A customer sends a long email explaining what they need. Someone reads it, pulls out the key details, enters them into the customer system, decides who should handle it, and writes a reply asking for the information they left out.
Elsewhere, a colleague is opening three supplier quotes and building a comparison spreadsheet. Another is turning yesterday’s meeting notes into tasks.
All of this is necessary work. It also eats up time before anyone reaches the part that needs their experience: advising the customer, choosing the supplier, or delivering the project.
These are useful places to consider AI. But do you need it to automate your business?
No. Many valuable automations work perfectly well with straightforward rules. AI becomes useful when a process involves interpreting information that arrives in different words, formats or layouts.
Understanding the difference helps you invest in improvements that make everyday work easier.
Plenty of useful automation needs no AI.
Suppose someone completes your website enquiry form. They enter their name, email address and the service they are interested in.
An automation can take those details, create a customer record, assign a follow-up task and notify the appropriate person. Everything it needs is already in clearly labelled fields.
The same principle applies when you want to:
- Send a reminder when a task reaches its due date.
- Create a project folder after a quote is marked as accepted.
- Transfer an approved customer address between systems.
- Send a standard acknowledgement using an existing email template.
Each process follows a defined instruction: when this happens, do that.
Adding AI to these steps can introduce cost and uncertainty without improving the result. A calculation should use a formula. A deadline reminder should use the recorded date. Copying a known address does not require interpretation.
Sometimes, the most effective improvement is simply connecting the tools you already use.
Where AI adds something useful
Business information often arrives in a less convenient form than a completed form.
Customers explain their requirements in emails. Suppliers send documents with different layouts. People leave voice notes containing a mixture of decisions, questions and things to remember.
Before that information can move through a process, someone has to organise it.
AI can help extract relevant details, categorise a request, summarise a discussion or prepare a draft response. An automation can then use the resulting information to carry out the next steps.
The following examples show how that combination could work in practice. Each would need to be configured and tested around the business’s actual documents, systems and approval requirements.
1. Turn a detailed enquiry into a prepared sales opportunity
Imagine a prospective customer sends this email:
“We’re moving into new premises in November and need help getting the network and staff computers set up. There will be around 18 of us initially, possibly 25 next year. We already have laptops but aren’t sure what else we need. Could someone call me on Thursday afternoon?”
Ordinarily, somebody has to read the message, identify the opportunity, record the details and work out what to ask next.
A useful AI-assisted process could prepare that first pass automatically.
It could identify the required service, current and expected staff numbers, available equipment, target timing, and requested contact window. Those details could become an organised enquiry brief, with the original email attached.
It could also compare the email against a checklist of information needed for this type of work. In this example, the premises address, confirmed moving date and existing internet arrangements are missing.
The process could prepare a draft reply asking for those details. Defined routing rules could assign the enquiry to the relevant team, while anything that does not fit an agreed category goes to a general review queue.
The person handling the enquiry would open a record containing the customer’s requirements, the gaps to clarify and a suggested response. They could review it and concentrate on the conversation.
There are important distinctions to preserve. “Possibly 25 next year” is a future estimate. “Thursday afternoon” is a requested contact window, not a confirmed appointment. The automation should retain those qualifications.
Pricing, availability, and delivery promises still need an authorised person’s approval.
The value is a better-prepared first conversation, with less time spent reconstructing what the customer needs.
2. Turn supplier documents into a useful comparison
Comparing quotes sounds simple until the documents arrive.
One supplier includes delivery. Another lists it separately. A third offers a lower initial price but adds a recurring support charge. The information you need may be scattered across tables, notes and an accompanying email.
Someone has to bring it together before the prices mean anything.
An AI-assisted workflow could read each quote and prepare a comparison using the fields your business actually cares about: the product or service offered, quantities, price, currency, delivery charges, installation, lead time, recurring costs and stated exclusions.
Each entry could link back to the original document. Where a detail is absent or unclear, the comparison should say so.
Consider two hypothetical equipment quotes. Supplier A quotes £4,800 including delivery and installation. Supplier B quotes £4,500, plus £150 for delivery and £400 for installation.
Once those figures are extracted and checked, standard calculations show totals of £4,800 and £5,050, assuming the same quantities and tax basis. The comparison could also highlight that Supplier B’s delivery date has not been confirmed.
