Requests your team can settle
without waiting on you.

Why requests wait for you

A request is anything someone asks the business to do: a customer who wants a credit, a supplier who proposes new terms, a colleague who needs a day off or wants to reorder stock. In many owner-led businesses most of these could be settled by someone in the team, and they still end up waiting for the owner, for one of three reasons.

The first is that the limit exists only in your head. The person who takes the request knows you sometimes accept it, but not up to what amount, for which customers or in which cases, so asking you is the only safe answer. The second is that the request arrives incomplete: the order reference is missing, there is no photo of the damage or the dates are unclear, and someone has to go back to the customer before anyone can decide. The third is that nobody else owns that kind of request, so it lands with you by default, on top of everything else you were doing.

Each of these requests is quick for you to answer, but the customer, the supplier or the colleague waits until you have replied, and when you are in meetings or away the waiting grows. We have seen it in businesses we work with: the team needs the owner’s yes and waits, because the owner is buried in requests that someone else could have settled.

Write the limits once

Before any AI is involved, you write down, once, who may say yes to each kind of request and up to what point: credits for damaged goods with the customer service lead up to the limit you set, delivery changes with customer service as long as the new date fits the delivery calendar, reorders with operations from the suppliers you have approved, days off with the team leader under the roster rules, prices on a quote with the sales lead within the range you allow. Whatever falls outside those limits comes to you.

Request limitsEdited by you

Who can say yes, and up to where

The page the AI checks every request against

• Kind of requestWho decidesUp to, and above it
• Credit for damaged goodsWith a photo and the order referenceCustomer service leadUp to the credit limit you setAbove it: to you, prepared
• Delivery date changeAsked by the customerCustomer serviceWhen the new date fits the delivery calendarOtherwise: to operations
• Reorder of stock or suppliesFrom the stock listOperationsFrom approved suppliers, at the agreed priceA new supplier: to you, prepared
• Day off or roster swapAsked by a team memberTeam leaderUnder the roster rulesA clash on the roster: to operations
• Price on a quoteBelow the price listSales leadWithin the range you allowBelow it: to you, prepared
• New contract, new supplier, a change to your pricesAnything that commits the businessYouAlways, with the facts gathered first

Each kind of request has a named owner and a named stand-in for when they are away.

Our illustration of a request limits page for a fictional business. The limits themselves are yours to set.

That page is short, and it is the document the AI checks every request against. An AI cannot apply a limit nobody has written, so a business that skips this step ends up with an assistant that either asks you about everything or guesses. Writing it is also the moment to spot the requests you approve out of habit and could hand over entirely.

Each kind of request also gets one person who owns it, with someone named to stand in when they are away, so that nothing returns to you by default. When a limit changes, you or that person edit the page, and the next request is checked against the new version rather than the one somebody remembers.

What the AI does with each request

Once the limits are written, the AI reads each request where it arrives: the shared inbox, the form on your website, or the channel your team uses in Google Chat, Teams or Slack. It works out what kind of request it is and checks that it has what is needed to decide. If something is missing, it asks the person who sent it, in the wording you have approved, so a customer who reports damage is asked for a photo and the order reference before anyone in the team has opened the email.

It then gathers the facts the decision depends on from the systems where they are kept: the order and its delivery date, what was agreed with that customer before, the credits already given. It checks the request against the written limit and sends it to the person who owns that kind of request, with a short summary, the facts, the rule it applied and a proposed reply. That person approves, changes or refuses the proposal, and the reply leaves under their name.

Our illustration with a fictional business. It is not the interface of Gmail, Google Chat or Xero.

Drafting the reply with the right context is the part we describe in using AI to answer your customers’ emails. Here the difference is who receives the request, and which limit decides whether anyone else needs to see it.

What still comes to you

Some requests should keep reaching you, and the written limits say which ones. They come to you prepared: the request, the facts gathered, the limit it does not fit and a proposed answer, so that deciding takes the time of reading one page rather than the time of finding everything out again.

• Anything above the limits you set, such as a credit larger than the customer service lead may give, or a price below the range you allow.
• A kind of request nobody has written a rule for yet. Once you have answered it, the rule is written, and the next one goes to the person who owns it.
• Anything that commits the business: a new supplier, a new contract, a change to your prices.
• A complaint that could turn into a dispute, or a request that touches someone’s employment.
• A request the AI cannot place with confidence, which goes to a person rather than to its best guess.

How a decision is prepared for you is set out in the decisions waiting on you, prepared by AI, and how your team finds the answers it needs without asking you in answers your team can find without asking you.

What the AI may do on its own

How much the AI does without a person is part of what we agree with you before building anything, one kind of request at a time. It can prepare each request and pass it on while a person decides; it can prepare the reply and the action so that the person only has to approve them; or, for routine requests inside a written limit, it can settle them alone, record each one and send any uncertain case to a person.

We do not force that third level where the variables, the access it would need or the judgement involved would make it irresponsible. Actions that move money or speak to a customer, such as a credit note or a reply, are the ones we keep behind a person’s approval at first, until requests of that kind have been handled correctly for long enough to trust the limit.

Whatever it does, the AI acts with the access it has been given, and that access follows each person’s role, as we explain in adopting AI securely. Where the approval sits, and what the approver checks each week, is covered in handing a repetitive task to AI while keeping the approval.

Where to start

Start with the kind of request that waits for you most often and whose answer rarely changes: credits for damaged goods, delivery changes, reorders from the usual suppliers. Write its limit, name the person who owns it, and let the AI bring each request in complete and pass it on, with that person approving every one for the first weeks.

What tells you it works is whether that kind of request still reaches you. When it no longer does, the next kind follows, and each request you answer that no rule covered becomes a rule, so it stops coming back. A short weekly summary of what was settled, by whom and under which limit, keeps you aware of what happens in the business without anyone having to ask you.

How we start

When we set up AI around a business, we organise the business knowledge your AI needs, set up the working environment for you and your team with permissions that reflect each person’s role and clear points for human approval, and build the automations with your team, testing them against your actual processes. Your team learns through its own day-to-day tasks, so the person who owns a kind of request knows how to read what the AI prepares and when to overrule it.

The first step with us is a free AI pre-audit: 45 minutes on a call where we look at how you and your team use AI today and which work to hand over first. Bring the requests that waited for you last week. You leave with a clear next step, with or without us. See how we help, who we are and how we handle security. In the news today: Microsoft’s Frontier programme now covers Outlook, Teams, OneDrive and SharePoint.