A chain that buys kitchen exhaust site by site pays for the same engineering forty times: forty specification reviews, forty quotation cycles, forty slightly different units, and a spare-parts shelf nobody can reconcile. A framework supply agreement exists to stop that repetition. It replaces a stack of one-off projects with one master terms document, one technical annex, one price schedule, and a call-off mechanism any regional project manager can operate without reopening negotiation. What follows is the structure that survives contact with a real building programme: the four commercial numbers to settle before the first site, the specification clauses that keep sites comparable, the call-off paperwork that decides who pays when a schedule slips, and the clauses to hold when the supplier pushes back.
Separate the three instruments buyers call a framework
Most multi-site procurement failures begin with vocabulary. Three different instruments all get called “a framework”, and they create completely different obligations.
A project tender covers one site, one scope, one closed price. It is the correct tool for a single restaurant and the wrong tool for a rollout, because every commercial and technical question is answered from scratch at each location. The visible cost is engineering time. The invisible cost is drift, because each site’s specification is only as good as whoever wrote it that month.
A framework agreement fixes master terms, technical specification and unit price schedule for a defined term, typically 24 to 36 months with an optional extension, and leaves quantities to be drawn down as sites reach release. Volume is usually not guaranteed. What is guaranteed is the mechanism: how a site is specified, how it is priced, how long delivery takes, and what happens when something goes wrong.
A panel framework appoints two to four suppliers under the same master terms and runs a mini-competition at each call-off. It costs more administration per site but keeps price tension alive across a long programme. It suits estates where site conditions vary so widely that one supplier cannot cover every format.
Whichever structure is chosen, four documents have to exist, and each has a different reader. The master agreement is read by lawyers and finance. The technical annex is read by the engineer and the installer. The price schedule is read by the project manager who has to defend a number. The call-off order form is read by everyone, because it is the only instrument that creates an obligation for a specific site.
A framework with no call-off order form is a letter of intent with extra pages. If the annex does not state exactly what a call-off order must contain, the first site will be ordered by email, the second will be ordered differently, and the third will be ordered twice.
The four numbers to settle before site one
1. What the unit price is actually per
A price schedule that says “canopy units” is not a price schedule. The basis has to be explicit, because every basis shifts risk to a different party.
| Price basis | What it makes easy | Where it bites later |
|---|---|---|
| Per metre of hood length | Fast comparison, scales with the cookline | Ignores depth, number of filtration stages, and fan duty |
| Per m3/h at a stated duty point | Puts airflow and static pressure on the same line | Meaningless unless the pressure and the measurement point are written down |
| Per cabinet or per complete unit | Clean for standardised boxed products | Hides the premium for non-standard hood lengths and bespoke plenums |
| Per treatment stage plus per fan | Makes the price of the filtration visible | Pushes airflow design back onto the buyer |
Whatever basis is used, the schedule must also state what the price excludes: ductwork, installation labour, electrical connection, commissioning, freight, duties, structural support, and any builder’s work. The exclusion list is where most budget overruns on a rollout actually originate, not in the unit price.
2. Price validity and indexation
Twelve months firm from signature is a normal ask and most suppliers will accept it. The second year is where frameworks quietly become expensive. Three elements have to be agreed in advance: the index, the source of the index, and the ceiling.
Name the index precisely — steel plate, copper, aluminium, ocean freight on the relevant lane, local installation labour — and name the publication that reports it. Then cap the annual movement, typically at three to five percent, and require a 60-day written notice supported by the index printout before any re-rating takes effect. Crucially, state that a re-rating applies only to call-off orders placed after the notice date, never retroactively to orders already accepted.
The common mistake runs in both directions. A contract with no index at all pushes all commodity risk onto the supplier, who prices the risk in upfront and then delivers late when the risk crystallises. A contract with an open re-rating clause pushes the same risk onto the buyer with no ceiling at all. Neither survives a two-year materials cycle.
3. Volume bands and how rebates are reconciled
Tiered pricing is normal; tiered pricing reconciled against a forecast is not. Write the rebate as a percentage of shipped value, reconciled annually against actual call-offs, credited against the next invoice. State clearly whether installation value counts toward the band, because a supplier and a buyer will otherwise read that sentence differently at year end.
Take-or-pay commitments deserve separate treatment. Accept one only if the estate is genuinely committed, and cap the exposure — a common compromise is a commitment at around 60 percent of forecast, with no penalty on unplaced sites. A framework that penalises the buyer for sites the property team failed to deliver is a finance liability dressed as a supply contract.
