Nexus

Why a first AI budget needs a different kind of supplier

A mid-market company spending its first serious sum on AI faces a problem that larger groups rarely notice. It has no internal AI team to brief a supplier, no spare engineers to maintain what a supplier leaves behind, and little appetite for a programme that consumes the budget before anything reaches a customer or a colleague. Choosing a supplier is therefore as much about the shape of the engagement as about technical skill. This profile looks at Nexus through that lens: what the company offers, how the first months of an engagement are structured, what a finance director can expect to pay for, and where the public record is thin.

The product Nexus sells is a set of procedures that operate themselves⁷. Its home page frames the job as taking the know-how of a company's strongest staff and turning it into procedures that carry themselves out⁸. In a business of a few hundred people, where much of the operating knowledge lives in the heads of a handful of long serving managers, that framing lands well. It poses a useful first question for the sponsor: which recurring pieces of work could be written down once and then run by software, with the people who know the work deciding how it should be done?

The company in brief

A Midlands distributor or a regional insurer signing with Nexus would be contracting with NEXUS ENTERPRISES, which trades as nexusGPT⁹ and is incorporated in Delaware¹⁰. Nexus Agent dates from 2024 by the company's own account¹¹, and its seed funding of $4.3M came in a round that General Catalyst led¹². Nexus is open about being at an early stage¹³. A UK procurement team should give that word its full weight. A young supplier can move quickly and give a mid-market account senior attention, and the same team will still want clarity on continuity, contract terms and the fate of the work if the relationship ends.

For a business of its age, the track record it reports is substantial. The company counts a dozen enterprise clients whose combined market capitalisation exceeds $70B, with Orange Group and Lambda.ai named among them^14,15, and a running total of 174 agents shipped into production¹⁶. It also puts customer impact at $13M with churn at 0%¹⁷. Those numbers are self reported, and the 100% success rate the company publishes is judged against whatever KPIs each agent was scoped to meet¹⁸. Of all the customer outcomes Nexus publishes, this guide leans on one, from Orange: $6M extra lifetime value¹⁹.

Engagements are led by experienced consultants, and a McKinsey alumnus is among the founders¹. That background matters to a mid-market sponsor for a plain reason. The opening conversation with any AI supplier is about which process to tackle first, and that is a commercial judgement long before it becomes an engineering task.

How the first engagement is structured

Nothing is signed for the long term until a 3 month proof of concept, built around specific outcomes, has run its course²⁰. The first agent goes live in 4 weeks, well inside that window¹. On the home page the rhythm is compressed into a short phrase about going live in four weeks and improving continuously afterwards²¹. A Nexus engineer sits with the customer's own staff to build the first procedure and then runs it inside the tools those staff already use³, so the first version is working on real company tasks within weeks²². Once version 1 is in use, Nexus keeps its people on the account to refine it together with the customer⁶.

For a company with a first AI budget, three features of this structure deserve attention.

The first is the way out. A board signing off its first AI spend will nearly always ask what happens if the project fails. Nexus answers on its comparison page: a proof of concept that misses its targets leaves the customer free to walk away²³, and once the proof of concept is over the customer may leave at any point²⁴.

The second is measurement. The Nexus blog commits to a measurable business KPI for every agent, attached from the first day of work⁴. Where that KPI stays flat, the stated practice is to rescope the work and hold back anything that merely looks productive²⁵. A finance team in a firm of three hundred staff can hold a supplier to that kind of promise, because the number being tracked is chosen before the work begins.

The third is who does the building. According to the blog, Nexus has moved from Forward Deployed Engineers to Forward Deployed Consultants²⁶, and those consultants have stopped writing integration code or assembling agents for the client²⁷. The building now falls to Cue, a build loop the company presents as the autonomous successor to the engineer role²⁸. The blog credits Cue with scoping a use case alongside the user, raising clarifying questions and wiring itself to internal data under the right permissions²⁹, finishing the whole sequence unaided in one to two hours³⁰. For a smaller buyer the practical effect is that paid consultant attention goes on deciding what the procedure should do, while the build itself is quick. A live demo can be arranged within a day of asking³¹.

