AI for Employee Onboarding in Professional Services: Clearing the Joiner Before the Clock Starts

This is about the fee earner. A solicitor moving practices, an audit senior joining an assurance team, a consultant arriving from a competitor. The distinguishing fact of this desk is that the person cannot be put on chargeable work, and in some roles cannot lawfully start at all, until the firm has run their history against its own book. Not the firm's history. Theirs. Everything else waits behind that. This is also not engagement acceptance, which clears a piece of work as you take it on. This clears a person.

The offer to start gap and what has to clear in it

Between signed offer and first chargeable hour sits a queue that almost nobody owns end to end. Conflicts against the joiner's prior matters. Independence declarations if the firm audits. A practising certificate verified with the body that issued it. Regulated references. Ethical walls built and switched on. Systems provisioned. Each has a different owner, system and cycle time, and they run in sequence because nobody sees the whole thing.

An orchestrating agent is worth building before anything clever. It holds the clearance queue as one record per joiner, knows which items block a start date and which merely block a matter, opens every workstream on offer acceptance, and shows risk and HR the same picture. The value is not speed on any single check. It is that the checks stop waiting on each other.

The prior client list arrives partial

The lateral hire owes their old firm confidentiality, so the list you get is incomplete by design, not by carelessness. Client names they are permitted to disclose. Matters described generically. Codes without counterparties. Some arrives as prose in an email, some as a spreadsheet typed from memory.

An agent does useful work on that mess. It parses whatever form the list arrives in, normalises names into candidate entities, marks each line as identified, ambiguous or unusable, and writes back one request for the gaps that matter, phrased so the joiner is not asked to breach their own obligations. It must be explicit about coverage: these resolved, these unresolved, these deliberately withheld. A clearance record that quietly treats a partial list as a complete one is worse than no record.

Running a joiner against the live client and matter database

The check is a search problem with a low tolerance for misses. The joiner's entities go against live clients, closed matters inside your retention window, current adverse parties, pipeline prospects and entities already walled for other reasons. Trading names, abbreviations, post merger renames and the same operating company appearing three ways across three systems all defeat exact matching, and loose fuzzy matching buries the reviewer in noise instead.

An agent can do the resolution and the assembly. For each possible hit it gathers the matter, the role the firm played, which side it was on, whether it is open, and what the joiner's own involvement looked like. The reviewer gets a short, ordered file of real candidates with evidence attached, not a raw match list.

When the conflict is with a parent or an affiliate

The expensive misses are rarely direct. The joiner acted for a subsidiary; your firm is adverse to its parent. The counterparty is a fund, and the conflict sits in a portfolio company two levels down. A group has been carved up since the matter closed.

This is where an agent earns its place, because walking a corporate tree across registry data, your own client group hierarchies and the engagement records is the kind of patient, wide traversal people skip when they are busy. It should show its route: this entity, this relationship, this source, dated. Whether the relationship is close enough to matter is a risk decision, and it stays one.

Independence, personal interests and family declarations

Audit and assurance firms carry a second problem. Independence attaches to the person, their household and their financial position, not just their work. The joiner declares holdings, loans, insurance, pension arrangements, directorships, and the employment of a spouse or close family member at an entity the firm audits.

Agents handle the mechanical half. Reconcile declared holdings against the restricted entity list, resolve tickers and fund names into issuers, walk funds through to underlying holdings where policy requires it, and flag a family member's employer against the audit client base. Where a disposal is required, the agent tracks it to completion with dated evidence. Materiality, and the interest technically outside the list but awkward in substance, goes to the independence partner.

Credentials verified with the body that issued them

A practising certificate, professional body membership, CPD standing, an audit registration, a licence for a reserved activity. A scan from the candidate evidences nothing. Each item has to be confirmed with whoever issued it, and the confirmation recorded with the source, date and method.

Agents suit this because it is persistence across sources that behave inconsistently, some searchable, some responding by inbox, some by post. The agent opens each verification, reads what returns, compares it against what the joiner declared, chases on schedule and escalates when your window passes. It must never mark an item verified from a document. It also carries the expiry forward, because a certificate that lapses later is the same failure.

