The useful comparison for AI-assisted communication is rarely a perfect message written by a well-informed professional against a machine-generated alternative.
In practice, the choice may be between a suitable update delayed by three days and a draft prepared earlier for the relationship owner to check, adapt and send. Used carefully, AI can reduce the administrative effort between an intention to communicate and the client receiving something useful.
The same capability can erode a relationship at scale. A fluent message may be accurate in isolation and completely wrong for the client's circumstances. Speed and consistency are operational measures; neither proves that communication has improved.
The source article opened with percentages about how attorneys and clients perceive care. No source was attached, so those figures have been removed. The underlying commercial problem is recognisable without them: professional services firms depend on clients feeling known, and that judgement is shaped by timing, relevance and accountability.
Use AI where structure consumes the time
Good candidates have repeated structure, reliable inputs and a named person responsible for the meaning.
Routine status updates. A drafting tool can assemble agreed milestones, decisions and next actions from approved sources. The relationship manager still checks accuracy, explains exceptions and adds the observation that comes from knowing the engagement.
Plain-language summaries. Complex reports can be easier to approach when a concise explanation sets out the issue, consequence and required decision. A qualified professional must compare the summary with the underlying document. In legal, financial or other high-consequence work, a plausible omission can matter as much as an obvious error.
Meeting follow-up. AI can turn notes into a proposed record of actions and owners. Participants need a way to correct it, and the firm must decide whether the source material is appropriate for the selected system.
Scheduling and reminders. Low-judgement logistics may remove avoidable delay. The workflow should still recognise complaints, vulnerability, bereavement, restructuring or other contexts where an automated sequence could be insensitive.
Translation and accessible alternatives. AI may help create a draft in another language or at a different reading level. Specialist review remains necessary where nuance affects rights, obligations or advice. Accessibility also involves structure, format and compatibility, rather than simplified wording alone.
The source describes a 200-person consultancy reducing routine update drafting from 45 minutes to 12. It also claims 60% to 80% reductions across Distinction engagements. These may be valid internal observations, yet they need engagement records, a definition of the measured task and publication permission. They should be treated as evidence to verify, rather than a universal promise.
Decide which messages cannot enter the workflow
The first design exercise should identify exclusions. Complaints, material service failure, adverse advice, sensitive personal circumstances and relationship-threatening decisions need direct professional ownership. Drafting assistance may still be possible in some governed environments, although automation should never decide the response or send it without deliberate review.
The source includes an instructive financial-services account. A firm improved response-time and consistency metrics after expanding AI-assisted communication, then lost two clients who said the service no longer felt personal. One client in a restructuring received an upbeat delivery update that ignored redundancies and a board change. Nobody had reviewed it.
That case needs internal verification before it is presented as Distinction experience. Its mechanism is credible and important: a system can see project status while missing the relationship context that changes how, or whether, a message should be sent.
Create an escalation list with the people who manage difficult client situations. Include keywords only as one signal, since a rules list will miss context. Give users a simple way to pull a draft out of the automated route. A stopped message can be a successful workflow outcome.
Human review is an accountable act
“Human in the loop” is too vague to govern a client communication. The named reviewer needs access to the source, authority to change or reject the draft, time to perform the check and responsibility for the final message.
A useful review asks:
- Are the facts supported by the approved source?
- Does the message reflect what I intend to say?
- What relevant client context is absent?
- Is the tone appropriate to the event?
- Are confidentiality, privilege, privacy or contractual restrictions engaged?
- Should this be a call or a personally written response instead?
Approval cannot become a click performed while the draft remains unread. Sample completed communications, compare them with source records and examine corrections and near misses. If review quality declines as volume grows, reduce the automation or change the workload.
The professional remains accountable for the communication. A product interface describing text as a “draft” does not transfer that responsibility to the supplier.
Data handling shapes the viable design
Client-specific drafting may involve personal data, confidential information, commercially sensitive plans or privileged material. The team must know what enters the system, where it is processed, how long it is retained, whether it is used to improve a model and which suppliers or territories are involved.
The UK's Information Commissioner's Office provides guidance on AI and data protection. A firm should involve its privacy, security, legal and sector-risk owners in proportion to the information and use. Client terms, professional duties and internal information-classification rules may set stricter boundaries than general data-protection compliance.
Use approved systems and minimise the data supplied. Retrieval should respect the user's permissions. Templates and prompts need version control when they influence regulated or contractual communication. The design should also preserve an appropriate record of the source, draft, reviewer and final output.
These controls may rule out a proposed use. That is a design result, not a failure of ambition.
Fit the assistance into the existing work
Adoption depends on whether the workflow removes the blank-page and information-gathering burden. A separate application that requires repeated copying can create friction and new data leakage risk. An embedded tool can be easier to use, though integration alone says nothing about suitability or governance.
Map the complete task:
- What event triggers communication?
- Which approved information sources are needed?
- Who receives the proposed draft?
- Who owns review when the usual person is absent?
- How is sending prevented before approval?
- Where is the final record stored?
- How are corrections, complaints and opt-outs handled?
Then test the workflow with a small group and representative scenarios, including awkward ones. Compare assisted and unassisted work for accuracy, time, edits, escalation and recipient response. Users should be free to reject the draft.
Measure relationship quality as well as throughput
Response time and drafting effort reveal whether the operation has changed. They do not establish client value.
Track a balanced set of signals:
- Time from a genuine trigger to a useful response
- Factual corrections and material omissions
- Percentage of drafts rejected or substantially changed
- Complaints, escalations and inappropriate-send near misses
- Client feedback about clarity, relevance and feeling understood
- Differences across relationship owners and client groups
- Adoption, including reasons people avoid or bypass the workflow
Establish a baseline before the pilot. Ask clients about communication in normal relationship conversations and give them a practical route to express preferences. Avoid treating a broad NPS movement as proof of one workflow's effect.
AI assistance should allow professionals to communicate at the right time with better preparation. It should never become a rationale for increasing message volume or removing the person who knows why the communication matters.
The source's best observation survives the evidence clean-up: the enemy of good client communication is often time, especially when a full inbox erodes sound intentions. AI can recover some of that time. The firm still has to spend it on judgement, context and care.
If you want to understand which of your current client communication workflows are the best candidates for AI assistance - and how to design the human-in-the-loop process that maintains quality - book a client communication workflow review. We've also put together a client communication AI readiness checklist (free download below) that covers the implementation criteria I've outlined above - it's a useful starting point before you commit to any tooling.
