phainly
Talk to us

LUCID

Pilots open

Every “no”, explained. Every review, real.

LUCID is the explanation and human-review layer for automated credit decisions. It sits beside your existing scoring and makes every decline clear to the customer, defensible to the supervisor and fixable on the next application.

Request a CCD2 diagnostic Try the demo

CCD2 gives borrowers four new rights. LUCID delivers each one.

When a creditworthiness assessment relies on automated processing, the borrower can ask for all of the following. Most lenders today can answer none of them at scale.

A clear explanation

LUCIDPlain-language reasons, the logic behind them and the data used, traced to the exact transactions, rules and model version.

Human intervention

LUCIDOne tap on the decline screen opens a case for a qualified reviewer, with the whole file ready.

To give their view

LUCIDThe customer adds context and documents in the same flow. Reviewers see it next to the model's reasoning.

A review of the decision

LUCIDEvery review is timed, logged and tracked to its outcome, then sealed into an evidence pack.

LUCID supports your compliance work. Your legal team decides what your firm needs under national law.

One decision, three readers.

Follow a declined pay-in-6 application from the customer to the reviewer to the supervisor. Every name and figure here is synthetic.

Case R-2611-0417, pay in 6, €680

09:14Your bank

We can't approve €680 today.

Here is why, and what would change the answer.

  1. Your regular commitments take 52% of your monthly income. For this amount we need 45% or less.

  2. Two direct debits were returned in the last 90 days.

  3. Your income varied by 31% from month to month. We look for 25% or less.

What would change the answer

€400 can be approved today, in 6 payments of €66.67.

From 15 January your car loan ends and commitments fall to 39%. You could then apply for the full €680.

Explanation E-14, generated from the data and rules used in this decision. 4 Nov 2026, 09:14.

  • Reasons ranked by weight.The customer sees the factors that mattered most, in plain words, with the limit they missed.
  • Evidence behind every reason.Each reason opens the exact transactions it rests on.
  • A path to yes.A smaller amount today, or the date when the answer changes, worked out from the cash-flow forecast.
  • A person is one tap away.Review requests and the customer's own account go straight to your review queue. Press “Ask a person to review” to follow the case.

What LUCID does.

Six modules on one decision ledger. Start with the ones CCD2 needs first and add the rest as you go.

Explain

Reasons a customer understands and logic a supervisor accepts. Each sentence links to the transactions, rules and model version behind it.

Path to yes

Shows what would change the outcome and when: a smaller amount today, or a date to re-apply based on the customer's cash-flow forecast.

Human review

Queues with service levels, the customer's own account and four-eyes options. Seeded test cases show that reviewers really check.

Evidence

An append-only, hash-chained ledger of every decision. Export the full pack for a complaint, an audit or a supervisor.

Ask

Credit, risk and compliance teams ask questions in plain language. Each answer comes with the SQL that produced it.

Embed

A software kit and UI components for web and mobile: decline explanation, path to yes, review request and consent, in your brand.

Underneath sit more than twenty explainable signals drawn from bank transactions, such as income regularity, recurring commitments, subscriptions and cash-flow gaps. Each one carries its own explanation.

Ask the ledger anything.

Your teams ask in plain language. LUCID answers from the decision ledger and shows the query behind every answer, so anyone can check it.

Why did declines rise last week?

SELECT rule_id,
       COUNT(*) FILTER (WHERE iso_week = '2026-W44') AS last_week,
       COUNT(*) FILTER (WHERE iso_week = '2026-W43') AS week_before
FROM decline_reasons
WHERE iso_week IN ('2026-W43', '2026-W44')
GROUP BY rule_id
ORDER BY last_week - week_before DESC
LIMIT 3;

Declines rose by 186 week on week. 127 of them came from rule R-17, commitments above 45% of income, after October rent indexation. 214 declined customers are within €50 a month of passing.

Example questions on synthetic data.

Live without replacing anything.

LUCID works with any scoring model, decision engine or core system. Three steps take you from decision data to live customer screens.

  1. Connect your decisions

    Send each decision from your decision engine or loan system: inputs, score and the rules that fired. An API call or a nightly file both work.

  2. Map the reasons

    We turn your rules and model features into plain-language reasons in your customers' languages. Your compliance team signs them off.

  3. Go live in your app

    Add the decline, path-to-yes and review screens, connect the review queue and start the pilot. Integration is designed to take under six weeks.

One layer for three rulebooks.

The same requirement arrived in all three of our markets in 2026: explain the automated decision, offer a person and allow a challenge.

EU

  • 20 November 2026Consumer Credit Directive: explanation, human intervention, the customer's view and a review.
  • 10 July 2027Anti-Money Laundering Regulation: meaningful human intervention in automated customer decisions.
  • 2 December 2027AI Act: high-risk duties for creditworthiness assessment.

UK

  • 5 February 2026Data (Use and Access) Act: safeguards for significant automated decisions, including human intervention and the right to contest.
  • Consumer DutyFirms must support customer understanding and avoid foreseeable harm.

UAE

  • February 2026Central Bank guidance on AI and machine learning: explanations, human oversight of credit decisions and a route to contest them.

Designed by financial-regulation lawyers and machine-learning engineers, and funded by the European Innovation Council.

Questions lenders ask.

Does LUCID replace our scoring model?

No. LUCID works with any model or rule set. It receives the decision, its inputs and the rules that fired, then explains, routes and records them.

Is a generic reason code enough under CCD2?

CCD2 asks for a clear and comprehensible explanation, including the logic and risks of the automated processing. The EU Court of Justice has said explanations must describe the procedure and principles actually applied (case C-203/22, February 2025). LUCID is built to that standard. Your counsel decides what your firm needs.

Who makes the decision on review?

A qualified person at your firm. LUCID gives them the case file and records who decided what, and why.

Does it cover buy now, pay later and small loans?

CCD2 extends to most buy-now-pay-later products and to loans under €200, subject to national choices. LUCID works for any credit product with an automated decision.

Where is the data stored?

In your cloud region or on your own infrastructure. LUCID keeps references and hashes where it can, instead of copies of personal data.

Get ready for 20 November.

Our readiness diagnostic maps your automated credit decisions against CCD2 and shows exactly what is missing, from customer screens to review records.

hello@phainly.com

Request a CCD2 diagnostic

SIGILMandates and proof for AI agents that act for your customers.Design partners for 2027 KEELTreasury and brakes for the money AI agents spend.Early access