Keeping a practice ahead of regulatory change

The situation

A practice advising investors, funds and founders has to know when the rules change. Changes arrive from several regulators, as web pages, notices and long documents, at any hour.

Clients expect to hear what matters to them, and what it changes, before they ask. Reading everything in full is not possible for a small team.

Why it is hard

Most of what is published does not matter to this practice. A few documents matter a great deal, and they look like the rest.

The rule that matters can sit on page forty of a long document with a dull title.
A document that restates old rules looks like a change. It cannot be trusted until the notice that changed it is found.
Who a rule applies to decides whether it matters. A change for listed companies means little to a founder of a private one.
Anything sent to a client must be exact to the clause. A loose summary is worse than none.

The approach

We built a monitor that watches each source on its own schedule. It fetches politely, keeps a copy of everything it reads, and can trace every line back to it.

Each document’s text gets a fingerprint. A known page with new text becomes a new version, and old versions are never overwritten.
The type of document is read from its address and title, before it is opened.
Practice areas are tagged from keyword lists. A hit in the title counts three times a hit in the body.
Each document gets a score from four parts: practice area, the instruments clients hold, change language and a deadline.
The first two parts are multiplied by how far the rule reaches private companies. A rule for listed companies only counts for less.

The weights are 0.40 for practice area, 0.25 for instruments, 0.25 for change language and 0.10 for a deadline.

The score sends each document to a lane. At 0.78 or above, it gets a draft the same day.

At 0.30 or above, it goes into the daily digest. Below that, it is filed.

A restated text whose change is not yet established waits in its own lane until the source is found.

Drafts quote the source rather than paraphrase it. Every sentence cites its clause, and a list of banned words keeps out adjectives and advice.

The section on what to do about it is always left empty. That part belongs to a lawyer.

A lawyer approves, edits or rejects each draft, and a rejection needs a reason. Nothing goes to a client without that step.

Once enough decisions build up, the monitor suggests new thresholds from them. A person decides whether to adopt them.

A worked example

This is a synthetic month, scored the way the monitor scores it. Six of its documents show how the parts add up.

Synthetic example. Six documents from a synthetic month, scored the way the monitor scores them. Each part runs from 0 to 1.

Document	Practice	Instruments	Change	Deadline	Applies to	Score	Lane
New rule on reporting to fund investors	1.00	1.00	1.00	1.00	private companies (1.00)	1.00	Draft now
Change to employee share schemes	0.90	0.67	1.00	1.00	private companies (1.00)	0.88	Draft now
Disclosure change for listed companies	0.80	0.33	1.00	0.50	listed companies only (0.60)	0.54	Daily digest
Lending notice for intermediaries	0.50	0.33	0.67	1.00	intermediaries only (0.70)	0.46	Daily digest
Rules restated in one text	0.90	0.67	0.25	0.00	unclear (0.85)	0.51	Needs its source
Press note on a conference	0.20	0.00	0.00	0.00	unclear (0.85)	0.07	File only

The two changes for private companies reach the drafting lane. The listed-company change scores well on every part, and still lands in the digest.

That is by design. A document that reaches listed companies only can never score above 0.74, below the drafting line of 0.78.

Figure 1. A synthetic month of 60 published documents, by the lane each one lands in. Synthetic example.

Of 60 documents in the month, three need a draft the same day. Another two wait for the text that changed them.

The digest carries 27, three lines each. The other 28 are kept, and nobody has to read them.

Where it breaks

Keyword matching catches words inside other words, so some scores run high. The lawyer’s review is the backstop.
Most documents never say clearly who they apply to. The reach multiplier often cannot do its job, and treats them as unclear.
Scanned documents with no text layer cannot be read at all.
Some sources have no page it can read reliably, so they are still checked by hand.
It routes changes to kinds of client, not to named clients. Deciding which client to call is still a person’s job.

If your team reads every change in full, score them first and read only what reaches the top.