OIA · Dry Run - controlled demonstration prototype · Worked example Hartwell M&E Group · £45m turnover · 180 staff · Maturity 2.4 / 5 · 4-week diagnostic · Bleed Rate £1.87m (4.2% of turnover)
ILLUSTRATIVE · Operational Intelligence Audit™ · Deliverable 04 of 07

Dry RunCONTROLLED DEMONSTRATION PROTOTYPE

A practical demonstration of AI-assisted tender processing using Hartwell M&E Group's own tender data. Track A — Tender Intelligence Prototype.

Prepared for
Hartwell M&E Group
Prepared by
Built AI Ltd
Date
May 2026
Dry Run™ — Controlled Demonstration Prototype
← Audit PackDeliverable 04 of 07BAI-HME-AUD-004
Section 01

What This Prototype Demonstrates

This is not a full software build or system integration. It is a controlled demonstration showing how one workflow could operate using Hartwell's own tender data.

A

Track A — Tender Intelligence Prototype

Selected based on Opportunity Roadmap ranking: highest data readiness (5.0), fastest speed to benefit (5.0), lowest operational risk (5.0)
InputOne real tender pack from Hartwell's recent bid pipeline
TenderMechanical Services — Braeburn Court, Phase 2 Refurbishment
ClientRedacted (NHS Trust — healthcare estate)
Value£780,000 (estimated)
Pack size142 pages across 6 documents (spec, schedule, drawings list, PQQ, T&Cs, pricing schedule)
Processing time47 minutes (AI-assisted) vs 11.5 hours (manual estimate)
Why Track A was selected. The Opportunity Roadmap (Deliverable 03) ranked tender production second overall but first for data readiness and speed. Tender packs are self-contained document sets requiring no system integration. The prototype can be produced during the audit period using a real pack from Hartwell's pipeline, giving the leadership team a tangible demonstration of output quality and time saving before committing to Phase 1.
01
Tender pack uploaded
0 min
02
Document classification & parsing
3 min
03
Scope extraction & structuring
8 min
04
Assumptions & exclusions draft
12 min
05
Clarifications & risk register
14 min
06
QS review & validation
10 min
Section 02

Before & After Comparison

How the same tender pack was processed under Hartwell's current workflow versus the AI-assisted prototype workflow.

Current process

Manual tender production

Scope review: Estimating Lead reads all 142 pages manually, highlights scope items, notes queries in a Word document. 4.5 hours
Assumptions: Drafted from scratch. No standard library. 3 internal review cycles before submission. 2.5 hours
Clarifications: Notes scattered across email, marked-up PDFs and notebook. Compiled into a single document at the end. 1.5 hours
Risk register: Not produced for most tenders. Risks discussed verbally at bid review. Not documented
Compliance check: Manual checklist against submission requirements. PQQ responses populated individually. 3.0 hours
Total estimated time
11.5 hours
Prototype workflow

AI-assisted tender production

Scope extraction: AI parses all documents, identifies and structures 68 scope items across 12 work packages. QS reviews structured output. 8 min + 10 min review
Assumptions: Generated from scope analysis + Hartwell's standard assumptions library. 14 assumptions drafted, 3 flagged for QS input. 4 min + 8 min review
Clarifications: 9 clarification queries identified from spec gaps, contradictions and missing information. Pre-formatted for client submission. 6 min + 5 min review
Risk register: 11 risks identified and scored across commercial, technical, programme and compliance categories. 5 min + 6 min review
Compliance matrix: Submission requirements mapped automatically. PQQ responses pre-populated from Hartwell's existing data. 3 min + 5 min review
Total time (AI + human review)
47 minutes

Performance comparison

Total processing time
11.5 hrs
47 min
93% reduction
Scope items identified
51
68
+33% more items
Assumptions drafted
8
14
+75% coverage
Clarifications raised
4
9
+125% queries
Bid risk register
None
11 risks
New output
Human review retained
100%
100%
Unchanged
Critical point. The AI workflow does not replace the QS or estimating team. It produces a structured first draft in minutes that the team then reviews, validates and refines. The 47-minute total includes 34 minutes of human review. Every output is reviewed and approved before use. The time saving comes from eliminating manual reading, extraction and formatting — not from removing human judgement.
Section 03

Prototype Outputs

The following outputs were produced from the Braeburn Court tender pack. All content is redacted but structurally representative of the prototype produced during the audit.