4. Lead time and what starts the clock
Lead time means nothing without a defined start event. Catalogue units typically run four to six weeks; custom hood lengths, non-standard plenums and cabinet configurations run eight to twelve. Write that the clock starts only when three things exist together: approved drawings, a purchase order, and the deposit. If only two of the three are in place, the clock has not started and nobody can claim a delay.
Two additions make the lead-time clause usable in practice. First, a capacity reservation letter for peak season, stated as a monthly unit ceiling rather than a vague promise of support. Second, an expedite clause that states what air freight costs, who pays it, and at what notice it can be triggered — typically at cost plus handling, buyer-funded, and only for units already released to production.
The technical annex is what keeps forty sites comparable
Duty tiering by site format
The annex should be a small number of tables, not a prose specification. Start with site formats, because a chain’s estate is never one format. Counter-service units with fry batteries and griddles behave nothing like a bakery counter, and neither behaves like an open display cookline in a food hall.
| Site format | Duty tier | Airflow basis | Treatment expectation |
|---|---|---|---|
| Counter-service QSR, fry and griddle led | Heavy | 2,000 to 2,500 m3/h per metre of hood | Grease stage plus odour stage, odour media changed on measured interval |
| Cafe, bakery, sandwich, no open flame | Light | 2,000 m3/h per metre | Grease stage, odour stage optional by lease condition |
| Open display cookline, food hall stall | Heavy to very heavy | 2,500 m3/h per metre, plus capture geometry check | Grease stage plus odour stage, front-of-house discharge constraints apply |
| Solid fuel, charcoal, wood-fired | Very heavy | 4,000 m3/h per burner as a starting point | Treated as a ducted-only case in most jurisdictions; spark arrestance required |
Fixed formats are only half the estate. Sites that move — event kiosks, seasonal terraces, temporary pop-ups during a refit — do not fit a fixed-asset framework at all, and the mobile catering cart range answers them better than a bespoke hood with a removable duct. Building that category into the annex from the start avoids the situation where a regional manager improvises a solution outside the contract and the estate acquires a one-off asset nobody can service.
Airflow basis and the cross-check that catches errors
Two independent sizing methods belong in the annex, because each catches the other’s mistakes. The first is the cookline method: 2,000 m3/h per metre of hood for enclosed work, 2,500 for open display cooking, and approximately 4,000 m3/h per burner for high-output ranges. The second is the capture method: a face velocity at the hood opening of 0.4 to 0.5 metres per second, checked against the actual opening area and the overhang.
The annex should demand that the two methods agree within about 30 percent. When they do not, the geometry is wrong rather than the arithmetic — an overhang that is too shallow, a hood mounted too high, or a side that is open to a cross-draught. Where the building can support a full duct run and the volume is genuinely high, the ducted ESP purifier range covers that end of the programme; where it cannot, the specification has to reach for a different architecture rather than a smaller fan.
The no-drift list: mandatory at every site
This is the part of the annex that earns its keep. Each item is written as a pass-or-fail requirement, not a preference, so that an installer cannot quietly substitute.
Access panels at no more than three metre centres along every duct run, and at every change of direction, sized for cleaning rather than for inspection. Grease-tight joints throughout, with no flexible duct in grease service. Treatment cells and high-voltage power supplies identified by part number, not by description. Odour media stated as a mass in kilograms with a stated contact time, because “carbon filter” describes a shape, not a performance. A measured sound pressure level at one metre. And a documentation pack per site, listed in the annex and not left to negotiation.
Where the specification template has to serve a ductless range hood family across several hundred units, the value of the no-drift list becomes obvious: one cell part number, one power supply part number, one media specification, and a spare-parts pool that stays small enough to be auditable.
The equivalence trap
Sooner or later a supplier will ask for the words “or equivalent” in the technical annex, and the buyer will agree because the sentence sounds reasonable. It is the single clause most likely to fragment an estate. Define equivalence on measurable attributes instead: airflow at a stated static pressure, cell efficiency established by a type test rather than a marketing figure, odour media mass, sound power level, and the physical dimensions of service access. Then require written proof against those attributes before a substitution is accepted. Without that discipline, site twelve receives a different cell, the spares pool splits in two, and the framework has quietly become a series of projects again.
Call-off mechanics: the paperwork that decides who pays
The call-off order is the only document that binds a specific site, so its required contents belong in the master agreement. At minimum it restates the site address and access constraints, the duty point, the drawing revision letter, the delivery term and who unloads, the commissioning window, and the trigger for acceptance and retention release.
