Fair enough. The menu has accounts, contacts and deals. Every customer has a history. But that is not what we built it for, and the difference changes how your team spends its day.

The trouble with a CRM in a small company

A CRM is a database that somebody has to feed. It works well when the team logs every call and every quote, and moves each deal to the next stage on the day it actually moves. In a small sales team, that logging is the first thing to go when the day gets busy.

Requests arrive by email and WhatsApp. The reply has to go out today. Updating the CRM gets pushed to later, and often to never. Over time the CRM says one thing and the ERP holds the real version of the orders. Anyone who wants to know where a customer stands ends up asking a colleague.

What Crafteer.ai does

Crafteer.ai starts from the work that comes in. It reads the company's emails (Outlook or Gmail) and WhatsApp messages and works out what is being asked for, such as a quote or an order. It pulls out the details that matter: the customer, the product codes, the quantities, the deadline.

Then it does what a person would do next. It goes to the ERP (PHC, Primavera or Sage) to check prices and stock, drafts the order or quote and writes the reply. When someone on the team approves, everything goes to the ERP and the customer. Each process runs as a procedure written by your company, step by step.

Internal work runs on its own. Anything going out to a customer waits for a person. The email or message appears in the chat as a "Waiting for you" card, with the draft ready. Whoever is on duty clicks Approve and send. If they want to change something, they click Continue in chat and ask for the change, typed or spoken.

There is one more thing a CRM does not do by itself. Crafteer.ai follows each customer's history and flags when something is out of the ordinary, like a quote that has gone unanswered for days. It brings the follow-up message already written, waiting for approval like everything else.

So what about the CRM?

It stays up to date as a by-product of the work. When Crafteer.ai handles a quote request, the contact is linked to the right account and the deal is created or moved forward. The quote that went out sits in the customer's history. Nobody had to open a record and fill in fields.

Accounts, contacts and deals are still there for the team to look up. The reason to use Crafteer.ai lies elsewhere, though: in the work that no longer has to be done by hand.

How it differs from an email client

There are very good tools for managing an inbox. Superhuman is the best-known example. It makes email faster, with keyboard shortcuts and AI help for drafting replies. For people who live in their inbox, it saves real time.

In a small business, however, writing the reply is usually the shortest part of the job. Before it, someone went to the ERP to look up the price, checked the stock, worked out the lead time and sometimes entered an order. An inbox tool helps you write the email. It does not go into the ERP for you.

For Crafteer.ai, email and WhatsApp are the way into a process. What we care about is everything that happens between the message arriving and the reply going out.

How it differs from Zapier or n8n

Zapier and n8n connect applications through workflows: when this happens, do that. They work very well for tasks with fixed rules and data that always arrives in the same shape, such as moving website form submissions into a spreadsheet.

Customer requests are rarely like that. One customer puts the product code in the subject line. Another sends a photo of a handwritten list, someone else forwards a three-page PDF, and there is always the one who asks for "same as last time, but ten more boxes". A rules-based workflow needs someone to design it for each case and fix it when the format changes. More cases means more workflows to maintain.

In Crafteer.ai, you describe the procedure the way you would explain it to a colleague who started this week, in words or by voice. The AI reads each request as it comes and follows the steps. When information is missing or something does not add up, it asks the team in the chat rather than guessing.

So what is Crafteer.ai?

We call it an AI workspace for sales and back-office teams. Your team works in a chat. Crafteer.ai reads the requests, does the work in the systems the company already uses and puts in front of people only what needs a decision. Along the way, the CRM keeps itself current.

If your team spends a good part of the day copying details from emails into the ERP, that is usually where we start. You can see the whole flow on How it works, or read our guide to handling quote requests by email.

AI does not learn that way. It is not in the room, it does not overhear conversations and it knows nothing about your company. To do the work the way your team does it, someone has to tell it how. That is what a procedure is for.

In short: an SOP is a step-by-step description of how your company runs a process. AI needs one to do the work the way your team would, the same way every time, and to know where to stop and ask a person for a decision.

What an SOP is

SOP stands for standard operating procedure, often just called a procedure. It describes the steps of a process: what starts it, where each piece of information comes from, what to do in each case and who decides when a case falls outside the rule.

Many companies have written procedures for quality or audits, kept in a folder nobody opens. The procedures we mean here are different. They describe everyday work, like handling an order or answering a quote request, and they are actually used every time the process runs.

Why AI needs one

A good AI model reads and writes very well. It can make sense of a confusing message, work out which products the customer wants and write a polite reply. But it does not know whether the customer has a special price, what to do when stock runs short, or whether an order below a certain value pays for delivery.

