Recommended AI Features to Focus On

Based on: AI & Innovation sections (questions 5–10) of all 23 competitor reports, plus limitations (question 18) Date: September 2026

Summary

Across the 23 products, AI has split into three tiers:

Table stakes (2026): GenAI drafting of RFx documents and summarising of supplier responses. Every Gartner leader and every AI-native newcomer has these.
Where products now compete: agents that execute work rather than assist it, such as autonomous sourcing of low-value spend, bid evaluation with award recommendations, and supplier onboarding and risk checks. Most of these launched between late 2025 and 2026, so production maturity is still low across the market.
Data moats: price benchmarking and predictive analytics. These depend on transaction data at scale (Coupa’s $8T+ community dataset, Fairmarkit’s 2.7M-supplier network) and can’t be copied quickly.

Our direct tendering competitors are far behind on this curve. bids.do has no documented AI. Bidzar has ML supplier selection but no GenAI. Promena has only OCR and SRM risk analysis. Only Zinit has a real AI proposition (60-second RFP generation, supplier matching, AI scoring, partly autonomous negotiation).

The 6 features below are chosen to (a) close the table-stakes gap quickly, (b) take the agentic ground the regional players have not reached, and (c) start building our own data moat from day one.

The 6 Features

1. Conversational Intake → Auto-Drafted RFx

What it is: A requester describes the need in plain language (or uploads a spec, email or old tender) in Azerbaijani, Russian, Turkish or English. AI turns it into a structured request, suggests the category, and drafts a complete RFI/RFQ/RFP (lots, specs, evaluation criteria, T&Cs) for the buyer to review.

Market evidence:

Near-universal among leaders: Zip (AI Procurement Concierge), Zycus (Merlin Intake), Tonkean (ProcurementGPT), ORO Labs (AskORO), Omnea, Ivalua, SAP Ariba (Intake Management: +12% productivity).
Intake straight into a sourcing event: Keelvar Kai and Fairmarkit’s Intake Agent go from one prompt to a launched event.
Regional competitors: Zinit’s “RFP in under 60 seconds” is its headline feature. Bidzar, bids.do and Promena have nothing comparable.

Why focus: This is the table-stakes feature. Without it we lose demos to Zinit. Multilingual drafting is our differentiator: global suites are English-first, and Zinit only offers translation.

Effort: Low–medium (LLM + category templates + our existing RFx data model).

2. AI Bid Evaluation & Comparison

What it is: Suppliers submit in any format (PDF quotes, Excel, scanned documents). AI extracts and normalises the prices and terms, checks each bid against the mandatory requirements, flags missing documents and anomalies, and produces a ranked side-by-side comparison with a plain-language award recommendation. The buyer always makes the award decision.

Market evidence:

Ranked recommendations: Levelpath (ranked comparisons with award recommendations), Fairmarkit (Evaluation Agent with TCO and scenarios), SAP Ariba (Joule supplier-response summary).
Extracting quotes from non-standard documents: Oracle (Negotiation Surrogate Response Assistant, 26C), Ivalua (automated response summaries).
Regional competitors: Zinit has AI-driven scoring. Mercell deliberately keeps scoring rule-based to comply with procurement law.

Why focus: Evaluating bids is the most labour-intensive step of tendering and the step buyers most want help with. It fits our tender-centric users better than generic intake. Keeping a human on the award keeps it compliant for public-sector and regulated buyers.

Effort: Medium (document extraction + normalisation + scoring engine tied to the tender’s evaluation criteria).

3. Autonomous Sourcing for Low-Value & Repeat Purchases

What it is: Below a configurable threshold (for example under 10,000 AZN, or catalogue and repeat items), an agent runs the full event with no buyer involved: it picks suppliers, invites them, sends reminders, collects bids, closes the event at the best time and proposes (or auto-awards) the winner within policy. Buyers only handle exceptions.

Market evidence:

Specialists: Fairmarkit (low-value requests run end to end; “Total Agentic Sourcing”, Apr 2026) and Keelvar (80–90% automation reported by Siemens and Caterpillar; Optimal Close and Rate Refresh agents).
Suites: Oracle (Autonomous Sourcing Agent, Feb 2026), Coupa (Navi Autonomous Sourcing Agent), Zip (Sourcing Superagent, Sept 2026).
Regional competitors: Zinit’s negotiation is only partly autonomous. No other regional competitor has this.

Why focus: This is where the market is moving, and no regional competitor offers it end to end. It has a clear, measurable ROI story (cycle time, buyer hours saved, savings on spend that was never sourced before). It also produces the transaction data that Feature 6 needs.

Effort: Medium–high (orchestration, policy thresholds, full audit trail). Build it on top of Features 1 and 2.

4. Supplier Discovery, Matching & Risk Screening

What it is:

Matching: for every event, AI recommends suppliers from the customer’s own base and from our network, based on category fit, past performance, location and how likely each supplier is to respond.
Onboarding checks: new suppliers are screened automatically against local company registries, sanctions and adverse-media sources, and have their financial health checked. The AI pre-fills and reviews onboarding questionnaires.

