Case Study: Taking Quoting Off the Owner's Phone at a Singapore SME

A 14-person air-conditioning servicing and light M&E retrofit firm in Singapore was issuing every single quotation through one phone — the owner's. Eleven weeks after we started, 83% of quotes went out without the owner seeing them, and median turnaround fell from four days to under one. The change that did it was not software. It was writing down the three things the owner had been carrying in their head: what each line item costs, what the rules are for discounting and site conditions, and who is allowed to decide what. Tooling came last, and it was deliberately boring.

One disclosure before the detail: this is a composite drawn from several engagements, with figures taken from one firm's own quote log. The sector and headcount are real; the company is not named, and a few specifics are blended so the client stays unidentifiable.

What did the bottleneck actually look like?

It did not look like a crisis. It looked like a busy owner. Enquiries arrived by WhatsApp, phone and the website form. Site staff sent photos and voice notes. The owner read everything, decided the price, typed the quote into the accounting system on a laptop at night, and sent the PDF. Roughly 60 quotes a month, every one of them through the same person.

The symptoms were the ordinary ones. Quotes went out slowly, and the slow ones lost to whoever replied first. Nobody else could answer a customer asking "what would that cost, roughly?" without promising a call back. When the owner took nine days' leave in the prior year, the quote log showed four issued in that window and a visible dip in jobs starting three weeks later. And the pricing itself was invisible — not secret, just never written down, so it could not be checked, taught or improved.

Why didn't hiring a quotations executive fix it?

Because they had already tried. A coordinator had been hired 18 months earlier specifically to take over quoting, and had quietly stopped doing it within two months. The reason was structural, not personal: the new hire had no priced catalogue to work from and no authority to commit to a number, so every quote still ended with "let me check with boss." That is not delegation. That is adding a step.

This is the standard failure mode. A person-shaped bottleneck does not become a process just because you point a second person at it. Without the pricing logic and the decision rights written down, you get the same bottleneck with a longer queue in front of it.

What had to be written down before any tool was chosen?

We started in a spreadsheet, with three years of issued quotes exported from the accounting system — about 4,100 quote lines. Two patterns emerged within a day:

• Those 4,100 lines collapsed into 212 distinct items once wording variants were merged ("R410a top-up", "gas top up R410", "refrigerant topup" were one item typed three ways).
• 31 items accounted for 78% of quoted value. Those 31 got proper costed rates — material, labour hours, and a standard margin. The long tail was left as custom-priced, which is honest and takes nothing away from the result.

That catalogue was the whole project, really. Everything after it was assembly. It also had a side benefit the client did not pay for: a clean, coded item list with consistent descriptions and tax treatment is the same master data an e-invoicing rollout needs, so their eventual InvoiceNow onboarding started from a tidy catalogue instead of a cleanup project.

How do you set decision rights an owner will actually honour?

Thresholds, written on one page, with the escalation triggers named explicitly. Theirs came out as:

1. Coordinator issues any quote built entirely from catalogue items, up to S$25,000, with discount of up to 8%, on standard 30-day terms.
2. Owner approval required for: anything above S$25,000, any discount beyond 8%, any payment term past 30 days, any first-time commercial building management, and anything involving work at height or hot works.
3. Site survey mandatory before quoting on any retrofit, no exceptions — this was the rule the owner had been applying intuitively and had never said out loud.

Two details made this stick. First, the owner signed the page, which turned it from a suggestion into a standard. Second, we agreed on a weekly 20-minute review of every quote issued under delegated authority — not approval, review after the fact. That gave the owner a way to see the pattern without being in the path, and the thresholds were adjusted twice in the first quarter as confidence grew.

What changed in the system — and what deliberately didn't?

No new CRM. The catalogue went into the item master of the accounting package they already paid for, one quote template replaced four stylistic variants, and enquiries from all three channels were routed into a single shared inbox with a named owner per enquiry. Total new recurring software spend: nothing.

We did not build an approval workflow, integrate WhatsApp, or buy a quoting add-on. Every one of those was proposed and deferred, because a tool bought before the pricing logic exists just encodes the chaos faster. The constraint here was knowledge, not software.

Where did AI help, and where was it kept out?

AI drafted scope-of-work wording from the site coordinator's notes — turning five bullet points and a photo caption into two clean paragraphs of customer-facing text, which a human then checked. That saved real minutes on every quote.

It was kept out of pricing entirely. A generative model will produce a confident number with no costing behind it, and in a business where margin lives in labour hours, that is the one place you want a lookup rather than a guess. Their written AI use policy also barred customer names, addresses and site photos containing identifiable details from consumer chatbot accounts — the scope-drafting step runs on a business-tier account with an explicit no-training setting, which is the difference between a controlled tool and PDPA exposure.

What did the first 90 days produce?

Measured from their own quote log, before-and-after on the same seasonal quarter:

• Quotes issued without owner involvement: 0% to 83%.
• Median time from enquiry to quote: 4 days to 0.8 days.
• Owner time recovered: roughly 9 hours a week, self-reported.
• Win rate: 31% to 37%. We flag this one as suggestive, not proven — one quarter, one firm, and a hundred other variables. The turnaround figure is the one we stand behind.

If you want to start this next week, do it in this order: export two or three years of quotes, collapse them into a distinct item list, cost the items that make up three-quarters of your value, write one page of thresholds and sign it, then pick a tool — and only then. Most owners reach for step five first, which is why most of these projects fail.

Frequently asked questions

How long does this take for a company our size?

For a firm under 20 staff with a few years of quote history, budget eight to twelve weeks of calendar time and perhaps 25 to 35 hours of actual work, most of it the owner's. The catalogue build is the long pole, and it cannot be fully delegated — only the owner knows why a price was what it was.

What if our pricing genuinely depends on judgement?

Then write the judgement down as ranges and triggers rather than fixed numbers. "Access above three metres adds 15–25% to labour, and anything beyond that comes to me" is a rule a coordinator can use. Aim to systematise the 70% that repeats and keep the 30% that is genuinely bespoke on the owner's desk — a 70/30 split removes the bottleneck without pretending the hard cases are easy.

Won't a CRM or quoting tool do all this for us?

No. A CRM stores and routes quotes; it does not know what your items cost or who may approve a discount. Install one before you have a priced catalogue and written thresholds, and you will have paid a monthly fee to move the same bottleneck into a nicer interface. Data and decision rights first, software second.