Half of SoFi's AI Coach chats are about investing. That is a distribution problem.

SoFi's head of advice and planning, Brian Walsh, told PYMNTS that roughly half of the interactions with the company's AI-powered Coach platform concern investing ^[1][2]. That is the most consequential number in the interview, because the investing question is the front door of almost every advisory relationship, and in this case it is being answered inside an app the customer already banks with ^[3].

The original digital wealth pitch was reach: software could serve consumers who lacked enough assets to attract a traditional adviser, delivering basic education and automated portfolio management at a fraction of the cost ^[4]. That framing put robo-advisers in a price fight. What SoFi describes is a different product shape. Coach, per PYMNTS, combines conversational guidance, account-level data, visual explanations and persistent memory into something closer to an ongoing advisory relationship ^[3].

The competitive substance is in the plumbing, not the answers. Walsh's own example is that telling a customer to invest spare cash may be right for someone with a funded emergency reserve and no expensive debt, and harmful for someone with a near-term expense or a high-interest card balance ^[5]. Consumers, he said, often hold accounts at five, 10 or even 15 institutions, which makes assembling that picture the actual work ^[6]. Add memory of family circumstances and prior conversations, and the user stops re-explaining their life every session ^[7]. Add charts, because a line item saying 30% of discretionary spending goes to one category lands differently as a pie slice towering over the rest ^[8].

None of that is a claim to replace planners, and PYMNTS does not make one; the argument is that software is absorbing the context, continuity and communication techniques that made human advice hard to scale ^[9]. SoFi says Coach was built from principles, methods and boundaries drawn from eight years and tens of thousands of human planner conversations, with rule sets designed to escalate an interaction to a human when it exceeds the intended scope of automated guidance ^[10][11]. Walsh frames the human-in-the-loop design as what makes AI scalable in a regulated environment, and as a two-sided productivity gain: software serves people who would otherwise get nothing, while planners work from data the digital interaction already gathered ^[12][13].

Read the ledger from the other side of the table. If half of Coach's traffic is investing, then everything else combined -- budgeting, debt, insurance, planning -- is the other half ^[14]. Investing is not a follow-on module here; it is the thing customers show up asking about, at the moment they have cash to place. A firm whose economics depend on winning clients at low asset levels depends entirely on being the first place that question gets asked. It cannot win an inquiry it never receives, and the incumbent with the deposit account, the transaction history and the memory of the last conversation receives it first.

Worth watching: whether SoFi ever discloses what share of Coach interactions escalate to a human, and what share convert into funded accounts, since those two rates decide whether this is a service feature or an acquisition channel ^[11]. Also worth watching whether the investing share holds as Coach's scope widens, and how the rule sets that define the boundary of automated guidance hold up when the guidance is wrong for a customer whose 15th account was the one nobody linked ^[6][11].