Alessandro Singh
Co-founder
Head of FSI, Retail and Manufacturing at Multiverse Computing
Enterprise selling to technical buyers, and the engineering to put AI to work in real operations.
Deploy & Operationalise AI
From AI that answers to AI that runs the work.
Most companies have deployed AI somewhere. Few run on it. We take each division from a chat assistant to processes AI runs every day, with an owner, a measure, limits and a fallback.
Three ways to use AI
- L1Deployed AIA chat assistant. Someone asks, it answers, people do the rest.
- L2Specialised AIIt works from your data and tools and prepares the work. A person approves.
- L3Operational AIIt runs the process within set limits, logs every step and hands people the exceptions.
Most companies stop at the first. We take each process as far as it should go.
By division
Pick the part of the business you run.
Each division has its own priorities, its own walkthrough and its own place to start.
Sales & growth
Go-to-market, pipeline and revenue that compounds.
Marketing
Content, campaigns and demand you can measure.
Customer service
Faster answers, with people on the hard cases.
Finance
Close, reporting and forecasting, with controls.
Operations & supply chain
Planning, scheduling and exceptions, run every day.
People
Hiring, onboarding and policy, fairly and lawfully.
Product & engineering
Shipping with AI in the workflow, quality measured.
Legal & risk
Review, drafting and the rules for everyone else.
Leadership
Where AI pays back, governed and funded.
See the difference
The same request, three ways.
A chat assistant waits to be asked. Operational AI starts when the work arrives, runs within set limits and hands people the exceptions. Switch between them.
Every division page has its own walkthrough.
closeweave.com/ai/finance/walkthroughIllustration
The request
InvoiceHalden & Rowe Components Ltd
Invoice INV-20931
Total £12,480.00, due in 30 days.
Please note our new bank details.
Autonomy level 1
Deployed AI
A chat assistant. Someone asks, it answers, people do the rest.
- A person types
Here is an invoice and the purchase order. Do they match?
- AI replies
They do not match. The invoice is £12,480 and the purchase order is £11,900, a difference of £580 on line 3. You may want to check with the supplier.
- Person
Payables clerk. Finds the goods receipt, emails the supplier, updates the bank details on file as asked, and posts the invoice.
3 steps · one to a person · about 25 minutes per invoice
- How long
- About 25 minutes per invoice
- Who does the work
- A person does every step
- What keeps it safe
- None: new bank details accepted without a check
- What goes to people
- Everything
Autonomy level 2
Specialised AI
It works from your data and tools and prepares the work. A person approves.
- Trigger
INV-20931 arrives in the payables inbox.
- Tools
erp.purchase_order("PO-4471") · erp.goods_receipt("GR-8812") · vendor.bank_details()Returned: The invoice is £580 (4.9%) above the order, all on line 3. Quantities match the goods receipt. The bank details differ from the record.
- AI replies
Two issues: a price difference on line 3, and changed bank details. A query to the supplier is drafted and payment is held.
- Person
Payables lead. Calls the supplier on the number already on file, not the one on the invoice.
Called back - Action
Query sent; the invoice parked with both reasons recorded.
5 steps · one to a person · about 6 minutes of a person's time
- How long
- About 6 minutes of a person's time
- Who does the work
- AI prepares, a person decides
- What keeps it safe
- Changed bank details always held
- What goes to people
- Every decision
Autonomy level 3
Operational AI
It runs the process within set limits, logs every step and hands people the exceptions.
- Trigger
Every invoice, around the clock.
- Rule
Three-way match against the order and the goods receipt.
Tolerance: 2% or £250. Duplicates and changed bank details are always held.
- Action
Invoices within tolerance are approved and posted. Money goes out only in the normal payment run, released by a person.
- Person
Exception: goes to a person
Payables lead. This invoice is held twice over: 4.9% above the order, and new bank details. The lead gets the evidence and calls the supplier back on the number on file.
Called back - Record
[08:12]Decision, evidence and approver recorded for audit.
- Report
Weekly to the finance lead: the straight-through rate, time to clear exceptions, holds, and every correction.
6 steps · one to a person · minutes, and only for exceptions
- How long
- Minutes, and only for exceptions
- Who does the work
- AI runs matching; people handle exceptions and release payments
- What keeps it safe
- Tolerances, fraud holds, an audit trail
- What goes to people
- Exceptions, bank changes, every payment release
Controls
- Owner
- Financial controller
- Measure
- Straight-through rate and days to close
- Limit
- 2% or £250; AI never releases a payment
- Log
- Every decision with its evidence
- Fallback
- Invoices queue for the team
- Stop
- Pause matching in one switch
Illustrative: names, figures and timings are invented.
Deployed or operational?
Five questions tell you which you have.
If you cannot name its owner, its measure and its fallback, it is deployed, not operational.
| Question | Deployed | Operational |
|---|---|---|
| Owner | The team that built the pilot | A named person in the business, with time set aside |
| Measure | A good demo | A number the department already tracks, reviewed every month |
| When it fails | Someone notices, eventually | Monitoring raises it, and a fallback keeps the work moving |
| Who can change it | Whoever has access | Changes are reviewed, tested and logged |
| Running cost | Unknown until the invoice | Budgeted per process and compared with the value |
The autonomy ladder
Four levels. Each process earns the next one.
Treat it like a new hire: a job description, a sign-off limit, a manager and a monthly review. Some processes should stay at level two, and we will tell you which.
- 1
Level 1
Assist
AI drafts, finds and summarises. People decide and act.
- 2
Level 2
Automate with approval
AI completes steps. A person approves before anything leaves the building.
Where some processes should stay
- 3
Level 3
Run with exceptions
AI runs the process. Exceptions and edge cases go to people.
- 4
Level 4
Run and improve
AI runs and improves the process within limits you set and review.
Decisions about people carry legal duties
When AI helps make decisions about people, the law adds duties. Hiring, credit and access to services are common examples.
In the UK, Articles 22A to 22D of the UK GDPR, in force since 5 February 2026, set the rules for significant decisions made solely by automated means. The safeguards include telling people, letting them make representations, a route to human review and a right to contest. Decisions that use special category data, such as health, are restricted further.
In the EU, the AI Act classes uses such as recruitment and credit scoring as high-risk, with those duties applying from 2 December 2027. People must also be told when they are dealing with an AI system.
We design for these duties from the start, with human review where it is needed, alongside your legal or data protection lead.
How we do it
We build it in your systems, then hand it over.
- 1
Diagnose
Two to four weeks
We look at the numbers, the team and the systems, and agree what to fix first.
- 2
Plan
Agreed in writing
One plan with owners, measures and dates, shared with everyone who pays for it.
- 3
Embed
Usually eight to twelve weeks
We do the work with your team: selling, building, running, and fixing what breaks.
- 4
Hand over
When it holds
Documentation, a named owner and a trained team. We step back when the work runs without us.
Who does the work
The founders lead the work.
The people who design it build it and run it with you. Our advisors join where their experience fits.
Rohith Varma
Co-founder
VP Sales and Strategic Advisor at Multiverse Computing
Revenue built from zero in new markets across the UK, Europe, the Gulf and Asia-Pacific. Government and regulated buyers, and investor relationships.
Felipe Padilla Gomez
Strategic advisor
Senior Director, AI for Deeptech and Corporate Venture Office, Hitachi
Former CTO roles at Hitachi Vantara and Vodafone. Industrial AI and corporate venturing, after more than 30 years in technology.
Start with one process that pays back.
We map the work, pick the process to run first, and build it with your team.