Output 01

Structured Scope Extraction

68 scope items identified and structured across 12 work packages
8 min AI · 10 min review
Extract — Work Package 03: Heating Distribution
WP03-01Strip out and removal of existing LTHW pipework to Wards 4, 5 and 6 including all associated valves, fittings and supports. Allow for asbestos survey findings — refer Clarification CL-03.
WP03-02Supply and install new copper LTHW flow and return pipework to serve 42 no. LST radiators. Sizes from 15mm to 54mm per specification Section 5.3. Route as indicated on Drawing M-201 Rev C.
WP03-03Supply and install thermostatic radiator valves (TRVs) and lockshield valves to all new radiators. TRVs to be tamper-resistant type in patient-facing areas per NHS Estates guidance.
WP03-04Commission and balance complete heating distribution system including flow rate verification and TRV calibration. Provide commissioning certificates per CIBSE Code W.
WP03-05Pressure test all new pipework installations to 1.5× working pressure. Provide photographic evidence of test gauge readings and hold-time records. Minimum 2-hour hold.
QS review note: The AI identified 17 more scope items than the manual extraction (68 vs 51). The additional items were primarily commissioning requirements, testing obligations and evidence deliverables embedded within specification clauses that the estimating team typically extracts later in the pricing phase rather than during initial scope review.
Output 02

Assumptions & Exclusions Schedule

14 assumptions generated · 3 flagged for QS review · 6 exclusions structured
4 min AI · 8 min review
RefAssumptionSourceStatus
A-01Works will be carried out during normal working hours (07:00–17:00 Mon–Fri). No allowance for out-of-hours, weekend or bank holiday working.Spec 1.4.2Accepted
A-02The existing building structure is suitable for supporting new pipework and equipment loadings without additional structural support or enabling works.StandardAccepted
A-03Asbestos survey results will be provided by the client prior to commencement. Removal or encapsulation of any identified ACMs is excluded from this tender.Spec 2.1.8Flagged — QS to confirm exclusion wording aligns with T&Cs clause 14.3
A-04Ward decant programme and patient relocation will be managed by the Trust. Contractor access will be to cleared wards only. No allowance for working around occupied areas.Spec 1.6.1Flagged — programme risk. Decant schedule not included in tender pack
A-05Existing electrical supply capacity is sufficient for new mechanical plant. No allowance for incoming supply upgrades or transformer works.StandardAccepted
A-06All builders work openings, fire stopping, making good and decoration to disturbed areas will be carried out by others under a separate contract unless specifically identified in the scope.Spec 1.5.3Accepted
A-07Pricing is based on Drawing M-201 Rev C and Specification Issue 02 dated March 2026. Any subsequent revisions may require re-pricing.Drawing refAccepted
A-08BMS integration is limited to hardwired point-to-point connections to the existing BMS head end. No allowance for BACnet/IP networking, new outstations or BMS graphics reprogramming.Spec 7.2Flagged — spec ambiguous on integration scope. Refer CL-07
Showing 8 of 14 assumptions · 3 of 6 exclusions shown in full document
Output 03

RFI & Clarification Log

9 clarification queries identified from specification gaps, contradictions and missing information
6 min AI · 5 min review
RefClarification querySource referencePriority
CL-01The specification references “low surface temperature radiators to NHS Estates HTM 05-02” but HTM 05-02 was superseded by the 2023 revision. Please confirm which edition applies and whether maximum surface temperature limit is 43°C or 41°C.Spec 5.3.4Pricing impact
CL-02Drawing M-201 Rev C shows 42 radiator locations but the schedule of rates lists 38 positions. Please confirm the correct quantity.Dwg M-201 / SoRPricing impact
CL-03Section 2.1.8 states “contractor to allow for asbestos management” but no R&D survey results are included in the tender pack. Please confirm whether survey results will be issued pre-tender or whether a provisional sum should be included.Spec 2.1.8Pricing impact
CL-04No decant or ward availability programme is included. Please confirm the planned ward release sequence and estimated access dates per phase.Spec 1.6.1Programme
CL-05The T&Cs reference liquidated damages at clause 9.4 but the damages rate and sectional completion milestones are not defined in the tender documents. Please confirm.T&Cs 9.4Commercial
Showing 5 of 9 clarifications · Full log included in prototype pack
Output 04

