There has been a ton of commentary about the MIT Stanford study indicating the overwhelming majority of AI pilots fail. If 2024/5 were the years when it was ok to try and fail, it feels like 2026 is where it needs to “get real” or you will not win support for your POC.
Moreover, with multi-use case AI tools like AI negotiation, it may deliver pedestrian ROI on the initial deployment but amazingly good returns on the scale opportunity so you need eyes on the long term prize.
The best stories I have heard recently about getting pilots successfully deployed and scaled were from Cyril Pourrat when chatting on my podcast. I really recommend you listen as he speaks with great clarity about empowering the team to try new things and then supporting them to sell it internally to other departments and colleagues, it was a true bottom up approach. Cyril was adamant that a top down, command and control attitude kills innovation.
The real reason AI pilots get blocked
When senior stakeholders say “no,” concerns often stem from structural areas. Four of the most common concerns are:
- We can’t govern it. (What’s the policy? Who’s accountable?)
- We can’t control it. (Will it say something stupid to a supplier?)
- We can’t measure it. (Will we actually see ROI?)
- We can’t scale it. (Is this another one-off tool?)
If you can address those four things up front, you stop selling “AI” internally and start selling a controlled commercial process. Let’s break this down into five simple steps.
Step 1: Start with an AI Negotiation use case that makes approval easy
Pick a pilot use case that is:
- Repeatable (same concept, easy to run several times)
- Low relationship risk (not your top strategic suppliers)
- Commercially meaningful (enough volume to prove value quickly)
- Operationally simple (simple negotiable terms, ideally anywhere up to 5 negotiable terms at once)
Great starting points tend to be simple service renewals; rate-card negotiations; or benchamark-able buys. Try to avoid advocating for pilots in strategic suppliers or heavily bespoke contracts.
Step 2: Pitch the AI Negotiation pilot, not the tool
Rather than pitching the AI tool itself, pitch the experiment or campaign you want to run:
- Objective: what outcome you’re proving (e.g. reduce renewal uplifts, extend payment terms, increase compliance)
- Scope: categories, geographies, contract types, and explicit exclusions
- Guardrails: what the AI can/can’t concede, escalation rules, approval thresholds
- Success metrics: Any relevant measurable KPIs
- Stakeholders: who owns policy, supplier comms, legal review, infosec sign-off
- Timeline (Nibble’s Proof Of Concept campaigns typically run on a 3-month “test – launch – analysis” structure)
- Decision rule: what “go/no-go” looks like for scale post-POC
Step 3: Get broader teams on board (and excited about AI Negotiation!)
The fastest way to secure buy-in from Legal, InfoSec and Procurement leadership is to be very explicit about control. When will a human get involved, and how? Can you share full transcripts with the agent, rationale, and outcomes per negotiation for full auditability? What guardrails are in place to protect the business?
If stakeholders feel the tool you want to pilot is governed procurement, rather than experimental AI, approval becomes much easier.
Step 4: Build a SIMPLE ROI model for AI Negotiation
Keep the maths simple, focusing on value you can attribute. For example:
- Price / rate movement vs baseline (or vs requested uplift)
- Payment term improvements (cash impact is real and easy to explain)
- Avoided uplifts on renewals that would have auto-rolled
- Time saving (this one may grow in scaling)
A strong pilot will feature one core KPI you can define and track so that defining success becomes an easier job for everyone involved.
Step 5: Design the AI Negotiation pilot so scale is the obvious next-step
The last thing you any of us wants is the hard-fought pilot to fail at the handover to scale. Thinking beyond the POC, have a plan with where you want to take the technology next if it proves successful. Incorporate a scale plan where possible. For example,
- Phase 1: Prove AI negotiation with the rapid POC on a smaller batch of suppliers
- Phase 2: Expand volume and add a secondary, adjacent use case with more suppliers
- Phase 3: Embed fully in workflow, full integration, operationalise reporting
The POC is not your destination – it’s essentially stage 1 of your rollout.
Just One More Thing
Erik Brynjolfsson has a great piece in the FT explaining why early signs of productivity gains from AI are low initially and how we might be on the edge of seeing it in growth this year:
“General purpose technologies, from the steam engine to the computer, do not deliver immediate gains. Instead, they require a period of massive, often unmeasured investment in intangible capital — reorganising business processes, retraining the workforce and developing new business models. During this phase, measured productivity is suppressed as resources are diverted to investments.”
https://giftarticle.ft.com/giftarticle/actions/redeem/6e934658-d748-48eb-825a-e3cda1251e16
