We look for parts of the work where AI can take repetitive reading, sorting or searching off people’s plates — and we use simpler automation everywhere else.
We look at where people spend time reading, sorting, searching or repeating steps — then decide whether a simple rule, a better connection, staff training, or a small AI feature with human review is the right answer. Where data lives — in the EU with Mistral, on US servers with privacy controls, or on hardware inside the organisation — follows how strict the requirement is. Machines have to match that work and those rules.
How we think
Useful AI helps where people currently read, sort, summarise or search. Known rules belong in ordinary software. People stay responsible for judgement, approvals and exceptions.
01Problems you may recognise
Not every organisation has all of these. They are patterns we see when information, handoffs or repetitive reading become the bottleneck.
Emails, forms, documents or tickets need reading and sorting before work can even start.
The answer exists somewhere in the company, but finding it reliably takes longer than it should.
The amount of reading and sorting grows with demand, while hiring cannot keep up.
Expensive AI is doing work that a clear rule, a better form or a connection between systems could have done more reliably.
01bIn this area
These are the concrete offerings. Open one if you already know the kind of work.
AI
For organisations that want AI to become part of how work gets done — with governance, adoption and measurable operational value built in from the start.
Open this serviceAI
For teams with repeatable processes where employees are still manually moving every case from one predictable step to the next.
Open this serviceAI
For organisations that need serious AI capability without overspending on infrastructure they do not need — or underbuilding infrastructure that fails once it reaches production.
Open this serviceAI
For teams that spend hours pulling the same fields from the same kind of documents, emails or tickets.
Open this serviceAI
This is not a chatbot trained to sound confident. It is a controlled interface to the knowledge your business actually trusts.
Open this service02Fit
02bAlso important
03How work should be split
Known rules belong in ordinary software. AI helps with language and interpretation. People keep judgement and high-impact decisions.
Automation
The rule is known and stable
Move a record, send a notification, sync systems, check a field, create a task
AI
Someone must read, sort or make sense of less structured information
Understand an email, extract a document, summarise a case, search knowledge, draft a reply
Human
Judgement, relationships or important commitments are required
Approvals, exceptions, sensitive decisions, customer relationships
Unnecessary AI costs more to run, is harder to predict, and needs more checking. Ordinary software should handle ordinary rules. AI is for reading and language that actually matter.
04What we can implement
Pull out the important fields and send the case on — with a person checking when the system is unsure.
Understand a request well enough to send it to the right queue without someone reading every line.
Answer from approved company material and show where the answer came from.
Suggest next steps or summaries while people stay in control of what gets sent or promised.
Prepare replies, notes or reports for a person to edit — not unsupervised send.
Highlight the odd ones so specialists spend time where judgement is needed.
Connect AI into orders, warehouse, documents, email or internal tools so it is a working step, not a separate chatbot.
Size cloud, local or hybrid AI environments for the models, usage, latency and data rules — not a shopping list. Too much wastes money; too little fails in production.
05How we choose
Practice-specific checks for this engagement. They are not a ceremony every project must complete in order.
Map attention
What currently takes up employee time?
Spot reading work
Which steps require reading, language or judgement?
Try the rule first
Can a simple rule, form or connection solve it?
Test whether AI is worth it
Does it improve quality or speed enough to justify the extra complexity?
Plan for uncertainty
What happens when the system is unsure?
Start small
Ship the smallest useful version with review and measurement.
Options we weigh
Not every engagement uses every option. We pick the simplest mix that fits the workflow.
Reuse
Keep what already works.
Connect
Link systems so people stop copying by hand.
Automate
Use ordinary software for predictable steps.
Apply AI
Use AI where reading, language or sorting adds value.
Build
Create custom software where standard tools genuinely do not fit.
06Cost & complexity
No invented savings percentages — just the levers that usually matter.
Not every task needs the strongest model. Simple sorting can stay cheap; harder analysis can step up.
Known mappings, checks and transfers should not become AI calls.
Ground answers in your own sources instead of letting the system invent from general knowledge.
Drafts and suggestions are cheaper mistakes than unsupervised promises.
Hardware has to fit what you are running — operations, documents, a product, the warehouse floor — not a catalogue. On-site versus cloud is a fit choice, not a default. When data must stay in the EU, we use Mistral. When EU residency is not required, US servers can be used — encrypted, with access limited, so the host is not reading the content. If data cannot sit on any outside server, dedicated AI hardware inside the organisation is the most private option, and it costs more. Microsoft Foundry and AWS Bedrock are integrations, not a way of keeping data in the EU.
07Examples
Example situations — not client case studies or promised results.
Problem
A team reads the same kind of supplier documents every day to type fields into another system.
Possible direction
Extract the fields automatically, send unsure items to a person, write approved results into the system you already use.
Why: Spend people time on the exceptions — not on every routine document.
Problem
Staff ask the same policy questions because answers live in scattered files.
Possible direction
An internal helper that answers only from approved documents and shows its sources.
Why: Reduce search time without inventing answers.
Problem
Someone proposed “an AI platform” before the work was clear.
Possible direction
Pick one reading bottleneck, prove value there, keep surrounding steps as ordinary automation.
Why: Avoid paying for a large AI programme when a small feature would do.
08Questions
No. Many wins are structured extraction, sorting or routing with smaller systems — or no AI at all when a rule is enough.
Approved sources, hard business rules, a person for important steps, a clear fallback when the system is unsure, and access that matches your organisation.
Yes. A single feature inside an existing tool is often the right first step. Broader platforms only make sense when the work and ownership are proven.
It depends on the work you are running and where data is allowed to live — hardware has to match both, not a shopping list. When data must stay in the EU, we use Mistral. When EU residency is not required, work can run on US servers — still treated as private: encrypted, with access limited, so the host is not reading the content. If data cannot sit on any outside server, we can build dedicated AI hardware inside your organisation; that is the most private option and it costs more. We can connect to Microsoft Foundry or AWS Bedrock (Microsoft’s and Amazon’s AI platforms) when those are the services you already use — that is an integration, not a way of keeping data in the EU.
No. Machines have to match what you are actually running and where data is allowed to live. We advise on that fit — including proprietary hardware inside the organisation when data cannot sit on any outside server. We will not specify more than you need, or something too weak for daily use. We are not a shop for everyday computers.
09Related
AI features usually sit inside custom software or logistics workflows — connected to systems you already run, as a working step rather than a separate chatbot.
Software built around how your operation actually works.
ExploreOne shared logistics picture so floor, planners and customers can coordinate on the same order.
ExploreWebsites and tools that do a real job — not just look finished.
ExploreNext step
Prefer a short conversation about the workflow before deciding technology? That is usually enough to know whether this practice is the right starting point.
Tell us which work currently requires reading, sorting or searching. We will help separate rules, AI and human judgement without assuming a big platform.