What should you automate first?

You've seen enough demos to know what AI can do. The harder question is where to start in your own business, where every process is half documented and the people who run it are already busy.

The first system you put into daily work decides whether there is a second. When it runs cleanly, the people around it start asking what else it could take on. When it sends one wrong quote, the next conversation about AI in your business starts from that quote. So the choice deserves more care than the tool does.

Four questions that sort any process

Write down the ten processes that take the most time across sales, operations and customer service. Ask the people doing the work, not the org chart. Then put four questions to each one.

How often does it happen?

Daily beats weekly. A small task that happens every working day costs more over a quarter than the big task that happens twice a year, and it is easier to automate, because you already have hundreds of examples of it to test against.

How much judgment does it need?

Separate the steps anyone trained could do from the steps that need someone who knows the customer, the margin or the history. Reading an invoice, matching it to an order and queuing it for approval is the first kind. Deciding whether to accept a late delivery from a key supplier is the second. The first kind automates cleanly. The second stays with a person, and the system's job is to put everything they need in front of them.

Where does the information come from?

A process that starts from a web form, an email or a document in a shared inbox is ready. A process that starts from a phone call nobody writes down, or a spreadsheet that lives on one laptop, needs the input fixed first. That is often a cheap fix, but it comes before the automation, not after.

How would you know it worked?

If you can't name the number that should move, you can't tell whether it did. It might be the time to first reply, the hours spent re-keying each week, or the share of requests resolved without an escalation. Pick the measure now, take the reading before anything changes, and hold the project to it.

Automate around the person, not instead of them

The projects that go wrong are usually the ones where someone tried to remove every human step. The ones that last leave the decision where it was and remove the work around it. A supplier invoice is read, matched to the purchase order and queued. If the amount doesn't match, it goes to the finance lead with the order, the invoice and the difference side by side, and they approve or reject in a minute instead of hunting for the paperwork. Designing that exception path is most of the work, and it is where the trust comes from.

Ask of every candidate: where does a person need to decide, and what would they need in front of them to decide quickly? If you can answer that, the rest of the process is a candidate.

Where the first wins usually are

Across the businesses we talk to, the same handful of processes come up first, because they pass all four questions.

In sales

• Answering new enquiries. They arrive every day from the same few channels, the first reply follows a pattern, and the measure is obvious: how long a lead waits before someone replies. See how fast you should reply to a new lead.
• Follow-up on open quotes. Nobody enjoys it, it gets dropped when the week is busy, and every missed follow-up is revenue that was already half won.

Both sit in Revenue Growth.

In operations

• Invoice and order handling. Read, matched, queued, with the exceptions routed to someone who can decide. How to automate invoice processing walks through it.
• Keeping records in step. When a customer's details change in the CRM, they change in accounting and the project tool too, so nobody is reconciling three systems on a Friday.
• The weekly report. The numbers your team assembles by hand from four sources, arriving on their own.

Operational Efficiency covers these.

In customer service

• Triage. Sorting incoming requests, pulling the order or account from your systems, and drafting a reply for an agent to check and send.
• Answers from approved information. The same answer whoever is asked, with the source attached.

See Customer Experience.

What to leave for later

• Anything that happens a few times a year. You won't have enough examples to test against, and the saving is small.
• Anything where the inputs are a mess. Fix the form or the inbox first.
• Anything customer-facing where you can't yet say how a mistake would be caught. Start with the internal step before it, then extend once the system has earned it.
• Anything that depends on a system with no connection options. Check that before you commit, not after.

None of these are never. They're second or third, once the first system is running and your team knows what good looks like.

Decide the measure before you build

Whichever process you pick, write down three things before any configuration is touched: the number that should move, its value today, and who will read it after launch. The value today is the one people skip, and without it you'll be arguing about whether anything changed. Take it from your inbox, your CRM or a week of timesheets. A rough reading taken before the change is worth more than a precise one taken after.

This is also how you'll know whether the work earned its cost. A system that removed hours of re-keying from each week has a number next to it. A system that made things better in general does not, and it's the one that gets cut when budgets tighten.

What to do this week

1. List the ten processes that take the most time across sales, operations and service.
2. Score each on the four questions: frequency, judgment, inputs, measure.
3. Pick the one that is frequent, low on judgment, starts from a form, an email or a document, and has a measure you can take today.
4. Write down where a person decides in that process and what they'd need in front of them.
5. Take the measure's reading now, before anything changes.

If you'd like a second opinion on the list, that's what the first call with us is for. Bring the ten processes and we'll tell you which one we'd build first, why, and whether it needs AI at all.