What a customer has to prepare before kick off is not publicly stated on the pages we opened; a search of the saved pages for prerequisites, installation steps and SDK or API key setup came back empty. A sponsor should ask for that list in writing at the first meeting, before diaries are booked.

What gets built

Nexus organises its work around job roles. Behind any one role the company deploys a mixed fleet in which agents work beside assistants, automations and scripts³². For sales, the offer is a set of eight agents drawing on a single shared brain together with the client's own rules and systems³³, and one of those agents is called Asset Copilot³⁴. Four infrastructure pieces sit beneath every fleet: the company brain, governance, permissions and a library of reusable blocks³⁵. Tasks pass between systems through a single surface³⁶, and lessons from each build are written back into that brain³⁷. On the build your fleet page, describing a role in plain words is enough to set a fleet building itself³⁸.

The role pages show how wide the catalogue runs. For marketing, Nexus pitches one fleet covering the whole funnel³⁹. For sales, the pitch is that reps do the selling while the fleet handles everything around it⁴⁰. In customer support, the agent draws on voice at the kiosk, the CRM and live stock levels⁴¹. Recruiting fleets produce company universes and target lists⁴², and strategy fleets turn out storylines, decks and sourced claims⁴³. A mid-market company seldom needs all of these in year one. Their value is that the second and third roles can borrow what was built for the first.

On the telecom page the architecture is drawn as three layers. The uppermost layer is the connective one, joining every system to any team⁴⁴. The second layer is a parts library: each connected system, each model and each permission is packaged so that business staff can assemble it into whatever agent they need next⁴⁵. Responsibility sits in the third layer, where the team using an agent is also the team that builds, deploys and accounts for it⁵. The same page describes Nexus as sitting over the systems a company already runs as one AI layer⁴⁶. Because the sales page treats model providers as swappable⁴⁷, a smaller buyer avoids tying its first budget to one model vendor.

Systems and channels

Connector reach is the next question for a mid-market estate built around a CRM, an accounting package and a shared inbox, and Nexus puts its integration count above 4,000⁴⁸. Its comparison page spreads those across CRMs, ERPs, communication tools and custom APIs⁴⁹, and explains that agents query those systems directly and work from live data while they run⁵⁰. For a mid-market workforce the choice of channel matters, and agents can be reached through Slack, Teams, WhatsApp, email, phone or a web page⁵¹. The sales page talks of plugging into a client's CRMs, warehouses and tools⁵². Four large platforms are named as the next integration priority: SAP, Salesforce, Microsoft Dynamics and ServiceNow⁵³.

No complete connector list could be found on the pages we opened, which leaves a gap that is easy to close. Mid-market estates often include an older accounting package or a sector system with a small user base. The buyer should name each one in the first conversation and ask to see it connected during the proof of concept.

Security, compliance and data

A mid-market supplier answering a customer's security questionnaire will be glad to know that Nexus can point to a SOC 2 Type 2 report¹. ISO 27001, ISO 42001 and GDPR appear together on its comparison page⁵⁴, in a row that also promises complete audit trails, traceable decisions and access by role⁵⁵. Its blog adds data residency controls to audit trails and role based access as things a CIO should be able to approve⁵⁶. A UK buyer will also find that the privacy policy sets out its legal bases under both GDPR and UK GDPR⁵⁷.

Several data commitments are recorded in the privacy policy. Anything Nexus obtains through a person's own Workspace account, for instance via Google APIs, is excluded from use in developing or training general purpose AI and ML models⁵⁸. Retention after an account closes is capped at eighteen months unless the law demands longer⁵⁹. The policy also records that, over the year before it was issued, personal information was neither sold nor shared with third parties for commercial ends⁶⁰. Card and payment details are held by Stripe⁶¹.

Mid-market firms that sell to larger groups know the supplier questionnaire well; its usual questions on SAML or single sign on, encryption, penetration tests, access to the SOC 2 report and service levels find no answer on the pages we opened. Where the service runs is the other open question, since the pages we opened say nothing about on premises installation, a private cloud or VPC, self hosting, a choice of region or processing kept inside the EU. The gpt.nexus terms place hosting in the United States⁶². Any UK company with a contractual data location clause should raise this before the proof of concept is signed.