Market evidence:

Supplier matching: Fairmarkit (2.7M-supplier network with a response-likelihood score), Zinit and Bidzar (selection based on behavioural patterns and financial indicators), Levelpath (supplier graph ranking).
Onboarding and risk checks: ORO Labs (unified risk scoring; 80% shorter onboarding, 4x more fraud checks), Omnea (Dow Jones monitoring), Zip (AI Risk Orchestration / TPRM, Sept 2026).

Why focus: In our markets, a lack of competition (too few bidders) is often a bigger problem than price. Bringing in more qualified local suppliers is a direct value driver. Screening against local registries is something global vendors don’t do for our region.

Effort: Medium (the matching model is simple; building local data-source integrations is the main work).

5. Trustworthy AI: Audit Trail, Explainability & Data Privacy

What it is: A cross-cutting layer, not a single screen:

Audit trail: every AI suggestion, score and automated action is logged with its inputs, its reasoning and the person who approved it.
Explanations: each recommendation shows why it was made.
Guardrails: customers set policy limits on what agents may do alone.
Privacy: customer data is never used to train shared models, and data can be hosted in-region.

Market evidence:

Auditable AI: Ivalua (AI traceability built in; never trains LLMs on customer data), ORO Labs (“intelligence + orchestration + governance”, ISO 42001), Zip (deterministic logic for high-impact steps; MCP within its permission model), Levelpath (agents act within customer-set governance limits).
User concerns: SAP Ariba users worry about hallucinations in mission-critical steps. Mercell keeps humans in control of award decisions.

Why focus: Public-sector and state-owned buyers, a large share of the tender market, will not adopt AI they can’t audit or explain to a review committee. Treating trust as a feature turns the “hallucination risk” objection into a selling point. It also makes Features 2 and 3 approvable.

Effort: Medium, but it has to be designed in from the start. Adding it later is expensive.

6. Price Intelligence & Negotiation Assistant

What it is:

Price benchmarks: AI benchmarks every quote against historical tender prices on our platform (anonymised and aggregated with customer consent), contract rates and market indices, and flags quotes that are unusually high or low.
Negotiation: for shortlisted suppliers, an assistant suggests target prices and counter-offer points, and can run structured multi-round negotiation.

Market evidence:

Price benchmarking: Coupa (community intelligence from $8T+ spend), Zip (benchmarking via NPI, Tropic, Beroe and others), Omnea (Price Intelligence, June 2026), Keelvar (rate benchmarking and automatic re-tendering when rates drift).
AI negotiation: Zinit (multi-round negotiation orchestration), Fairmarkit (negotiation guidance developed with Chris Voss).

Why focus: This is the data moat. Few global vendors have benchmark data for regional tender prices, and every event on our platform adds to it. It also supports the savings claims buyers need to justify the platform.

Effort: Low to start (internal historical benchmarks). It gets more valuable as volume grows, which makes it a strong phase-2 feature.

Priority & Roadmap

#	Feature	Market status	Regional gap	Effort	Phase
1	Conversational intake → auto-drafted RFx	Table stakes	Only Zinit has it	Low–Med	Phase 1
2	AI bid evaluation & comparison	Differentiator	Zinit partial	Med	Phase 1
5	Trustworthy AI (audit, explainability, privacy)	Emerging requirement	None	Med	Phase 1 (foundation)
4	Supplier discovery, matching & risk screening	Differentiator	Zinit/Bidzar partial	Med	Phase 2
3	Autonomous sourcing for low-value spend	Leading edge (2026)	None	Med–High	Phase 2
6	Price intelligence & negotiation assistant	Data moat	Zinit partial (negotiation)	Low → grows	Phase 2–3

Why this order: Phase 1 closes the demo gap with Zinit and lays the trust foundation that public-sector buyers require. Phase 2 moves from AI that assists to AI that executes, which is where the global leaders went in 2025–2026. Feature 6 improves automatically as Phases 1–2 bring transaction volume onto the platform.

Deliberately Not Prioritised (Yet)

Capability	Seen at	Why not now
No-code agent builder / agent studio	Levelpath Ranger Studio, ORO Agent Builder, Oracle AI Agent Studio, Ivalua	For large enterprises with in-house ops teams. Pre-built agents deliver more value to our segment
Supply-disruption prediction & spend forecasting	Coupa, GEP, Jaggaer	Needs cross-customer data at a scale we don’t have yet
MCP access for external AI assistants (ChatGPT, Claude, Copilot)	Zip, Omnea, Oracle	A quick win once our API is mature. Worth adding in Phase 3 but not a buying criterion today
Invoice AI (OCR, 3-way matching, touchless AP)	Oracle, Levelpath, Tonkean, Promena	Outside the tendering core. Better handled through ERP integration