Technology & technical assessment
Understand what AI says about your business, where its information comes from and what your systems need to support better discovery, controlled access and useful adoption.
We choose the methods that suit your question, then use the findings to write requirements your teams can review.
Assessment priorities
Inspectable evidence
Reliable source information
Integration and evaluation
The technical foundation
An inaccurate recommendation can come from outdated product data, unclear website content or a third-party source. An internal assistant can struggle because approved information is hard to retrieve or permissions are poorly defined.
We investigate the sources, systems and processes behind each finding. Our recommendations explain the issue, the proposed change and how your team can check the result.
Our methods
Public AI answers and internal AI workflows need different forms of evaluation. We agree the relevant evidence and methods for each engagement.
01
Observe how your brand appears across agreed questions, markets and AI platforms. Record the prompt, date, test conditions, answer and cited sources. Distinguish a mention from a recommendation and check important claims against approved references.
You receive: Recorded answers, source analysis and findings from a clearly defined sample.
02
Review crawlability, rendering, internal links, content structure and information consistency. Examine whether pages answer the related questions AI search may explore through query fan-out. Check that structured data and product feeds agree with visible content and the underlying offer.
You receive: Prioritised technical requirements, content briefs and data-quality checks.
03
Review logs and available bot verification, then assess controls for the assets and systems in scope. Consider robots directives, snippet controls, authentication and edge rules according to purpose. Assess licensing or HTTP 402 paid-access feasibility where there is a commercial case.
You receive: An access-policy assessment, control requirements and a validation plan. Paid access also needs platform participation and a workable commercial agreement.
04
Assess how product discovery connects to catalogue data, availability, checkout and customer support. Review relevant UCP requirements for commerce. Where agents need tool access, assess Model Context Protocol (MCP) or existing APIs, with clear identity, permissions and confirmation for consequential actions.
You receive: An integration brief covering eligibility, data contracts, permissions, transaction handling and pilot acceptance criteria.
05
Assess whether an assistant can find the approved information a task requires. Where suitable, review retrieval-augmented generation (RAG), which supplies source material to a model when it answers. Test retrieval, answer quality, permissions, failure handling and human review against realistic cases.
You receive: A pilot design, evaluation cases and recommendations for source preparation, workflow controls and operating ownership.
An assessment in practice
Each finding records what we observed, the reference we checked and the recommended action. The example below shows how this works.
Can I collect this drill from my local store today?
Same-day collection is available at every store.
Collection depends on stock at the selected store and the order cut-off time.
The answer omits the location, availability and timing conditions.
Reconcile product-page and feed information, clarify collection conditions and review the cited source.
Commerce and product-data teams. Check the updated sources and repeat the location-specific scenario across the agreed sample.
Architecture and delivery
We review your existing platforms, data flows and operating responsibilities before recommending an integration. The proposed design should make access, ownership and ongoing support clear.
01
Identify the required sources, allowed uses and access boundaries. Review retention, provider terms and relevant regional requirements with your security and procurement owners. The pilot should use only the information it needs.
02
Agree quality criteria, test cases, review responsibilities and escalation routes. Assess costs and support needs alongside performance. Decide how changes to information or models will be reviewed after launch.
03
Give teams the proposed architecture, data contracts, dependencies and acceptance criteria. Make responsibilities and unresolved decisions explicit so engineering and programme owners can scope delivery.
Explore SWLR
Common questions
That depends on the work. An assessment can use existing analytics, logs, content and approved tools. We review what you already have before recommending software or an integration. A proposed solution should have a clear purpose and an agreed owner.
No. Accurate structured data can support a clear description of your pages and products, but it does not guarantee selection or a citation. Google states that its AI search features require no special schema. Useful content and sound technical SEO remain relevant foundations.
No. Controls can affect future access to assets within their scope. They do not remove information already collected or stop every third-party copy from being used. We assess the available controls, their effect on discovery and their practical limits.
Signal → Evidence → Action
Tell us which answer, data issue or workflow you need to investigate. We can discuss the evidence and the technical scope needed to make a useful decision.
Discuss your AI strategy