Without a procedure, the AI improvises. Sometimes it gets it right and sometimes it does not, and in a business that is not acceptable. A wrong answer to a customer or an order entered incorrectly costs money and trust.

With a procedure, the AI does what the team would do, in the same order. Every step is logged, so at any moment you can see what was done and why.

When only one person knows how it is done

Writing procedures also solves a problem companies had long before AI.

In most companies, processes and business rules are not written down anywhere. The people who do the work every day are the ones who know them. Often only one or two people know how to do a task from start to finish: which customers have special terms, which carrier is used for each area, what to reply when an item runs out.

While those people are around, everything works. The trouble starts when they are not:

• During holidays, requests wait for whoever knows, or are handled by a colleague who has to guess.
• During sick leave, the same happens, with no end date.
• When someone leaves the company, they take years of rules that nobody ever wrote down with them.
• When someone new joins, they learn only what they manage to see and ask in the first few weeks.

Writing the procedure takes that risk out of the business. The rules stop depending on one person's memory and sit somewhere the whole team can see and correct them. Because Crafteer.ai follows the procedure every time the process runs, the task keeps being done the same way, whoever is on holiday.

An example: an order by WhatsApp

A regular customer sends this message to the company number:

Morning, please send 40 boxes of the blue gloves and 12 of the big degreaser, same as last time. Needed for Thursday.

A procedure for handling WhatsApp orders might look like this:

1. Identify the customer by mobile number, from the contact record. If the number is not linked to any account, ask the team in the chat.
2. Work out which products they are. Customers rarely give the product code: they write the name they use day to day, which is often not the one in the ERP. Match the request against the customer's purchase history and previous orders to find the right codes. Here, “blue gloves” is product 2210 and “big degreaser” is 2215, the 5-litre one, which this customer orders every month. If more than one product could fit, ask the team.
3. Check the ERP: the stock of each product and the price that applies to this customer.
4. If everything is in stock, create a draft order with the requested delivery date.
5. If a product is short, prepare a message for the customer saying what can go now and asking whether they would rather receive a partial delivery or wait for the rest.
6. Prepare the confirmation for the customer, with the quantities and the delivery day.
7. Wait for the team's approval. The order and the message sit in the chat, ready to approve.
8. Once approved, send the order to the ERP and the confirmation to the customer, log everything on the customer's account and update the deal.

From step 1 to step 6, Crafteer.ai works on its own. Step 2 is where the most time is saved: someone who knows the customer knows what “the blue gloves” are, and Crafteer.ai gets there through the purchase history, without anyone having to look it up.

Notice too that the procedure does not try to foresee everything. It says what to do in the normal case and where to stop and ask.

How you write one in Crafteer.ai

In Crafteer.ai, you write a procedure by describing the process in words or by voice, the way you would explain it to a new colleague. The description is organised into steps, which the team can review and correct whenever the process changes.

A few tips that help:

• Start with the most common case. Exceptions can be added later, as they come up.
• Say where the information lives. “The price is the one on the customer's price list in PHC” works better than “check the price”.
• Write down what to do when in doubt. A product that cannot be identified, an unknown customer, an odd quantity. In these cases the best move is nearly always to ask the team.
• Be clear about what waits for a person. In each procedure your company chooses which steps run on their own and which wait for approval, and can change that whenever it likes.

During implementation, the LucidHorizon team writes the first procedures with you, based on real requests from your company.

Where human approval sits

Approval sits wherever your company decides, procedure by procedure. In the example, the team chose to approve the order and the message to the customer before they go out.

Both appear in the chat as a “Waiting on you” card, with the draft ready. Whoever is following the work clicks Approve and send, or Continue in chat to ask for a change, typed or spoken. As soon as someone approves, the order goes to the ERP and the confirmation to the customer. In the time it took to enter one order by hand, you approve several.

When the procedure says to ask the team, for example because two products could fit, the question appears in the same chat.

Frequently asked questions

Is an SOP the same as an automation?

A classic automation follows fixed rules and stops when a case comes up that it did not foresee. A procedure in Crafteer.ai describes the work in words, the way the team does it, and the AI can handle requests written in many different ways, such as an order with no product codes. When it is unsure, it asks the team instead of guessing.

Which processes should I start with?

The ones that repeat many times a week and nearly always follow the same steps: orders, quote requests, stock questions and lead times.

Can the AI send messages to customers without approval?

Only if your company sets it up that way. In each procedure you choose which steps run on their own and which wait for approval.

Further reading

To see how this applies to quote requests, read the guide on answering quote requests by email and WhatsApp. For WhatsApp orders, see the distribution page.

Good morning. Could you quote us for 40 boxes of the blue nitrile gloves and 15 boxes of the big degreaser, same as last month, delivered to our Braga shop by the 30th? Thank you.