Drawing approval needs a named approver on each side and a stated period, commonly ten working days. Deemed approval on silence should exist only if the contract says so explicitly — silence is not consent by default, and a supplier who proceeds on silence owns the rework. Variation orders must be in writing, issued only by named individuals, priced against the schedule first and against an agreed day-rate table second. Any variation executed without a written order is a gift to the supplier and a problem for the auditor.
Four events account for most slipped schedules in a rollout, and all four are contractual rather than technical: late drawing approval on the buyer’s side, a site that is not ready when the units arrive, a variation performed before it is priced, and a duplicate purchase order raised by a second region. The fourth is worth naming in the contract. Every framework eventually meets a site manager with a purchase card, so state who may place orders and state that orders from anyone else create no obligation.
Spare parts and service continuity: the part chains forget
Spare parts are usually handled as a courtesy until the first failure, at which point they become the most expensive line in the estate. Write them into the framework as an obligation with numbers.
Begin with a critical spares list: treatment cells, collector plates, ceramic insulators, sealing strips, high-voltage power supplies, baffle filters, odour media, and fan bearings and belts. Then choose a stocking model. Consignment stock, where the supplier owns inventory at a regional hub and invoices on use, suits slow-moving high-value items such as cells and power supplies. A buyer-owned pool suits fast movers such as seals and filters. Most estates end up with both.
Attach an availability period — five years from the last unit shipped is a normal ask — a last-time-buy notice of at least twelve months, and a form-fit-function replacement rule with written equivalence proof, so a discontinued cell does not silently redefine the installed base.
The pool can be sized from the maintenance interval rather than from instinct. In heavy-duty kitchens a cell wash cycle runs every two to four weeks and odour media changes every three to six months; in light-duty kitchens the cell interval stretches to around six weeks and media lasts longer. Multiply the interval by the unit count and the number of sites, and the required pool becomes arithmetic instead of guesswork. On a forty-site estate, holding roughly one spare cell per five installed units is a reasonable starting point. A single emergency air-freight shipment of one cell frequently costs more than the entire pool.
Finally, define the return process: swap-before-return for cells and power supplies, a stated turnaround in working days, freight paid by the supplier inside warranty and by the buyer outside it, and an inspection report issued for every returned item. Without the report, warranty disputes become arguments about opinion.
Warranty architecture across a rolling estate
A forty-site rollout never has one warranty. It has forty start dates, which is why the contract has to convert them into cohorts. Pick a single start event and write it: delivery, commissioning sign-off, or handover. Commissioning sign-off is the fairest, provided it carries a backstop such as thirty days after delivery, so that a buyer who delays commissioning does not extend the warranty indefinitely.
Then group sites into quarterly cohorts. Forty sites then produce twelve expiry dates rather than forty, and renewal, inspection and spares planning all become one exercise per quarter instead of forty small ones.
The wear-item exclusion list must be explicit rather than implied, with a priced replacement schedule attached to it. Treatment cells, collector plates, sealing strips, ceramic insulators and odour media are consumable in service; saying so in the contract prevents a warranty claim becoming a negotiation about semantics.
One further deliverable belongs here: a serial number register, updated within ten days of each commissioning, recording site, unit model, unit serial, cell serial, power supply serial, commissioning date and measured airflow at handover. This single register is what makes warranty administration, spares forecasting and later resale of the equipment possible at all.
The per-site compliance file
Inspectors and insurers ask questions on site, not in a meeting, so the framework should define a file that exists per site and can be retrieved within five working days. Its contents are predictable: the declaration of conformity for each model installed, the type test report supporting the stated treatment efficiency, the wiring diagram, the operation and maintenance manual in the language of the site’s jurisdiction, and a spare parts list with part numbers.
Assign custody. The buyer holds the master copy; the supplier refreshes it whenever a revision is issued. A compliance file that lives only in the supplier’s email archive is not a compliance file, and on a multi-site estate the retrieval clause is worth more than the discount negotiated on the price schedule.
Exit, transfer and the stranded-estate test
Every framework should be written with its own ending in view. Termination for convenience at 90 days notice is normal for the buyer; termination for cause needs a cure period, commonly 30 days, so that a genuine failure is not litigated as a misunderstanding.
What transfers on exit matters more than the notice period. Drawings, the serial register, spare stock at book value, unexpired warranty obligations on installed units, and any open commissioning activity all have to be listed. Where controls or monitoring are proprietary, either require an open protocol or place the configuration documentation in escrow.
