance migration, a codetermination migration, and a trading migration. Each with its own non-negotiable order of priority.

A compliance migration. Every single requirement — MiFID II, Basel II, IT security — had to be described in detail: what we solve how. Every change had to be documented, because the supervisory authority later audits exactly this documentation. It wasn't enough to satisfy the function — we had to be able to explain in every audit why precisely this solution path and not another. This continuous burden of justification is the actual cost-and-time factor in a banking platform migration. Whoever doesn't have it in the project rhythm from the start never catches up.

A codetermination migration. Voice recording of 1,200 employees is not a purely technical matter; it is a hard codetermination issue. The works council was completely opposed to recording at the start — interestingly, not against recording itself, since its regulatory necessity was beyond doubt. What disturbed the works council was the manual effort for downstream classification of the recordings by employees: which securities had been discussed in which call? Answering that question only afterwards is error-prone — and the employee burden it generates was unacceptable to the works council.

An AI that could automatically classify recordings did not exist at this maturity in 2016.

We solved the problem at a different point: with a metadata system at the trading desk that allowed the trader to enter the securities IDs (WKN) during the call. Data is generated where it is already known — not reconstructed afterwards from memory. With that, the manual post-processing went away, the regulatory requirement was met, and the employee no longer had to spend hours at the end of the week classifying recordings he had known about while making the call.

That was the moment the works council could carry the solution.

The second pillar of the codetermination architecture was access to recording data. Except when the supervisory authority requested data, every access was strictly under two-person principle. Each individual access had to be presented to the works council with justification. This discipline is built into the system, not appended — which means it cannot be bypassed under time pressure, even with good intent.

A trading migration. Switching ten trading desks under 99.99% high availability across two sites while ongoing trading cannot be paused is the operationally most delicate of the three migrations. Switch-over tests, failover behaviour, recording continuity, timestamp accuracy — all under the pressure that a gap in the recording is more expensive in an audit than the migration itself.

A central key person on the bank side was the chief security officer. In a constellation where compliance, codetermination and trading continuity can pull against each other, the CSO's voice is often the one that decides which solution is forensically sound in case of doubt. Bringing that voice in early, honestly and in detail kept the project from looping multiple times.

1. System follows function. When codetermination and compliance run through a platform project, the migration is not primarily a technology migration. It is a migration of the process, with technology following along — and it is led not by the technology, but by what the regulatory and codetermination function requires. Whoever reverses the order in a banking migration and starts with technology ends up with a technically clean solution that doesn't hold in audit and doesn't pass the works council. The order is not negotiable.
2. Bids in regulated industries are qualified on the formal list and decided on the persona. As long as every vendor masters the requirement list, the difference is not won by an additional function but by a concretely tangible demo from the daily life of those who actually use the platform. In the executive office it was the assistant. In any other bank it will be a different persona — the trader, the auditor, the helpdesk lead. Whoever doesn't know that persona doesn't win the bid. Whoever knows her and can show her demo authentically wins it even against formally stronger competitors.
3. Codetermination conflicts are hints toward a solution, not obstacles. The works council at Landesbank Bremen was not against recording — they were against the manual classification effort afterwards. That wasn't a negotiating position; it was a precise description of the actual problem. Capturing the securities IDs during the call rather than reconstructing them afterwards wasn't a concession to the works council — it was the right solution, which without the works council might never have been found. Whoever reads codetermination voices as obstacles to overcome forfeits this form of help.