That gives the buyer specific questions to resolve before placing an order.
The person reviewing the comparison still needs to confirm that the products, specifications and terms are comparable. A blank warranty field means “not found in the supplied information”; it does not establish that there is no warranty.
You could use the same approach to prepare purchasing requests, collect information from delivery documents or assemble a supplier information register.
The useful outcome is reaching a purchasing decision with the relevant information already organised and the gaps visible.
3. Turn meeting notes into work that actually gets followed up
A productive meeting can create an unproductive afternoon of administration.
Someone still has to write up the discussion, identify the actions, add them to the project system and tell the people responsible. Until that happens, agreed work may exist only in someone’s notebook.
AI can help turn written notes or a suitable transcript into a proposed action list. An automation can then carry approved actions into the tools the team uses to manage work.
For example, imagine your notes from a client meeting contain these three points:
- “Maya to send the revised drawings by 14 November.”
- “Ask the client whether installation can happen outside normal hours.”
- “We might include the upstairs area in a later phase.”
A well-designed process would treat them differently.
The first is an action with a named owner and a stated deadline. The second needs an owner before it can be assigned. The third is a possibility to record, rather than an agreed task to schedule.
The proposed actions could appear in a review screen alongside the relevant notes. After approval, the workflow could create tasks, attach them to the correct project, notify their owners and prepare a client follow-up email.
This can also be useful after site visits. A team member could record a short debrief while the details are fresh. The recording could be transcribed and used to prepare a visit report, proposed actions and questions for the office to follow up.
Names, dates, measurements and commitments need checking against the source. If something was never agreed, the system should leave it open for confirmation.
The benefit is getting from conversation to coordinated action with less administrative work in between.
The connections matter as much as the AI.
You can already paste an email into an AI tool and ask for a summary. That may help, but you are still responsible for moving the information, updating the customer record and creating the task.
A connected process can bring those steps together.
For an enquiry, that might mean a designated inbox starts the process, AI prepares the brief, a person checks the response, and approval releases the agreed follow-up actions.
The AI handles a specific interpretation task. Ordinary rules control where the information goes and what happens next. People make the decisions that need their knowledge or authority.
The design should also account for uncertainty and failures. An unreadable attachment or an unavailable customer system should leave a visible item for someone to resolve. The team needs to know when work has stopped.
Put checking where it earns its keep.
AI can produce an answer that reads well but misses a qualification or gets a detail wrong. Good presentation is not proof of accuracy.
Checking should reflect the consequences of an error.
An internal draft summary may need a quick review. A price, payment instruction, purchasing decision or promise to a customer needs stronger controls before it is acted on.
Useful safeguards include keeping the original message or document accessible, clearly marking missing information, and requiring approval before consequential actions. Where appropriate, check numbers using formulas or known records.
The review step should be practical. Showing someone the extracted delivery date beside the original quote is more useful than asking them to approve a summary with no supporting material.
Access matters too: the workflow should use approved tools and accounts, process the information needed for its task and send the results only to the appropriate people.
Choose a first project with a clear payoff.
Look for work where somebody repeatedly reads, interprets and re-enters similar information.
Then identify exactly what a useful result would look like. “Help with emails” is broad. “Prepare new sales enquiries with requirements, missing details and a draft response” gives you something specific to assess.
A good starting point has enough volume or consequence to justify the work, source material you can check, and an output that is easier to review than to create from scratch.
Test it on a representative sample, including incomplete emails, awkward documents and ambiguous notes. Measure the whole process: time spent reviewing and correcting, as well as time saved preparing. Include software and running costs when judging the return.
You may discover that straightforward automation solves most of the problem. You may find that one well-placed AI step removes the part that previously kept someone tied to their inbox.
Both are worthwhile outcomes.
Start with the work you want to improve.
Your business does not need AI in every process. It needs reliable ways to turn incoming information into useful action.
When the next step is obvious, and the information is already organised, rules are often enough. When someone must repeatedly sift through emails, documents or notes before anything can happen, AI deserves a closer look.
Picture starting the day with enquiries prepared for follow-up, supplier quotes ready to compare and yesterday’s agreed actions already in the project system.
That is a practical ambition: less preparation work between receiving information and doing something useful with it.
Have a process like this in your business? Contact Company Compass to explore where automation and AI could reduce the admin and help your team move work forward.