// Playbook 05 — Vendor & partner selection
Selecting a transformation partner.
A six-week playbook for selecting and onboarding a transformation partner without losing optionality. Frame the brief. Shortlist on published criteria. Evaluate on scenarios and triangulated references. Onboard against a charter, not a SOW. Worked example throughout: Halcyon Financial and the partner they select — Lighthouse AI.
onboarded
to shortlist
· with exit case
accepted
The brief.
Worked exampleOne workstream needs help from outside.
Halcyon Financial is six weeks into the transformation programme set up in Playbook 04. Three workstreams are scoped: the knowledge-base AI-assist (8 weeks, internal), the ticket-triage POC (12 weeks, internal), and the CX tool pilot (16 weeks, vendor-led). The first two are running. The third hasn't started — because the partner hasn't been chosen yet.
The Transformation Lead has six weeks to run the selection. Sam Patel (Engineering Director) is the workstream Accountable. Anna Petrović (Head of Compliance) holds the security and data-residency line. Rachel Doyle, new to the cast, owns procurement and the vendor-risk register.
Twelve vendors will end up on the longlist. Three will make the shortlist. One — Lighthouse AI, a 60-person specialist with three years of comparable CX tool pilots in regulated industries — will be onboarded by week six.
the longlist
published criteria
· 3 vendor, 2 own
· not a SOW
// And the people you'll meet
The Transformation Lead
You, the readerAccountable for nothing. Responsible for the selection process, the partner charter, and the integration plan. The role this playbook is written for.
Sam Patel
Engineering DirectorAccountable for the CX tool pilot from Playbook 04. The internal owner who'll inherit what the partner builds. Defines the brief in Phase 1.
Anna Petrović
Head of ComplianceHolds the non-negotiables: SOC 2 Type II, AU data residency, security review. Sign-off on any partner who touches customer data.
Rachel Doyle
Head of Procurement & Vendor RiskJoins the cast for partner scoring. Owns the contract, the vendor-risk register, and the exit-clause language. The procurement seat the Transformation Lead doesn't try to do themselves.
Priya Nair
Head of CXWill inherit the system once the partner exits. On the evaluation forum — pushes hard on whether officers will actually use what gets built.
Tom Nguyen
CX Team LeadEmbedded with Lighthouse AI three days a week during the 30-day integration. The pair-builder, not the receiver.
Four phases. Frame, shortlist, evaluate, onboard.
The framework// Most partner selections fail in Phase 1 — the buyer hasn't defined what they're buying, so the vendor defines it for them.
Frame
Weeks 1–2Define what the partner owns AND what they don't. Run the buy-vs-build test. Write the exit case before the entry case.
Shortlist
Weeks 2–3Longlist from multiple sources. Score against published criteria. Cut to three — not six, not one.
Evaluate
Weeks 3–4Scenario-based tests on real data. References triangulated, not just vendor-provided. Total cost, not licence cost.
Onboard
Weeks 5–6Charter, not SOW. First 30 days as pair-building, not handoff. Quarterly partner review scheduled.
Phase one · Frame.
Weeks 1–2The goal of the first phase is protecting optionality. Before the first vendor pitch arrives, the buyer should have written down what the partner owns, what they don't, what the exit case looks like, and what's non-negotiable. Without that, the vendor's deck does the framing — and every deck is optimised for the vendor's strengths.
Define what the partner owns AND what they don't.
A one-page brief: scope, boundaries, hand-off point. Both sides of the line. The "doesn't own" list is where the protection lives. Without it, scope creeps silently and the partner becomes load-bearing in places no one signed off.
Run the buy-vs-build test honestly.
Could the internal team do this if reassigned? What would it cost in capacity? What would it cost in time? The answer often isn't "build" — but the answer should be reached on evidence, not by default. The vendor's job gets easier when the buyer can't say no.
Write the exit case before the entry case.
At the end of the engagement, how does the work continue? Who operates it? What's the transition cost? If the exit case isn't credible at week one, the partnership becomes irreversible at week six. Choose for the exit, not the entry.
