Build a UK Tender & Grant Finder Copilot in 10 Days

Missed opportunities cost real money. Most SMEs and charities rely on scattered email alerts, spreadsheets, or busy inboxes to track public tenders and grants. In this case-study walkthrough, we show how a 20‑person London services firm stood up a lightweight “Tender & Grant Finder Copilot” in 10 days. The copilot watches trusted UK sources, scores fit, summarises the essentials, and nudges the right people before deadlines hit.

We’ll keep it practical: what to watch, how to score, the guardrails that matter, and the KPIs board members will actually read. Where rules and platforms changed in 2025, we call that out clearly.

The sources that matter (UK‑specific)

• Find a Tender Service (FTS): the UK portal for high‑value public procurement. From 24 February 2025, FTS also carries new above and below‑threshold notices (except below threshold in Scotland). You can sign in, save searches and get alerts. gov.uk
• Contracts Finder: for lower‑value opportunities in England and non‑devolved territories, generally over £12,000 including VAT; accounts let you save searches and get email updates. gov.uk
• Find a Grant: the UK government’s single service to search government grants; departments and arm’s‑length bodies are required to advertise eligible grants here, and you can manage notifications. find-government-grants.service.gov.uk
• Open grants data (GrantNav by 360Giving): a free search engine for UK grant awards and publishers using an open data standard; useful for mapping funders and typical award sizes. 360giving.org

What we built (in plain English)

A small internal “copilot” that:

• Monitors the four sources above via saved searches and email alerts; ingests each new notice into a simple feed.
• Scores fit using a transparent rubric (eligibility, geography, value band, deadlines, strategic keywords).
• Generates a 6‑sentence, plain‑English summary with a link to the official notice, deadline, and a go/no‑go suggestion.
• Posts high‑scoring items into Teams/Slack and adds deadlines to a shared calendar; creates a lightweight Kanban card for actions.
• Tracks outcomes so we can learn which alerts actually convert.

The 10‑day build: a simple, low‑drama plan

Day 1 — Define “what good looks like”

Agree a one‑page brief: your service lines, target geographies, the value band you can credibly deliver, and any bright‑lines (for example, “no projects under £20k” or “must allow consortium bids”). Write a 10‑word mission for the copilot: “Surface 10 relevant, winnable opportunities a week with zero noise.”

Day 2 — Create the intake form

Capture the rules your copilot will use to triage: eligibility statements, keywords and their synonyms, default red‑flags (e.g., “security clearance needed”), and who should be notified for each service line. Store them in a shared document so anyone can edit without calling IT.

Day 3 — Design the user journey

Keep it obvious. One list of new items each morning; a “High fit” badge for scores over 80; a two‑click way to mark interest or dismiss. Push urgent items to Teams/Slack with a “Claim owner” button and add an iCal deadline automatically. Use a weekly digest to avoid alert fatigue.

Day 4 — Connect to the data (no scraping first)

• Set up official accounts and saved searches on FTS, Contracts Finder and Find a Grant; enable email notifications to a shared inbox the copilot reads. find-tender.service.gov.uk
• Start with broad queries then narrow: sector, CPV keywords, location filters, and synonyms you listed on Day 2.
• Add GrantNav (360Giving) for awareness and prospecting—use it to learn which funders back work like yours. 360giving.org

Day 5 — Build a transparent scoring rubric

Weighting that worked well for the case study:

• Eligibility match (must‑have): 0 or 40 points.
• Service line fit (should‑have): up to 25 points.
• Geography and delivery model: up to 10 points.
• Value band vs capacity: up to 10 points.
• Deadline feasibility (10+ working days): up to 10 points.
• Strategic priorities (e.g., aligns to growth area): up to 5 points.

Anything under 50 is auto‑dismissed with a reason. Above 80 posts to Teams/Slack immediately.

Day 6 — Summaries people actually read

Use a standard, six‑line structure: who’s buying or funding; what is being sought; who is eligible; likely value; deadline and how to apply; why we should care (based on your strategy). Always include the “View on GOV.UK/FTS” link; no rewriting or paraphrasing of legal terms—your team must click through for the authoritative source. gov.uk

Day 7 — Shadow‑mode and calibration

Run the copilot in parallel with your existing manual checks for five working days. Each afternoon, compare its “top 10” against what humans spotted. Tune synonyms, thresholds, and weights. Keep a list of false positives and misses; add rules to fix them.