Bid Risk Register

11 risks identified and scored · 4 commercial · 3 technical · 2 programme · 2 compliance
5 min AI · 6 min review
RefCategoryRisk descriptionLIScoreMitigation
R-01CommercialAsbestos exposure — no survey data provided. Risk of programme delay, additional cost and CDM notification if ACMs encountered during strip-out.4520Clarification CL-03. Provisional sum. Contractual exclusion in A-03.
R-02CommercialDrawings vs schedule quantity discrepancy (42 vs 38 radiators). Risk of under-pricing or post-award dispute on final account.5315Clarification CL-02. Price to drawing (42 nr). Note assumption.
R-03ProgrammeNo ward decant schedule provided. Risk of access delays, phasing constraints and abortive attendance if wards not released on time.4416Clarification CL-04. Programme assumption A-04. Provisional prelims.
R-04TechnicalBMS integration scope ambiguous. Risk of under-pricing if new outstations, graphics or IP networking are required beyond hardwired connections.3412Clarification CL-07. Assumption A-08 limits scope to hardwired connections.
Showing 4 of 11 risks · Full register in prototype pack
What the QS said. The Estimating Lead reviewed the prototype outputs and confirmed: “The scope extraction picked up commissioning and testing items we would normally catch at pricing stage, not at initial review. The clarifications on the radiator quantity and the HTM reference are exactly the sort of things that cause problems on final account if you miss them at tender stage. We would have caught the quantity issue but probably not the HTM supersession.”
Section 04

What This Proves

The prototype demonstrates five specific capabilities that directly address the tender admin burden and bid quality issues identified in Deliverables 01 and 02.

Time can be recovered

The prototype processed 142 pages in 47 minutes (including human review) versus the 11.5-hour manual estimate. Scaled across 67 annual tenders, this represents a potential saving of 730+ hours per year in the estimating team.

Quality improves, not just speed

The AI-assisted process identified 33% more scope items, 75% more assumptions and 125% more clarifications than the manual process. It also produced a bid risk register — an output that Hartwell does not currently generate at tender stage.

Human review is retained

Every output was reviewed and validated by the Estimating Lead before acceptance. The workflow produces structured first drafts — the QS makes the final judgement on every item. No output is sent to a client without human review and approval.

Capacity can increase

If the 38 hours of admin per tender is reduced by 30–40%, the estimating team could handle 15–20 additional tenders per year without additional headcount. The 11 tenders declined for capacity reasons in the last 12 months represent £836k in potential revenue.

93%
Time reduction on this tender

This is a single tender. The compounding benefit comes from scale.

67 tenders per year × time saving per tender = capacity equivalent of 1.5–2.0 additional estimating staff. Combined with improved scope coverage and risk identification, the workflow creates both commercial efficiency and bid quality improvement simultaneously.

Section 05

Prototype Controls

How the prototype maintained governance, data handling and quality controls throughout the demonstration.

ControlHow it was applied in the prototypeStatus
Data handlingTender pack documents were processed within Built AI's controlled environment. No client data was stored permanently, shared with third parties or used for model training. Documents were deleted after prototype completion.Applied
Human review gateAll five outputs were reviewed and validated by the Estimating Lead before acceptance. Three assumptions were modified and one clarification was reworded during review.Applied
Output labellingEvery output carried a “Draft — Subject to QS review” watermark and version identifier. Final versions were labelled as “QS Reviewed” with reviewer initials and timestamp.Applied
Source traceabilityEach scope item, assumption and clarification includes a reference to the source document, section or drawing from which it was extracted. Allows the QS to verify any item against the original tender pack.Applied
No client submissionPrototype outputs were produced for internal demonstration only. No AI-generated content was submitted to the client or included in a live tender response.Applied
Audit trailProcessing log records the input documents, AI workflow steps, review actions, modifications and final approval. Available as a separate evidence file.Applied
Section 06

Methodology & Linkage

How the prototype was produced and how it connects to the other audit deliverables.

ElementDetail
Input documents6 documents, 142 pages: M&E specification (78 pages), pricing schedule (18 pages), drawings list with 12 referenced drawings (8 pages), PQQ (14 pages), T&Cs (16 pages), tender instruction letter (8 pages)
ProcessingDocuments parsed, classified and structured using Built AI's tender intelligence workflow. Scope items, assumptions, exclusions, clarifications and risks extracted and cross-referenced automatically.
ReviewEstimating Lead reviewed all outputs during a 45-minute working session on 16 May 2026. Three assumptions modified, one clarification reworded, one risk score adjusted. All other items accepted without change.
Workflow packBidroom in a Box™ — tender intelligence workflow. This is the workflow recommended for Phase 1 implementation alongside Commercial Control Kit™.
Up next
GR · Guard Rail
← Or back to the Operational Intelligence Audit page