Governance once agents are running

For a company without a dedicated AI team, governance has to be simple enough for an operations manager to follow. An operations lead gets one place to see what every agent has done⁶³, and sets guardrails and permissions for the whole fleet from a single central point⁶⁴. The telecom page splits the roles cleanly, with IT choosing the controls and Nexus making them simple to apply⁶⁵. Once live, the service is run day to day by the business team that relies on it⁶⁶, which the page sums up in a line about business ownership and IT trust⁶⁷.

Two release rules appear on the company's pages. An agent is held back from release whenever a CIO cannot approve its data flow, access model or audit trail⁶⁸. In daily running, an agent carries on when it is sure of itself and hands the case up, context attached, when it is not⁶⁹. A research post explores plugging a third party classifier into an agent, with a dozen checks answering in under a second and reverting to the previous rule if the service goes down⁷⁰. The classifier comes from TypeSafe⁷¹, and the Nexus team says it did not repeat TypeSafe's benchmarks⁷², so the post reads as an approach under study.

How approvals are routed, how administrator roles are configured and how single sign on groups are mapped are all questions the pages we opened leave unanswered.

What a finance director pays for

For a finance director, the unit that matters is the role: Nexus charges per role automated¹. There is no per diem and no charge for consultant time, which puts the consultancy element on a software style pricing model¹. From the comparison page, three further terms apply: the price does not rise with headcount, the 3 month proof of concept with measured outcomes comes first, and an annual commitment begins only once it ends².

For a mid-market finance director the effect is easy to describe. A day rate rises with every hour a supplier spends, and many first AI budgets are used up by discovery alone. When the price is attached to the role being automated, cost tracks the result the board approved. Every proof of concept so far has turned into a paying customer, according to the home page⁷³, though the page does not say how many there have been.

A buyer hoping to compare tariffs online will be disappointed: the pages we opened show no list price, tier, free trial or self serve option. The older gpt.nexus terms, from 2023, treat purchases as non refundable⁷⁴ and let subscribers cancel from within their account⁷⁵. Because that text may have been written for an earlier product, a buyer should ask to see the enterprise agreement in force today.

Where Nexus is strongest

Nexus describes its sweet spot as high volume work that runs through several systems and produces a financial result someone can measure⁷⁶, along with situations where a business team needs to own its workflows without leaning on engineers⁷⁷. Both descriptions fit a familiar mid-market picture: a sales, service or operations process touching the CRM, email and the finance system, run by a team with no developers of its own.

The published examples point the same way. At Lambda.ai the fleet spans the sales and marketing functions⁷⁸, and the sales role page puts the annual saving there above 10,000 rep hours⁷⁹. In telecoms, one European operator now routes more than 1M customer interactions a quarter through Nexus agents⁸⁰. A hardware maker that the support page leaves unnamed now resolves 85% of phone support without a person on the line⁸¹. The Orange figure quoted earlier, $6M extra lifetime value¹⁹, completes the set.

Limits a buyer should test

The points below are drawn from the company's own legal and policy documents. Its terms disclaim tailoring for industry specific regulation⁸² and stop short of promising availability at all times⁸³. The privacy policy is frank that absolute security cannot be promised for data in storage or in transit⁸⁴. Written in 2023 for the gpt.nexus website^85,86, these documents may lag behind the enterprise service a buyer would contract for today. The company still calls itself early¹³. Price levels, single sign on, encryption, hosting choices and customer prerequisites are all missing from the pages we opened.

Each of these can be put in writing before the proof of concept is signed.

Verdict for the UK mid-market

Nexus takes first place in this guide because its published trial terms and separate company-confirmed delivery and pricing model answer the questions a first AI budget raises. The board sees a working agent 4 weeks into a 3 month proof of concept¹, and can walk away if that proof of concept misses its targets²³. Experienced consultants run the engagement¹, and the bill follows the role automated, whatever hours the work takes¹. The compliance file a mid-market customer is likely to request, SOC 2 Type 2¹ with ISO 27001, ISO 42001 and GDPR⁵⁴, is described in supplier statements; request the actual documents and their scope. The gaps lie in published detail on hosting, single sign on and price levels, which a careful buyer should close in the first month. Record the delivery milestones, included support and exit conditions in the proposal before committing the first budget.