Day 8 — Add guardrails and contextual notes

• Show source and date prominently; make “Open official notice” the primary action. find-tender.service.gov.uk
• Flag which regime applies (new Procurement Act vs legacy) to set expectations on timelines and notice types during the transition. gov.uk
• For grants, remind users that Find a Grant is now the single, mandatory place for eligible central government grants, but some non‑central funders advertise elsewhere—hence the value of open data via 360Giving. gov.uk

Day 9 — Workflow and accountability

Wire up an “owner” field and a simple Kanban: Qualify → Draft → Review → Submit. Auto‑create a deadline entry the owner cannot dismiss without a reason. Add a Friday review: what converted, what didn’t, and what to refine next week.

Day 10 — Go‑live with feature flags

Switch on alerts for one team first. Keep the weekly digest for everyone else until you’re confident. Do a 30‑minute teach‑in and a one‑page quick start. Use a blackout rule for weekends and after 6pm. When you’re happy, expand coverage. For a simple go/no‑go checklist before flipping the switch, see our one‑pager on a Go‑Live Gate and our rollout tips in Feature Flags for AI.

KPIs your board will understand

• Signal quality: percentage of alerts rated “High fit” by a human within 48 hours. Target 60%+ by week 4.
• Coverage: proportion of relevant opportunities found vs a manual benchmark during week 1 shadow‑mode. Target 85%+ by week 3.
• Time to qualify: median minutes from alert received to human decision (go/no‑go). Target under 10 minutes by week 2.
• Throughput: qualified opportunities per fortnight by service line.
• Conversion: submissions and shortlist rate; grants awarded or tenders won.

Costs, risks and the simple mitigations

Quality checks: a 15‑minute weekly ritual

1. Pick five alerts at random; verify links, dates and values on the official postings. find-tender.service.gov.uk
2. Audit scoring by asking “Would we actually bid?” If not, update the rubric.
3. Review misses by sampling FTS, Contracts Finder and Find a Grant manually; add synonyms or filters. gov.uk
4. Close the loop: did last week’s “High fit” items progress? If not, why?

Procurement questions to ask vendors of “opportunity scanners”

If you prefer to buy rather than build, use these questions in demos and reference calls:

• Which sources do you cover natively today—FTS, Contracts Finder, Find a Grant, 360Giving—and how quickly are new notices ingested? Ask for proof. find-tender.service.gov.uk
• Can we bring our own scoring rules and edit them without your help?
• How do you prevent or flag hallucinations in summaries?
• What’s your policy on linking to the authoritative notice? We require one‑click access.
• How do you handle the 2025 regime transition and legacy notices?
• What are your measurable outcomes after 60 days (precision, recall, time saved)?
• Show us three customer references that look like us (size, sector, region).

For a fuller vendor shortlisting approach, see our AI Vendor Due Diligence Pack.

Operating notes for charities and SMEs

• Grants vs tenders: Grants have funder‑specific rules and are increasingly visible on Find a Grant; many other funders publish open data via 360Giving, so use both. find-government-grants.service.gov.uk
• Devolved nations: Scotland, Wales and Northern Ireland maintain their own below‑threshold portals; your copilot should reflect your delivery footprint. gov.uk
• Deadline hygiene: Put a single service‑wide calendar in place before you scale the copilot.

What it cost and what it returned (case study)

In four weeks post go‑live, the team processed 84 opportunities (37 “High fit”). They submitted 9 tenders and 6 grant applications. Two grants were awarded and one tender was shortlisted, paying for the time spent several times over. The time‑to‑qualify dropped from ~25 minutes to under 9 minutes per item by week 3, mainly due to better summaries and calendar discipline.

The stack stayed deliberately lean: existing mail and collaboration tools, simple shared storage, and a lightweight rules engine. No custom integrations needed to start; official portals’ saved‑search notifications did the heavy lifting. find-tender.service.gov.uk

When to pause or pivot

• If precision (share of “High fit” alerts) is under 40% after week 4, revisit your rubric and keywords before adding more sources.
• If no submissions result from “High fit” items by week 6, the bottleneck is likely internal capacity, not discovery—fix ownership.
• Kill it if it becomes another inbox. One page, one list, one weekly ritual—keep it small and useful.

Want help standing this up?

We can help you calibrate searches, design the scoring rubric, and run the first two weeks of shadow‑mode so you go live with confidence. If you’re moving from pilot to production, these posts will also help: designing trustworthy UX patterns in Copilot UX that earns trust and our Go‑Live Gate.