Apply the stranded-estate test before signing: if the supplier disappeared tomorrow, could the estate still be maintained for five years? If the honest answer is no, the document is a dependency rather than a contract, and no price advantage compensates for it.
A worked comparison: forty sites over three years
The arithmetic below is a planning illustration rather than a quotation, but the ratios are what matter in a board paper.
| Line | Project by project | Under a framework |
|---|---|---|
| Specification and drawing effort | 40 to 60 hours per site, so 1,600 to 2,400 hours | Two templates plus 8 to 12 hours per site, so 320 to 480 hours |
| Price variance across the estate | Roughly 8 to 15 percent site to site for the same duty | One schedule, indexed and capped |
| Treatment cell variants in service | Four to six, depending on who specified each site | One, protected by the equivalence clause |
| Warranty expiry dates | Forty | Twelve quarterly cohorts |
| Emergency freight events | Commonly six or more per forty sites | Reduced by the consignment pool, not eliminated |
| Audit trail | Reconstructed from email when needed | Serial register plus per-site compliance file |
The saving that survives scrutiny is rarely the unit price; it is the engineering hours, the variant reduction and the expedite freight that never appear in a quotation.
The negotiation table: what they ask for, what to trade, what to hold
| Clause | Supplier’s opening position | What you can trade | The line to hold |
|---|---|---|---|
| Price validity | 90 days | A longer term or a larger forecast | Twelve months firm from signature |
| Indexation | Open re-rating on notice | A named published index | A cap plus 60 days notice plus no retroactivity |
| Volume commitment | Take-or-pay at full forecast | Up to about 60 percent of forecast | No penalty on sites the estate does not deliver |
| Lead time | Eight to ten weeks from order | Phased release of sites | Clock starts at approved drawings plus order plus deposit |
| Warranty start | Date of shipment | Delivery date with a 30-day backstop | Commissioning sign-off as the default |
| Wear items | Silent exclusion | An explicit exclusion list | A priced replacement schedule attached to it |
| Spare parts | Available on request | Consignment stock at a regional hub | Five-year availability plus twelve-month last-time-buy notice |
| Equivalence | Or equivalent | Substitution with written proof | Equivalence measured against named attributes |
| Service levels | No remedy offered | Service credits rather than penalties | Two percent of call-off value per week, capped at ten percent |
| Exclusivity | Exclusive for the full term | Exclusivity tied to performance | Exit after two consecutive service-level failures |
| Documentation | Standard pack, on request | Electronic delivery | Per-site file within ten days of commissioning |
A credit against the next invoice is enforceable and self-executing, where a penalty across borders is neither.
Red flags in a framework proposal
Ten patterns are worth checking before signature. A volume commitment priced off a forecast with uncapped escalation. An exclusive term longer than the price validity. Specification language that says “or equivalent” with no measurable attributes behind it. A warranty counted from shipment. Spare parts available “on request” with no availability period. A rebate calculated on the forecast rather than on shipped value. No named individuals authorised to issue variations. No obligation to maintain a serial register. Service levels with remedies but no measurement method. And an exit clause that says nothing about drawings, stock or unexpired warranty obligations.
Any one of these is negotiable on its own. Three together mean the document was written to be signed quickly, not operated for three years.
The one-page checklist to take into the meeting
Price basis named, with static pressure and measurement point stated. Exclusions listed explicitly. Twelve months firm, then a capped named index with notice. Rebate on shipped value, not forecast. Lead time measured from approved drawings plus order plus deposit. Duty tiers mapped to site formats. Two sizing methods required to agree within 30 percent. No-drift list written as pass or fail. Equivalence defined by attribute. Call-off order contents specified. Named variation authorities. Critical spares list with an availability period. Warranty start event fixed and cohorts aligned quarterly. Serial register as a deliverable. Per-site compliance file with a retrieval period. Exit terms covering drawings, stock and open obligations.
The conclusion an engineering team usually reaches after the first framework is that the document works less like a discount mechanism and more like a specification with commercial teeth. The unit price still matters, but what actually protects the estate is the annex: the duty tiers that stop a light-duty cafe being fitted with a heavy-duty canopy, the no-drift list that keeps one cell part number in service across every site, the interval-based spares pool that turns an emergency into a scheduled delivery, and the serial register that makes all of it visible a year later. That is the version of a framework worth negotiating, and it is the version a supplier can actually execute against.