Set the non-negotiables.
Security posture (SOC 2, ISO 27001, equivalent). Data residency. IP ownership. Exit-clause length. Any vendor who can't meet these doesn't make the longlist. The non-negotiables are the saved-no — every yes that comes later is faster because of them.
Two sides of the line, written down.
The Transformation Lead spends week 1 with Sam Patel and Anna Petrović framing the brief. The buy-vs-build test takes a morning: the internal ML team is already on the ticket-triage POC and the knowledge-base AI-assist. Capacity for a third workstream doesn't exist. Buy.
The exit case is the harder conversation. At week 14, who operates the CX tool? Answer: Sam Patel's team, with two engineers embedded from week 1 of the partner engagement. Capability transfer checkpoints baked into weeks 6, 10, and 14. If the partner can't structure the engagement around those checkpoints, the partner is wrong for this work.
- Partner ownsBuild the CX tool · train first 20 CX officers · run weekly demos · hand off at week 14
- Partner doesn'tOngoing operation · scaling beyond pilot · CX-wide change management · IP ownership
- Exit caseInternal team operates from week 14 · partner on retainer for 60 days · clean exit at day 90
- Non-negotiablesSOC 2 Type II · AU data residency · 30-day notice exit · capability-transfer plan
The "doesn't own" list is the most powerful sentence in the brief. Vendors are used to writing what they do; almost no buyer writes what the vendor doesn't. The list shows up six months later when scope creep starts — and the list is what makes the no easy.
Phase two · Shortlist.
Weeks 2–3A good shortlist is the most underrated artefact in the process. Three vendors who fit the brief beats six vendors who fit a category. The discipline is to build the longlist from multiple sources (not just one analyst report, not just one peer), and to write the scoring criteria before the vendor names enter — so the scoring isn't bent by who the analyst was selling that quarter.
Build the longlist from multiple sources.
Peer referrals (especially other Transformation Leads in the same sector). Analyst reports — but at least three different ones. Repeat-buyer recommendations from past partners. Conference circuit. The internal Engineering and Data teams' networks. If the longlist comes from one channel, the bias is baked in.
Score against published criteria.
Five to seven criteria, weighted by importance, written down before vendor names enter the spreadsheet. Regulated-industry experience. AI delivery track record (not strategy decks — actual shipped work). Team continuity at the partner. Exit-friendliness. Total cost. If the criteria appear after the vendors, the criteria are doing politics, not selection.
Cut to three.
Three is the right number. Two becomes a comparison; four becomes a beauty parade. Three lets you triangulate — and lets the shortlisted partners know the competition is real. The discipline is the cut. Six "interesting" vendors that don't get cut becomes twenty calendar weeks of process.
Twelve to three, scored before names.
The Transformation Lead and Rachel Doyle build the longlist over five working days. Sources: four peer referrals from other AU fintech Transformation Leads, three names from analyst reports (cross-referenced across Gartner, Forrester, and an independent), three from Sam Patel's engineering network, two from Halcyon's existing vendor stable.
The criteria are written first. Five of them, weighted: regulated-industry experience (25%), AI delivery track record (25%), team continuity (15%), exit-friendliness (20%), total cost (15%). Cost gets the lowest weight intentionally — most vendors compete on day rate; the criteria that matter are the ones the vendor can't manipulate after signing.
The cut to three is made by the Transformation Lead and Rachel in a 90-minute working session. Five vendors are eliminated for failing the non-negotiables (security posture or data residency). Four are eliminated for scoring below 3/5 on two or more criteria. Three remain.
Criteria that reward the vendor you've already decided you want. If the criteria include "global presence" and only one shortlisted vendor is global, the criteria are doing the wrong work. Write the criteria from the brief, not from the bench.
Phase three · Evaluate.
Weeks 3–4By Phase 3 you have three shortlisted partners. The next discipline is to get past the deck. Every vendor's deck looks impressive. The evaluation should produce evidence the deck can't manufacture: scenario performance on your actual data, references the vendor didn't provide, and total cost including the parts the SOW won't quote.
