If somebody in your company regularly asks several suppliers for prices, this scene will be familiar. The emails go out on a Monday. By Tuesday three have replied. On Thursday someone remembers that five are missing and writes again. On Friday the last one arrives, as a PDF you have to open to pull the number out of, and only then can the comparison be built.
Five days. And the part that genuinely needs judgement — deciding who to buy from — takes ten minutes on the Friday.
Where the time actually goes
This is the most common mistake when scoping the automation: believing the work is writing the emails. It is not. The emails take a short while, and you have been able to send them in bulk for twenty years.
The time sits in three places, and none of them is drafting:
- Waiting without knowing what for. Nobody has a list of who has replied and who has not. It is in the inbox, mixed in with everything else.
- Chasing. Somebody has to remember, check and write again. And because it is tedious, it happens late or not at all.
- Sorting. Replies arrive in incompatible formats. One puts the price in the body of the email. One sends a PDF. One comes back with a counter-offer and its own payment terms. Somebody has to translate all of that into comparable rows.
Ask yourself how long it would take you right now, without opening your inbox, to say which suppliers from the last round still have not replied. If the answer is not immediate, there is your problem.
The five steps
1. Measure before automating
Count for a month: how many rounds you run, how many suppliers per round, and how many days pass between the first request and having the comparison.
That number decides whether it is worth it. If you run two rounds a month across three suppliers, do not automate anything: you will never earn it back. If you send forty requests a month and each one eats two hours, that is eighty hours a month and close to a thousand a year, and the conversation changes completely.
2. Standardise what you ask for
Before touching any tool, define the minimum details every request carries: which material or service, quantity, unit, lead time, terms.
This looks like bureaucracy and it is the step that saves the most back and forth. A good share of those round-trip emails are not caused by a slow supplier: they are caused by a request that went out incomplete, so the supplier had to ask. Every question of theirs costs you two days.
3. Send it and keep a record
Sending is the easy part. What matters is that there is a record of who was asked for what and when, outside the inbox.
That record is what makes the next step possible. Without it, knowing who is missing means going through the inbox by hand, which is exactly what we are trying to stop doing.
4. Chasing, which is where the money is
This is the part almost nobody automates and the one that gives back the most time.
The agent checks every day who has not responded and writes to them again. Without anyone having to remember, and without it depending on whether that week happened to be quiet. One reminder on day two and another on day four move the response rate more than anything else, and it is work a person does badly precisely because it is repetitive and dull.
5. Consolidate messy replies
This is where a useful automation separates from one that creates more work than it removes.
A system that only understands replies in one particular format forces somebody to translate the rest by hand. And since suppliers answer however they please, that "rest" is most of them.
What works is the other way round: accept that replies arrive messy — a loose price, a PDF, a counter-offer — and let the agent turn them into rows of the same table, with price, lead time, payment terms and variance against the reference price.
What NOT to automate
The award.
Choosing a supplier commits money and sometimes a commercial relationship built over years. That decision has to reach a person, with the comparison assembled and the variances flagged, and be made in two minutes.
An agent that awards on its own is a black box, and a black box cannot be corrected when it gets something wrong. The goal is not to take the person out of the loop: it is that when they arrive, they do not have to start from zero.
A case with numbers
At KONBUD we built exactly this for construction firms. The agent reads the bill of quantities, identifies each material, writes to the suppliers, chases the ones who do not reply and assembles the comparison.
The pilot numbers, on a live project:
- 430 materials handled, across 13 trade categories
- 552 quote requests sent
- 460 replies processed and normalised
- One round across 12 suppliers: from 5 days to 4 hours
And the thing that surprised us most while building it: the hard part was not the intelligence. It was the dirty work — reading badly formatted emails, handling retries, reconciling units that did not match. That is where the value was, and it is the part no polished demo ever shows.
Swap "suppliers" for "hotels" and you have an events agency checking availability. Swap it for "repair shops" and you have an insurer. The pattern is always the same: requests that go out incomplete, third parties who take their time, and one person in the middle acting as glue.
Where to start tomorrow
Without buying anything and without calling anyone:
- Open the last round of requests you ran and count the real days between the first email and the comparison.
- Look at how many of those emails were reminders.
- Look at how many replies you had to rewrite by hand before you could compare them.
With those three numbers you already know whether you have a process problem or just a bad month. And if you have the problem, you know exactly where it is.