TechGuidr engagement · Economic research
Two AI products built for a research business
A research business needed customers to find its current thinking across a large archive. Answerable brought cited research answers into daily use; Foreseeable brought AI into its account teams’ workflow.
- Role
- TechGuidr led the product design, architecture and implementation.
- Delivery status
- Answerable: in production and daily use. Foreseeable: in production.
- Attribution
- Answerable and Foreseeable are TechGuidr-owned products. This story describes their development around a research business’s requirements.
The opportunity
The research business held more than 78,500 publications, around 89,000 chart assets and approximately 35,700 structured forecast data points. Customers needed to find and interpret the organisation’s current view across that material.
The difficult requirement was keeping an answer current and supported. A relevant publication could still contain an older position. A useful answer needed to account for that, show the supporting research and let the reader check it.
A research answer people can examine
Answerable combines retrieval across the archive with a synthesis of the relevant research. Source references and underlying publications are shown alongside the answer.
The implementation separates question classification, retrieval, synthesis and a grounding check. Recency and supersession rules help distinguish the current position from earlier material. Forecast data can be supplied alongside prose when the question needs it.
Readers can open the underlying publication and check whether its date, context and evidence support the answer. That review is part of the workflow.
AI inside the account team’s workflow
Foreseeable brings portfolio information, renewal signals and structured account diagnostics into a workspace for account teams. Proposed actions include reasoning, an owner and a date, with a person deciding whether to accept them.
Commercial calculations use deterministic logic. AI supports interpretation and drafting, while account teams decide which proposed actions to take.
What the delivery involved
TechGuidr carried out the product thinking, architecture, data preparation, engineering and interface work. Much of the implementation effort concerned the corpus, retrieval behaviour, commercial rules and how users would inspect an answer.
A data-access issue identified during ingestion was escalated to leadership alongside the product work. The case describes that escalation without exposing the affected configuration.
How the work changed
Both products changed who does what: the system does the preparation, and people keep the judgement and the decision.
Answerable
The reader searches the archive, pulls the relevant research together and decides what it means.
AI retrieves the relevant research and synthesises the current view. The reader checks the evidence behind the answer and decides.
Foreseeable
Portfolio health, renewal signals, account diagnostics and proposed actions, each with its reasoning and a trace to the source.
The decision on every action. A person accepts or rejects each one, every accepted action has an owner and a date, and the commercial calculations stay deterministic.
What the business told us about Answerable
The feedback was about the work itself. The business saw Answerable get economists most of the way to a useful answer quickly, which leaves them to do the part only they should do: the judgement, the challenge and the final answer.
- It makes economists’ jobs easier
- Less time finding and reconstructing what the firm already knows
- More of the experts’ time on judgement and the final part of the answer
- A change in how the work can be done, beyond a better way to search
Themes from the business’s feedback, in our words rather than quoted. Time saved has not been measured against a baseline, so we do not put a figure on it.
Engineering detail
The research pipeline combines keyword and vector retrieval, semantic reranking, chart processing and structured forecast information. The architecture below is a simplified account of the implementation.
- Hybrid vector and BM25 retrieval with semantic reranking
- Recency and supersession logic for historical research
- Separate synthesis and grounding-check stages
- Source cards and confidence indicators for reader review
- Commercial workflow states and deterministic calculations in Foreseeable
Proprietary corpus
- 78,500+ publications
- ~89,000 chart assets
- ~35,700 forecast points
- ~163,000 indexed chunks
- Decades of research
- QuestionClassified by intent
- Hybrid retrievalVector and BM25 search, reranked, with recency and supersession
- SynthesisThe house view, drawn across sources and forecasts
- ValidationA separate grounding check
- Grounded answerInline citations and source cards for verification
The outcome
Answerable entered production and daily use, and the business reports that it makes economists’ work easier, getting them most of the way to a useful answer quickly. Foreseeable went into production, bringing account diagnostics and proposed actions for a person to approve.
Explore Build & ProductioniseWhat this architecture now supports
The engagement above was a research-intelligence deployment, and this account describes it as it was delivered. Since then we have developed Answerable, on the same architecture, into an AI working environment for proprietary knowledge and data more broadly.
It now puts documents and structured data side by side for research, analysis, comparison, briefings and recurring intelligence workflows, with permissions, domain guardrails, expert review and evaluation designed in. Clients can use it as our product, or we can build the architecture inside their own environment.
These later capabilities are shown on demo environments, with configurations for legal, life sciences, engineering, wealth, customer support, markets and workforce work. They are not claims about the client in this case study.
- Documents and structured data analysed together
- Comparison across documents, datasets and time
- Briefings, observatories and recurring intelligence workflows
- Permission-aware access, domain guardrails and expert review
- Used as our product, or built inside a client’s own environment
Talk through a similar situation
Tell us what you are building, buying or trying to fix. A couple of sentences is enough.