InMarketing Today

Bank of America reported third-quarter net income of $6.9bn today, down 17% from $8.3bn a year earlier, after a $1.2bn hit from a fine for money laundering and sanctions violations. The United States Treasury announced the fine yesterday, saying the bank failed to report more than $5bn of suspicious transactions, and a new consent order requires it to work with a third party to find and report any it missed. Shares fell more than 8% in morning trading, and results fell short of analysts' expectations even without the fine.

The bank says remediation costs are already in its expense base, and its chief financial officer told reporters that thousands of staff are on the work and no substantial further hiring is planned. That describes effort, and effort is a weak defence when the new order follows one imposed in 2024. The open question is how much the bank must spend to prove to regulators that its compliance programme has meaningfully improved, and a headcount figure does not answer it.

If your firm has a regulatory finding open, or could face one, report progress in outcomes: suspicious activity identified, backlog cleared, deadlines met. Agree those measures with your compliance lead now, so that when a client or journalist asks how remediation is going, your people have something verifiable to say.

Starling is working with Nvidia to build a transactional foundation model that will be trained and hosted on a sovereign cloud run by the British provider Civo, with all data kept within United Kingdom jurisdiction, Finextra reports today. The model will learn from proprietary data covering more than 15 billion anonymised transactions, go into production in early 2027, and start with fraud detection before it augments the Starling Assistant. Finextra describes Starling as the first bank to train an artificial intelligence model this way.

Starling has framed its artificial intelligence work as a decision about where customer data lives, which is something a customer, a regulator or a board can understand and check. Its chief information officer, Harriet Rees, co-authored the United Kingdom government's financial services artificial intelligence adoption plan, which flagged reliance on a small number of global artificial intelligence and cloud providers, so the bank's choice matches its public position. The caveat is timing: nothing is in production until early 2027, so today's claim is a commitment, not a result.

If you are announcing an artificial intelligence programme, lead with the decision that carries the risk, such as where data sits, who can see it and which suppliers you depend on, before you describe the technology. State it in one sentence your clients' risk teams could verify, and put a date on when you will report results.

Matthew Stibbe of Articulate Marketing argues that a 'delivered' figure means only that a receiving server accepted the email, not that anyone saw it. Analysing 10.4 million sends, he finds five leaks between the send button and the inbox. In the portals studied, 24% of marketable contacts had never been emailed, hard-bounced contacts reached as much as 20% of a database, and a little over half of marketable contacts (two-thirds in some portals) had received ten or more emails since they last opened or clicked one. Spam filtering, he adds, takes about 12% of email, according to a Validity benchmark.

Deliverability rarely announces itself. It shows up as a slow fall in click rates and an anecdote about the chief executive's email landing in spam, and the cause is usually cumulative: years of sending to people who never asked to hear from you. A report that says 99% delivered can flatter a programme that, on Stibbe's own estimate for his portfolio, may be getting only one to two in ten emails looked at. For any firm that reaches clients and prospects by email, that is a reach problem no campaign report will show.

Ask whoever runs your email programme for three numbers this week: the share of marketable contacts never emailed, the share hard-bounced, and the share who have ignored ten or more sends. Then follow Stibbe's three steps: check your domain's authentication records, watch your sender reputation, and validate every list before it is loaded. If you do one thing first, suppress sending to contacts who have stopped engaging, which HubSpot calls graymail suppression, because a smaller list that reads your email is worth more than a larger one that does not.