Run scenario-based tests on real (anonymised) data.
Each shortlisted partner runs a defined exercise — three to five days — on a representative slice of your data, properly anonymised. Not a demo. Not a generic capability showcase. A specific task pulled from the brief. The output reveals what the deck can't.
Triangulate references — not just the ones offered.
The vendor provides three. You find two more — through your own network, LinkedIn, or community. Vendor-provided references are filtered by definition. The references you find yourself tell you what the filter took out. Two unfiltered references beat ten filtered ones.
Compare total cost, not licence cost.
Day rate × duration is the headline. The real number includes integration cost (internal time committed alongside the partner), change-management uplift, the cost of capability transfer, and the transition cost at exit. Cheap on day rate, expensive on exit, is the most common pattern.
Make the call in a documented forum.
Same discipline as the prioritisation forum from Playbook 03 and the steering from Playbook 04. Decisions, not minutes. Three lines: what was decided, why, who owns the next step. Forum membership: Transformation Lead, workstream Accountable, Compliance, Procurement, the user-community lead. Five seats, sixty minutes.
Scenario test, five references, one decision.
Each shortlisted partner runs a five-day scenario on 200 anonymised CX tickets pulled from Halcyon's archive. The task: triage and suggest resolution paths. Lighthouse AI's output demonstrates the deepest understanding of the regulated-industry constraints — its suggestions defer correctly on anything touching KYC or AML, where the other two over-reach.
References. Each vendor offered three. Rachel Doyle and the Transformation Lead found two more for each vendor — via Slack communities, AU fintech peer groups, and one carefully placed LinkedIn message. The vendor-provided references for Lighthouse AI told a story of strong delivery; the unfiltered ones added that the team turns over more than the vendor's deck implied. The honest reference sharpened the decision rather than breaking it.
The forum convened on a Tuesday. Transformation Lead (chair), Sam Patel, Anna Petrović, Rachel Doyle, Priya Nair. Sixty minutes. The decision was unanimous, but the discussion took fifty of the sixty — most of it on the total-cost picture rather than the headline day rate.
- Scenario test5 days · 200 anonymised tickets · Lighthouse strongest on regulated edge cases
- References3 vendor-provided + 2 found · Lighthouse honest on team turnover
- Total costHigher day rate · lower integration cost · cleaner exit · best lifecycle value
- Forum callUnanimous · documented in three lines · 48-hour publication SLA
The most useful reference call is the one with a buyer who didn't renew. Vendor decks list the renewals; nobody volunteers the ones who walked away. Tracking down a non-renewer takes effort and often reveals the most important pattern of all.
Phase four · Onboard.
Weeks 5–6Phase 4 is the difference between a partner who's working with you and a partner who's working at you. The artefact is a charter, not a SOW. The first 30 days are paired, not handed off. The quarterly review is on the calendar before day one. Done well, the partnership starts with the exit already structured.
Write a charter, not a SOW.
The SOW is what procurement signs — it lists deliverables and dates. The charter is what the working teams use — it lists what the partner owns, what they don't, the capability-transfer checkpoints, the exit triggers, and how disagreements get escalated. One page. Co-signed by the partner's delivery lead and the internal Accountable owner.
First 30 days as pair-building, not handoff.
The partner sits with the internal team three days a week. Pairing, not parallel. Capability transfer is observable from day one — not deferred to "after delivery." The internal team that will operate the system from week 14 is in the room from day one, learning by building, not by inheriting.
Schedule the quarterly partner review.
Before the partner starts, the first review is on the calendar. Same discipline as the programme's quarterly value review. Outcomes against the charter — not "how are we feeling about the partnership" but "is what the partner owns being delivered, and is what they don't own still off-limits."
- Design and build the CX tool — model, prompts, evaluation harness, UI.
- Train the first 20 CX officers · run weekly demos to the workstream forum.
- Pair with two Halcyon engineers (Sam Patel's nominees) from week 1.
- Capability-transfer checkpoints at weeks 6, 10, 14 · signed off by Sam Patel.