References for a regulated role

Regulated references ask the prior firm specific questions about conduct, and the answers have to be obtained, chased and filed. The chasing is agent work. Noticing a question that was answered around rather than answered is a conversation between people.

Information barriers that have to take effect in the systems

A wall that exists as a partner's email is not a wall. It has to be configured in document management, practice management, time recording, the research archive, the deal rooms and whatever knowledge search the firm has just switched on.

An agent can take the cleared conflict record and produce the configuration: which matters, which repositories, which groups, which exclusions, in the terms each system uses. It can then verify the wall rather than assume it, by testing whether the joiner's account can in fact reach what it should not, and re-testing after matter migrations and system changes. Silent wall decay between audits is a common failure.

Provisioning that lands on a chargeable code

Fee earner provisioning is not laptops. It is practice management access, a time recording identity, research and case law subscriptions, data rooms and per seat tooling a partner bought for one engagement and never cancelled. Much of it is chargeable, some recoverable to a client, and the renewal arrives long after anyone remembers who asked.

An agent drives the requests from the joiner's role and practice group, opens each in the right system, and holds the entitlement as a dated record with an owner and a renewal date.

The unbilled first week, and joining a matter already running

Every hour before the joiner's first chargeable entry is cost, and the reason for a slow start is rarely the person. It is that nobody has assembled the context for the matters they are inheriting mid flight.

Agents do that assembly well. Given the matters a joiner is joining, the agent produces a briefing from the firm's own material: engagement terms and scope, the billing arrangement and client guidelines, who is on the other side, what has happened recently, what is outstanding and where the documents live. It also surfaces the local mechanics nobody explains, such as how narratives are expected to read and what the client refuses to pay for. The partner reviews it rather than writing it.

What the joiner must not bring

Onboarding carries an inbound risk the firm owns: templates, precedents, contact lists and working papers from a previous employer arriving on a personal account or drive. Agents help at the edges here. They can run the attestation properly and keep its wording specific to what this joiner actually worked on. Detection of exfiltrated material sits with security tooling and a serious conversation, and should stay there.

The graduate wave and the single lateral

These look like one process and are not. A trainee intake is a scheduled batch with uniform requirements, shallow conflict exposure and heavy repetition, where the win is throughput and a clean exceptions queue. A partner hire is a single, commercially sensitive clearance where the win is depth and a defensible record, and being fast at the wrong thing is worse than being slow. Build one pipeline with two modes, and do not let the batch logic set the standard for the lateral.

Leavers, and the evidence trail

The same record has to unwind. Access removed across every system provisioned, subscriptions cancelled before renewal, walls reconsidered, matters reassigned, the conflicts record updated to reflect where the person has gone. An agent that provisioned from a structured entitlement record can deprovision from it and evidence what went, and when.

That evidence is the point. A regulator, a client audit or a later challenge asks what you checked, against what data, on what date, who decided and on what basis. Agents produce that continuously rather than reconstructing it under pressure.

Where AI fails here

Entity resolution is the weak point, and it is weakest where it matters most. Corporate group data is stale, incomplete and inconsistent between sources, so an agent walking a group tree will sometimes miss a link that exists and will confidently assert links that no longer do. Treat a clean result as an absence of found conflicts, never as an absence of conflicts. Partial prior client lists make this worse: the model reasons over what it was given, and the missing entries are precisely the ones the old firm considered sensitive.

Judgement does not transfer. Whether a historic matter is adverse enough to block, whether an interest is material, whether a wall is sufficient or the hire is simply not viable: those are regulated decisions with named owners, and a missed conflict is not a recoverable error you fix in the next release. Build for a reviewer who can see the reasoning and overturn it, and log the overturn.

Where to point a first build

Start with the clearance queue, then the joiner conflict search. The queue gives risk, HR and the practice group one view and pays back on its own. The conflict search is where the expertise lives. Run it alongside your current process for a full hiring cycle and compare the two on misses, not on speed.

If it helps to talk through how this lands on your own conflicts data, hello@adoptaipartners.com reaches us.