The Recommendation Record

IntroductionThe posting, and the method

FINRA is hiring Principal Investigators; one of four teams is Recommendations, which investigates suitability, Regulation Best Interest and recommendation-related misconduct.

The posting says the team will “analyze accounts, trading activity, fees, and disclosures to quantify harm” and assess supervisory controls and conflicts of interest. Among the duties it lists are building factual records from documents, correspondence and testimony, performing data analysis and clearly stating “findings, exceptions, and assumptions,” and documenting results for Enforcement attorneys.

This chapter does what an outsider can do with those duties. It reads one public report, maps each entry to the rule that governs it, works one quantitative test on a labelled synthetic account, and lists what a case file would need. It follows the same five steps as chapter 12: questions raised, analysis, recommendations, actions and follow-through.

Section 1 · December 2025 reportQuestions raised

Three entries from one month, each leaving open what a case file would have to answer.

FINRA’s report “Disciplinary and Other FINRA Actions,” reported for December 2025, lists actions taken in the prior period. Three entries concern recommendations or the controls behind them:

An entry is a summary, and the report does not say how many customers were affected or what they lost. Reading it as an investigator raises five questions the report does not answer:

1. Which standard applied to the representative’s recommendations? Rule 2111’s own page says the rule does not apply to recommendations subject to Reg BI. The entry cites neither, so the file would have to establish whether each customer was a retail customer and when each recommendation was made.
2. How many customers, and what harm? Seven exchanges are stated. Losses, surrender charges and benefits given up are not.
3. When did the first firm’s deficiency start and stop? The entry states no period, unlike the second firm’s.
4. What did the undertaking require, and how is it tracked?
5. Was there a link between the firm-level findings and any customer outcome? A procedures finding is not a finding that a customer was harmed, and the entries do not claim one.

Section 2 · The rulesAnalysis

Each entry sits under a different obligation, and the quantitative tests carry no numeric thresholds.

How Regulation Best Interest is built

Rule 15l-1 has a general obligation in paragraph (a)(1): when recommending a securities transaction or strategy to a retail customer, act in the customer’s best interest at the time, without putting the firm’s or the person’s interest first. Paragraph (a)(2) says the general obligation is satisfied by meeting four component obligations: Disclosure (a)(2)(i); Care (a)(2)(ii), with three parts (understand the recommendation, a reasonable basis that it suits this customer, and, for a series of transactions, a reasonable basis that the series is not excessive); Conflict of Interest (a)(2)(iii); and Compliance (a)(2)(iv), written policies reasonably designed to achieve compliance.

The quantitative test has factors, not thresholds

Rule 2111 Supplementary Material .05(c) says no single test defines excessive activity and names three factors that may support a finding: turnover rate, cost-equity ratio and in-and-out trading. FINRA’s page for the rule gives no numbers for any of them, and Regulatory Notice 12-25 presents the same factors as ones that “may provide a basis,” not as rules. Any cut-off an investigator quotes therefore comes from somewhere else, such as case law or past practice, and must be cited there, not to the rule.

A worked test on a synthetic account

The numbers below are invented for this exercise. They show how definitions move the answer, which is the part an investigator has to state as an assumption. The formulation used here, annualised purchases divided by account equity for turnover and annualised total costs divided by account equity for the cost-equity ratio, is a common one in practice; the rule page does not define either, and the chapter does not source a definition from FINRA.

Take an account with $750,000 of purchases and $18,000 in commissions, markups and fees over 10 months, an annualisation factor of 12/10 = 1.2, and two readings of equity: the average over the period ($150,000) or the starting balance ($120,000).

The same activity reads 6.0 or 7.5, and 14.4% or 18.0%, depending on one choice. Counting sales as well as purchases would double the first figure to 12.0, which is why the definition has to be written into the file, not assumed. The cost-equity ratio is also the return the account must earn each year before the customer is better off than holding cash; both figures describe the activity, and the rule asks whether the series was excessive for this customer, which turns on the customer’s profile as well.

Section 3 · What the file needsRecommendations

What an investigator would ask for, by obligation. This is a plan, not something done.

• Care, customer-specific: the account-opening record and investment profile at each recommendation date, the recommendation record, the customer’s own account of what they were told, and the product’s terms (for an annuity exchange, surrender charges, benefits given up, new fees and any new surrender period).
• Care, series of transactions: the full trade blotter and monthly statements, so turnover and cost-equity can be computed under a stated definition and tested against more than one.
• Conflict of interest: compensation schedules, commissions on the product recommended against the product given up, and any contests or incentives in the period.
• Compliance and supervision: the written supervisory procedures in force on each date, the exception reports the firm ran, and evidence of who reviewed what and when.
• Disclosure: the relationship summary and any account-level disclosures delivered, with dates.

Section 4 · Building the fileActions

Six steps, in order, before any conclusion is written.

1. Fix the period and the standard: which recommendations fall on which side of 30 June 2020, and which customers are retail customers.
2. Request the documents above and log what arrives, from whom and when.
3. Compute the activity measures under at least two written definitions and report the spread, as in the table above.
4. Compare each recommendation with the customer’s stated profile and record every exception, including those that cut the other way.
5. Quantify the customer-level effect only from documents: fees, charges and benefits that are on paper.
6. Write the factual record and a separate list of assumptions, so Enforcement can see which findings rest on documents and which on judgment.

Section 5 · MeasuresFollow-through

How the work could be shown to have held.

• Reproducibility: every computed figure comes from a script that a second reader can run on the same inputs, as ch13-verify.py does for this page.
• Sensitivity: a finding that changes under a reasonable alternative definition is reported with the alternative, not alone.
• Evidence sufficiency: each assertion in the record points to a document, and each unsupported inference is labelled as one.
• Review: a second investigator re-derives the headline numbers before the record goes to attorneys.

LimitsWhat this does not show

• Any investigative, enforcement or testimony experience. The author has none to claim; this is a reading exercise.
• The underlying cases. Only the report’s summaries were read. The orders, letters of acceptance, or the firms’ and representative’s own accounts were not, and the entries’ omissions may be filled by them.
• That any customer was harmed or how much. The three entries state sanctions and findings about procedures and recommendations, not customer losses.
• The December 2025 report in full. Entries outside the three, including two borderline items about restricted investments and unauthorized transactions, were not analysed.
• Which rule governed the representative’s recommendations. The chapter marks its mapping as an inference.
• The accounts in the worked example. They are synthetic.
• The rules after the date read. Rule text was read on 30 October 2026 as published then.

UpdatesWhat changed after the first version