top of page

AI for sorting inbound post and email into the right hands

  • 3 days ago
  • 3 min read

Updated: 12 hours ago

Introduction


In most small businesses everything arrives in the same place. A shared inbox and a pile of post contain enquiries, invoices, complaints, statutory notices, marketing, delivery confirmations and a legal letter with a deadline, and one person sorts them between other tasks. The consequences of a mis-sort range from a slow reply to a missed statutory deadline.

Triage is a good automation candidate because it is high volume, repetitive, and governed by patterns rather than judgement in the great majority of cases. It is also a place where the failure mode is asymmetric: the cost of routing an enquiry slowly is small, and the cost of not noticing a formal notice can be very large, so the design has to be built around the important minority.


1. AI for sorting inbound post and email is a routing problem


Not a reading problem.

Deciding what each item is and who needs it, quickly and consistently. The reading is done by whoever receives it, and the value is that they receive it within an hour rather than a day.


2. Define the categories from what actually arrives


A week of observation.

Enquiry, existing customer, invoice, statement, complaint, supplier query, statutory or legal, recruitment, marketing. Your list will be short and specific to you, and it should come from what you receive rather than from a template.


3. Route enquiries fastest


Where the revenue is.

New enquiries are perishable and everything else can usually wait an hour. Making enquiries the highest priority category, with a target for acknowledgement, is the change with the clearest commercial return.


4. Flag anything with a deadline for a person immediately


The asymmetric risk.

Legal correspondence, regulatory notices, tax matters, formal complaints and anything mentioning a time limit. These should never sit in a queue, and the rule should err heavily towards over-flagging.


5. Do not auto-reply to anything that needs judgement


The obvious trap.

An acknowledgement is fine. A substantive automatic response to a complaint, a legal letter or a technical enquiry creates a record you may have to live with, and it is read as dismissive.


6. Keep an audit trail of what arrived and where it went


Both operational and protective.

Date received, category, who it was routed to, and when. Disputes about whether something was received or actioned are common, and this record settles them.


7. Watch the misroute rate


The quality measure.

A sample check of where items went, weekly at first. Systematic misrouting of one category is easy to correct and impossible to notice without looking, and it usually points at an ambiguous category definition.


8. Handle personal data in correspondence carefully


A compliance dimension.

Inbound post and email contain personal data, sometimes sensitive. Where it is processed, who can see it, how long it is retained and how it is secured are obligations that vary by jurisdiction.


9. Use the categories as management information


An underused output.

Volumes by category over time tell you about complaint trends, enquiry seasonality, supplier query rates and where administrative effort is going. This is free information from a process you were running anyway.

Be careful about routing everything away from a human overview. Someone should still scan what arrived each day, because the item that does not fit any category is frequently the one that matters most.


Conclusion


Design the triage around the important minority, because the cost of missing a deadline is not symmetric with the cost of a slow reply.

Define the categories from a week of what actually arrives, route new enquiries fastest because they are perishable, flag anything with a deadline to a person immediately and err towards over-flagging, acknowledge automatically but never respond substantively to complaints or legal correspondence, keep a record of what arrived and where it went, sample the misroute rate weekly, handle personal data in correspondence according to your obligations, and use the category volumes as management information.


Related reading


 
 
 

Comments


bottom of page