Central purchasing for multiple locations: the before and after that decides your expansion

Four prices for the same kilo: where this starts

When the same protein cut is paid at 14,200, 15,800, 13,900 and 16,400 pesos per kilo inside one brand, the same week, from the same supplier, the group has no cost policy: it has four personal criteria living under one logo. That 2,500-peso spread, across 900 kilos a month between the four sites, evaporates 2.25 million pesos every month without anyone stealing a cent. Decentralized buying wins on speed —the chef calls, the supplier delivers that afternoon— and loses on price systematically, because the salesperson charges whatever each counterpart tolerates. A central purchasing unit for multiple locations flips that asymmetry: one counterpart, one volume, one price. The central unit wins, and the margin of victory gets measured in food cost points, not in personal rapport with the delivery guy. Roughly 40% of the saving from a central purchasing unit comes from volume discount and 60% from killing emergency buys at retail price.

Where does the saving from central purchasing actually come from?

No supplier sells you that second half because it works against them: internal discipline produces it, through an order calendar and a minimum inventory per site.

Decentralized buying runs to the corner store when onions run out on a Friday at seven, and pays a 30% to 60% premium in an act nobody records because it goes through petty cash. Multiply those episodes by four locations and thirty days. A group with consolidated buying drops 2 to 5 food cost points, while the decentralized one is still arguing whether the supplier honored the 3% he promised. The central unit wins because it controls price, not because it negotiates better. With two locations the central purchasing unit loses, and I hold that position even though I pushed the opposite for years.

The break-even point: I got this wrong for years

My argument was to structure from the second site because earlier is better, until the numbers corrected me: a buyer's salary with payroll charges runs near 4.5 million pesos a month, some 54 million a year, and the real saving on two sites over a 110-million annual spend barely reaches 4 points, that is 4.4 million. The math fails, and what happens in practice is worse: the owner takes the job on at eleven at night and drops it by March. The threshold that does work is THREE locations or 180 million in annual purchasing, whichever arrives first. Below that line, decentralized buying with a shared price master wins outright. The Bogotá group behind those four invoices had the central unit paid off in month eight. The numbers: consolidated annual spend of 1,940 million pesos, opening food cost of 34.8% and 31.2% by the close of the third quarter, meaning 3.6 recovered points worth 69.8 million a year.

Bogotá, four sites, month nine: the case with cash behind it

The buyer position cost 54 million annually with charges, and the price-master license another 7.2 million. Payback in 245 days, with the uncomfortable detail that the first quarter came out negative because two chefs kept buying outside and their petty cash had to be shut down. Without that shutdown no central unit is worth anything: a purchasing system that tolerates exceptions is a decorative price list. The site that resisted hardest ended up with the best variance, which did not surprise me. Through a 2026 that forced two protein and oil renegotiations in twelve months, the decentralized model arrived late to both rounds while the centralized one locked price by quarter. The reason is structural: negotiating site by site means four conversations, four calendars and four levels of urgency, so by the time the last location reacts the supplier has already moved his list twice. With consolidated volume there is one conversation and the argument is arithmetic.

Reaction speed when the market moves twice in one year

The honest counterweight exists and deserves saying: the central unit reacts more slowly to one-off surprises, and a site needing an unusual input for a Saturday event faces more paperwork. Fix it with an exception quota of 5% of monthly spend, authorized and audited. With that valve, the central unit wins on volatility; without it, the system sabotages itself. A site chef spends between 6 and 9 hours a week quoting, receiving salespeople and chasing invoices, time that shows up as zero in any model comparison because it already sits inside payroll. It is not zero. Four sites times 7 average hours make 28 hours a week, around 1,456 a year, close to three quarters of a full position devoted to a task the central unit performs once. That is the trap in decentralized buying: it looks free because its cost hides inside salaries you already pay. Give that time back to the line and the effect shows in things that do bill: recipe specs current, waste under control, the second cook trained.

The hidden cost nobody puts in the comparison: the chef's hours

Central purchasing for multiple locations wins this criterion without argument, and it is the least used case because it appears in no income statement. No central purchasing unit works if each site uses a different recipe spec, and that is the condition most groups skip before hiring the buyer. If site A orders 180-gram tenderloin and site B works it at 200, consolidation is fictional: the supplier quotes two references and you negotiate the volume of one. Standardizing the 40 to 60 core recipes, which in a typical menu concentrate 80% of purchasing, takes six to ten weeks of real work with the chefs. It comes first, not last. One figure gives scale: franchised restaurants in Spain billed 7.23 billion euros in 2024 across 7,967 establishments, according to Tormo Franquicias Consulting 2024, and that model stands on identical specification, not on each kitchen's judgment. Without a common spec, comparing the two models is not even valid.

What to choose given your operating profile?

With three or more locations under one brand, build the central purchasing unit and do it this quarter: consolidation cuts 2 to 5 food cost points and the position pays for itself before month nine.

With two sites, stay decentralized but add a shared price master and a joint quarterly negotiation, which captures around 40% of the benefit at close to zero cost. With locations under different brands or kitchens that share no inputs, the central unit loses: consolidate only common families —oils, cleaning, disposables, dry goods— and leave protein to each site. And past six locations across two cities, the question stops being central or not, and becomes central with its own warehouse or supplier cross-docking. At Masterestaurant, Diego F. Parra always starts with the recipe spec and the 180-million threshold; the org chart comes afterward. Central purchasing does not pay for itself through volume discount, which is what most owners expect.

What actually changes, and what does not?

It pays through price control. In the groups I have guided, negotiated discount delivered roughly 40% of the savings and killing emergency retail-price purchases delivered the rest.

No supplier deck ever shows that second block, because nobody sells it: internal discipline produces it. I got this wrong for years. I pushed the central desk from location two, on the logic that earlier structure beats later structure. The numbers corrected me: with two sites the buyer's salary eats the savings and the owner ends up doing the job at eleven at night. The real break point sits at three locations or roughly 45,000 dollars of monthly purchasing, whichever lands first. Centralizing purchasing does NOT mean centralizing the menu. The most expensive mistake in multi-brand groups is turning the master list into a straitjacket that kills the menu at the site with the best average check. Freeze commodities and packaging; leave 15% of signature inputs free per site.

What actually changes, and what does not — in practice?

The effect on territorial prefeasibility is indirect and brutal. Once you evaluate a new trade area with location intelligence and already know the exact landed cost of your signature plate, the candidate site's financial model stops being an act of faith.

Without a central desk, every territorial evaluation drags a 3 to 5% error margin on input cost.