Anyone who has handled requests like this knows the reply does not take two minutes. This guide walks through the usual process, where the time goes and how the same request is handled with Crafteer.ai.

How it is usually done

In many companies that sell to other businesses, the process goes roughly like this:

1. Someone opens the email or the message and works out who the customer is. Sometimes from the address or the number, sometimes from the company name in the signature.
2. They open the ERP and find the customer record.
3. They work out which products are meant. The customer writes “blue nitrile gloves” and the ERP has the item under another name, so they go through the customer's previous orders to find the codes. Then they check the price that applies to that customer.
4. They check stock and work out whether the requested date is achievable.
5. They create the quote in the ERP, or in a separate document, and export a PDF.
6. They go back to the email or WhatsApp, write the reply, attach the quote and send it.
7. If there is a CRM, they log the quote there. If the day is busy, that waits until tomorrow.

None of these steps is hard. The trouble is that they repeat dozens of times a week, and each one means switching screens.

Where the time goes

• Going back and forth between email, WhatsApp and the ERP. Every request means copying details from one place to another, sometimes line by line.
• Requests without product codes. The customer writes the name they use day to day, which is rarely the one in the ERP, or sends the list inside an attachment. Each line means one more search.
• Waiting. The request sits in the inbox or on the phone until someone has half an hour free. For the customer, that is the time that counts, and it is often when another supplier replies first.
• Typing errors. A wrong quantity makes it into the quote and, if nobody spots it, into the order.

How it works with Crafteer.ai, in five steps

1. The request arrives

The request arrives as it always has: in the company inbox, in Outlook or Gmail, or on WhatsApp, on the company's or the sales rep's number. Crafteer.ai handles both in the same way, and the customer does not change anything about how they ask.

2. Crafteer.ai reads the request

It recognises a quote request and pulls out the details: who is asking, which products, what quantities, where and by when. Customers rarely give the product code, so it matches the names they use against their purchase history and previous orders to find the right items in the ERP. If the list comes as an attachment, it reads the attachment. If two products could fit, it asks the team.

3. It checks the ERP

It goes to PHC, Primavera or Sage for the customer record, the prices that apply and the stock for each item. Nobody had to open the ERP.

4. It prepares the quote and the reply

It follows your company's procedure for quote requests: which terms apply, what to say when an item is out of stock, which wording to use. It prepares the quote and drafts the reply email.

5. The team approves

The reply appears in the chat as a “Waiting on you” card, with the quote and the draft ready, whether it is an email or a WhatsApp message. If it looks right, click Approve and send. If you want to change something, click Continue in chat and ask, typed or spoken, for example: “give 5% off the item that is out of stock”. Crafteer.ai redrafts it and puts it back in front of you.

Once sent, the reply is logged on the customer's account and the deal is updated.

What if the request comes in by phone?

The phone is not connected to Crafteer.ai yet, but the request does not have to end up on a scrap of paper. Whoever takes the call describes it in the chat, typed or spoken, during the call or straight after:

Rui Matos called asking for a quote for 40 boxes of the blue gloves, delivered to Braga by the 30th.

From there, the process is the same. Crafteer.ai identifies the customer and the products, checks the ERP and prepares the quote and the reply for approval.

What runs on its own and what waits for you

The rule is simple. Internal work runs on its own. Anything that goes outside, or that needs a decision, waits for a person.

Runs on its own:

• reading and sorting incoming emails and WhatsApp messages;
• pulling details from the request and its attachments;
• checking prices, stock and lead times in the ERP;
• logging the request and the quote on the customer's account.

Waits for the team:

• the reply with the quote, before it goes to the customer;
• discounts and exceptions the procedure does not cover;
• requests with missing information. In those cases, Crafteer.ai asks in the chat rather than guessing.

Where to start

You do not need to handle every kind of request at once. We suggest starting like this:

1. Pick one type of request. Quotes for catalogue products are a good start, because the steps almost always repeat.
2. Gather a few real requests of that type, by email or WhatsApp. Include the messiest ones. Those are the ones that show the exceptions.
3. Describe what the team does. The way you would explain it to someone who joined this week: where the price comes from, what happens when there is no stock, who decides on a discount. You can describe it in writing or by voice. We explain how in a separate guide on what an SOP is.
4. Review everything that goes out. At the start, it makes sense to approve each reply calmly and ask for changes. That is how the team builds trust in what Crafteer.ai prepares.

Implementation is done by the LucidHorizon team, after a demo. We connect your email, WhatsApp and ERP and write the first procedures with you. If you work in made-to-order manufacturing or distribution, each industry page has examples closer to your day-to-day work.