Somewhere in your business there is a Word document that gets opened, saved under a new name, and edited in four or five places. A name, an address, a couple of figures, a date. Then it gets read through, saved as a PDF, and sent.
Then it happens again. And again. Two hundred times for a price increase letter, or every single time a quote goes out.
Nobody has ever put this on a roadmap, because it does not feel like a system. It feels like typing. But it is one of the largest pieces of unautomated work left in a typical SME back office, and unlike most inefficiencies it gets worse as the business grows rather than better.
What it actually costs
Two things, and the second is the one that hurts.
The obvious cost is time. Three minutes a letter does not sound like anything until it is two hundred letters, at which point it is a person's entire week, done by someone who was hired to do something more useful.
The real cost is the errors. When a document is produced by copying the last one, a proportion of them will go out with something from the previous customer still in them. It is not carelessness, it is arithmetic: the failure mode is built into the method. And because each document is checked by the person who just made it, the mistakes that survive are exactly the ones that are hard to see when you already know what it is supposed to say.
Those letters are not neutral. A quote with the wrong figure is a commercial problem, and a letter with another customer's name and address on it is a data protection incident, however small. Both cost more to deal with than the letter took to write.
What good looks like
The version of this that works is less dramatic than people expect, and it does not involve replacing anything.
Your Word document stays the template. This is the part that makes it acceptable to the people doing the work. Nobody is being asked to give up Word, learn a new editor, or hand their letterhead to a developer every time the phone number changes. Whoever writes your correspondence keeps writing it in the tool they already use, marks the spots where the customer's details go, and uploads it.
The details come from the system that already holds them. The names and addresses are not typed in again, because they already exist in your accounts package, your CRM or your order system. Anything typed twice is a chance to be wrong twice.
The output has to be right without anybody proofreading each one. This is the actual test. A system that produces two hundred documents which then need two hundred checks has moved the work, not removed it. Getting to a point where the output can be trusted unread is the whole value, and it is also the hard part.
Why it is harder than it looks
Any developer can make this work for one document on their own machine in an afternoon. Making it work for every template your business will ever upload, without anyone checking the output, is a different job. This is the part worth paying for, and here is what we have actually hit.
A placeholder you typed is not one thing to the computer. When you type a marker into Word to show where the customer's name goes, Word does not necessarily store it as one item. If the spell checker or the grammar checker touched it while you were typing, it gets stored in fragments with invisible markup in the middle. The software filling the document in then does not recognise it, so it does not replace it, and the marker prints on the letter exactly as typed. Two hundred of them, posted.
And a broken one damages a different one. This is why it reads as random rather than as a bug. When a marker is not recognised, the software carries on scanning until it finds the next thing that looks like the end of one, which is somewhere in the middle of a perfectly good marker further down the page. So the placeholder that looks fine is the one that fails, and the one that caused it looks untouched. We had a client report where four fields on a letter never filled in, and none of them was the field that was actually broken.
One character can destroy the document. The sharpest example we have hit: the choice of which characters mark a placeholder is a setting, and it looks entirely cosmetic. Curly brackets seem like the obvious choice, and are the ones most people reach for. It turns out Word uses that same character constantly in its own internal markup, particularly around images. Filling in one letter matched far more than it should have and removed just over 1,800 characters of the real document, including the logo. Word considered its own file perfectly valid. Ours was not, and the only complaint anything made was that the file could not be opened.
The font is not on the server. A more mundane one, and a good illustration of why this needs testing in the place it will actually run. A letter that was reliably one page in Word came out as two pages when generated, with the signature stranded on the second sheet. Nothing failed and no error was raised. The machine producing the documents simply did not have the font the letter was written in, substituted a wider one, and every line grew by a fraction until the page overflowed. It was invisible on any developer's own computer, because their computer had the font.
There is a diagnostic lesson buried in these, and it generalises well beyond documents: the file a client sends you is usually not the file that failed. Opening the broken document to email it causes Word to quietly repair it on the way out, so it arrives working perfectly and reproduces nothing. More than once it has been quicker to reconstruct the fault from a screenshot than to keep asking for the original.
What it will not do
- It will not improve your letter. If the wording is unclear, automating it produces the same unclear wording two hundred times, faster. Worth a read through before rather than after.
- It will not fix bad data. Missing postcodes and half filled address fields become visible immediately, which is genuinely useful, but somebody still has to sort them out. Expect the first run to surface a tidying job nobody knew about.
- Fixed layouts cannot reflow. If the finished document has to match a printed layout exactly, a long address will not push the following paragraph down, it will print over it. Templates need to leave room for the largest realistic case, not the average one.
- It is not a mailing house. Producing the documents is separate from printing, folding and posting them, and the two need to be planned together if the volume is significant.
Where to start
Find the document your business produces most often and count how many go out in a month. That number, multiplied by the honest time each one takes, is the entire business case, and it is usually larger than anyone expects because the work is spread across several people and never appears as a line item.
Start with the highest volume, lowest judgement document you have. Price change notifications, renewal reminders and standard quotes are ideal, because the content is genuinely identical apart from the details. Leave the letters that need real thought alone, at least at first.
We build this into blu::campaign, which generates personalised letters from your own customer data at volume, using your own uploaded Word template. If somebody in your business is currently doing this by hand, get in touch and we will work out what it is costing you.