- Hand off to Halcyon's internal team at week 14 · 60-day retainer follows.
- Ongoing operation of the CX tool after week 14.
- Scaling beyond the pilot cohort to the broader CX organisation.
- Change management or communications across CX teams.
- IP ownership — Halcyon owns the build, including custom prompts and harness.
- Anything customer-facing without Anna Petrović's sign-off.
A one-page charter, pair-building from day one.
The charter is signed by Sam Patel (Engineering Director, Accountable) and Lighthouse AI's delivery lead on the Tuesday of week 5. One page. Two columns. Both sides of the line written down — what the partner owns, what they don't.
Day one of week 6, Lighthouse's three-person team arrives at the Halcyon office. Tom Nguyen joins them three days a week as the CX embed. Sam Patel's two engineering nominees sit in pairs from day one. The first build session happens on the Friday — not a demo, an actual ticket-triage run on real data.
The quarterly partner review is already on the calendar for week 12. Anna Petrović and Rachel Doyle are scheduled into it — review against the charter, not against the feeling of the relationship. The week 12 review will also re-confirm the week 14 exit checkpoint: is the internal team ready to operate without the partner?
Choose for the exit, not the entry.
Three cadences. Weekly, monthly, quarterly.
Operating rhythm// Once the partner is onboard, the relationship runs on three rhythms. Each protects something different.
The stand-up.
- Partner reports blockers
- Internal team confirms pairing time
- What's shipping this week
- Any non-negotiables at risk
The steering.
- Scope review vs the charter
- Change requests · yes / no / renegotiate
- Capability-transfer progress check
- Adoption metrics, not just output
The partner review.
- Outcomes vs the original charter
- Is the "doesn't own" list still holding?
- Exit case · is it still credible?
- Renew · renegotiate · or exit
Four ways this fails.
Common pitfalls// Every partner engagement that lost its way failed in one of these four ways. They're all easier to prevent than to fix.
The vague SOW.
The contract lists deliverables but never names what the partner doesn't own. Six months in, the partner is doing change management, operating the system, and writing internal training — and no one signed off on that expansion. Exit is now expensive.
Charter, not just SOW. Both sides of the line. Quarterly review against the "doesn't own" list. Anything outside the list is a yes/no/renegotiate decision — not a silent continuation.
The reference echo chamber.
All references the vendor offered. All from the same era. All happy by definition — vendors don't offer references who'll say something useful and difficult. The selection happens on a curated story.
Two references the vendor didn't provide. At least one non-renewer if findable. LinkedIn, peer networks, sector Slack groups. The unfiltered references are what make the decision honest.
Day-rate dazzle.
The cheapest day rate wins, and the procurement chart shows the savings. Twelve months later, integration cost, change-management uplift, and exit transition cost are double the difference. The savings were imaginary.
Total cost over the lifecycle, including exit. Day rate is one input out of four. Weight it low — vendors compete on the line they know you'll read first.
The capability that never transferred.
The partner did the work. The internal team watched the demos. At week 14, the system is live and nobody internal knows how to operate it. The "60-day retainer" becomes 180 days becomes a permanent dependency.
Pair-building from day one, not handoff at the end. Capability-transfer checkpoints written into the charter, signed off at each milestone. If the milestone slips, the engagement pauses — it doesn't extend.
By week six, you should have.
Starter checklist// Twelve items. If you can tick all twelve at week six, the partner is onboard, paired with the internal team, and the exit is already structured.
Using this in practice.
CloserThe cost of optionality is discipline.
Most partner relationships don't end badly. They end blurry — scope crept, capability didn't transfer, the exit case stopped being credible somewhere around month five. The discipline this playbook describes is the cost of optionality. None of it is hard. Almost none of it is done.
What travels is the shape — frame, shortlist, evaluate, onboard — and the principle that the most expensive partnership is the one you couldn't end. Halcyon's Lighthouse AI engagement is one shape. Yours will be different. The discipline will be the same.
If you're standing one of these up and want to talk through where it's getting stuck, I'm happy to.